AI security is a hot topic in the world of cybersecurity. If you don’t believe me, a brief glance at LinkedIn uncovers that everyone is an AI security expert now. This is why we end up with overly complex and sometimes nonsensical recommendations regarding the topic. But in the bustling market of thought leadership and job updates, we’ve seemed to have lost the plot. In most cases, it’s not AI security at all, but something else.
Misnomer of AI Security: It’s Security From AI
I recently delivered the keynote at the c0c0n cybersecurity and hacking conference in India. It was truly an amazing experience. One of my takeaways was encouraging a shift in perspective on the term “AI Security,” highlighting how we often approach this topic from the wrong angle.
The term “AI Security” has become a misnomer in the age of generative AI. In most cases, we really mean securing the application or use case from the effects of adding AI. This makes sense because adding AI to a previously robust application makes it vulnerable.
In most cases, we really mean securing the application or use case from the effects of adding AI.
For most AI-powered applications, the AI component isn’t the end target, but a manipulation or entry point. This is especially true for things like agents. An attacker manipulates the AI component to achieve a goal, such as accessing sensitive data or triggering unintended outcomes. Consider this like social engineering a human as part of an attack. The human isn’t the end goal for the attacker. The goal is to get the human to act on the attacker’s behalf. Thinking this way transforms the AI feature into an actor in the environment rather than a traditional software component.
There are certainly exceptions, such as with products like ChatGPT, where guardrails prevent the return of certain types of content that an attacker may want to access. An attacker may seek to bypass these guardrails to return that content, making the model implementation itself the target. Alternatively, in another scenario, an attacker may want to poison the model to affect its outcomes or other applications that implement the poisoned model. Conditions like these exist, but are dwarfed in scale by the security from AI scenarios.
Once we start thinking this way, it makes a lot of sense. We shift to the mindset of protecting the application rather than focusing on the AI component.
AI Increases Attack Surface
Another thing to consider is that adding AI to an application increases the attack surface. Increase in attack surface manifests in two ways: first, functionally through the inclusion of the AI component itself. The AI component creates a manipulation and potential access point that an attacker can utilize to gain further access or create downstream negative impacts.
Second, current trendy AI approaches encourage poor security practices. Consider practices like combining data, such as integrating sensitive, non-sensitive, internal, and external data to create context for generative AI. This creates a new high-value target and is a poor practice that we’ve known from decades of information security guidance.
Also, we have trends where developers take user input, request code at runtime, and slap it into something like a Python exec(). This not only creates conditions ripe for remote code execution but also a trend where developers don’t know what code will execute at runtime.
Vulnerabilities caused by applying AI to applications don’t care whether we are an attacker or a defender. They affect applications equally. This runs from the AI-powered travel agent to our new fancy AI-powered SOC. Diamonds are forever, and AI vulns are for everyone.
It’s Simpler Than It Seems
Here’s a secret. In the real world, most AI security is just application and product security. AI models and functionality do nothing on their own. They must be put in an application and utilized in a use case, where risks materialize. It’s not like AI came along and suddenly made things like access control and isolation irrelevant. Instead, controls like these became more important than ever, providing critical control over unintended consequences. Oddly enough, we seem to relearn this lesson with every new emerging technology.
In the real world, most AI security is just application and product security.
The downside is that without these programs in place, organizations will accelerate vulnerabilities into production. Not only will they increase their vulnerabilities, but they’ll be less able to address them properly when vulnerabilities are identified. Trust me, this isn’t the increase in velocity we’re looking for.
I’ve been disappointed at much of the AI security guidance, which seems to disregard things like risk and likelihood of attack in favor of overly complex steps and unrealistic guidance. We security professionals aren’t doing ourselves any favors with this stuff. We should be working to simplify, but instead, we are making things more complex.
It can seem counterintuitive to assume that something a developer purposefully implements into an application is a threat, but that’s exactly what we need to do. When designing applications, we need to consider the AI components as potential malicious actors or, at the very least, error-prone actors. Thinking this way shifts the perspective for defending applications towards architectural controls and mitigations rather than relying on detecting and preventing specific attacks. So much focus right now is on detection and prevention of prompt injection, and it isn’t getting us anywhere, and apps are still getting owned.
I’m not saying detection and prevention don’t play a role in the security strategy. I’m saying they shouldn’t be relied upon. We make different design choices when we assume our application can be compromised or can malfunction. There are also conversations about whether security vulnerabilities in AI applications are features or bugs, allowing them to persist in systems. While the battle rages on, applications remain vulnerable. We need to protect ourselves.
There is no silver bullet, and even doing the right things sometimes isn’t enough to avoid negative impacts. But if we want to deploy generative AI-based applications as securely as possible, then we must defend them as though they can be exploited. We can dance like nobody is watching, but people will discover our vulnerabilities. Defend accordingly.
The past couple of years have been fueled entirely by vibes. Awash with nonsensical predictions and messianic claims that AI has come to deliver us from our tortured existence. Starting shortly after the launch of ChatGPT, internet prophets have claimed that we are merely six months away from major impacts and accompanying unemployment. GPT-5 was going to be AGI, all jobs would be lost, and nothing for humans to do except sit around and post slop to social media. This nonsense litters the digital landscape, and instead of shaming the litterers, we migrate to a new spot with complete amnesia and let the littering continue.
Pushing back against the hype has been a lonely position for the past few years. Thankfully, it’s not so lonely anymore, as people build resilience to AI hype and bullshit. Still, the damage is already done in many cases, and hypesters continue to hype. It’s also not uncommon for people to be consumed by sunk costs or oblivious to simple solutions. So, the dumpster fire rodeo continues.
Security and Generative AI Excitement
Anyone in the security game for a while knows the old business vs security battle. When security risks conflict with a company’s revenue-generating (or about to be revenue-generating) products, security will almost always lose. Companies will deploy products even with existing security issues if they feel the benefits (like profits) outweigh the risks. Fair enough, this is known to us, but there’s something new now.
What we’ve learned over the past couple of years is that companies will often plunge vulnerable and error-prone software deep into systems without even having a clear use case or a specific problem to solve. This is new because it involves all risk with potentially no reward. These companies are hoping that users define a use case for them, creating solutions in search of problems.
What we’ve learned over the past couple of years is that companies will often plunge vulnerable and error-prone software deep into systems without even having a clear use case or a specific problem to solve.
I’m not referring to the usage of tools like ChatGPT, Claude, or any of the countless other chatbot services here. What I’m referring to is the deep integration of these tools into critical components of the operating system, web browser, or cloud environments. I’m thinking of tools like Microsoft’s Recall, OpenAI’s Operator, Claude Computer Use, Perplexity’s Comet browser, and a host of other similar tools. Of course, this also extends to critical components in software that companies develop and deploy.
At this point, you may be wondering why companies choose to expose themselves and their users to so much risk. The answer is quite simple, because they can. Ultimately, these tools are burnouts for investors. These tools don’t need to solve any specific problem, and their deep integration is used to demonstrate “progress” to investors.
I’ve written before about the point when the capabilities of a technology can’t go wide, it goes deep. Well, this is about as deep as it gets. These tools expose an unprecedented attack surface and often violate security models that are designed to keep systems and users safe. I know what you are thinking, what do you mean, these tools don’t have a use case? You can use them for… and also ah…
The Vacation Agent???
The killer use case that’s been proposed for these systems and parroted over and over is the vacation agent. A use case that could only be devised by an alien from a faraway planet who doesn’t understand the concept of what a vacation is. As the concept goes, these agents will learn about you from your activity and preferences. When it’s time to take a vacation, the agent will automatically find locations you might like, activities you may enjoy, suitable transportation, and appropriate days, and shop for the best deals. Based on this information, it automatically books this vacation for you. Who wouldn’t want that? Well, other than absolutely everyone.
What this alien species misses is the obvious fact that researching locations and activities is part of the fun of a vacation! Vacations are a precious resource for most people, and planning activities is part of the fun of looking forward to a vacation. Even the non-vacation aspect of searching for the cheapest flight is far from a tedious activity, thanks to the numerous online tools dedicated to this task. Most people don’t want to one-shot a vacation when the activity removes value, and the potential for issues increases drastically.
But, I Needed NFTs Too
Despite this lack of obvious use cases, people continue to tell me that I need these deeply integrated tools connected to all my stuff and that they are essential to my future. Well, people also told me I needed NFTs, too. I was told NFTs were the future of art, and I’d better get on board or be left behind, living in the past, enjoying physical art like a loser. But NFTs were never about art, or even value. They were a form of in-group signaling. When I asked NFT collectors what value they got from them, they clearly stated it wasn’t about art. They’d tell me how they used their NFT ownership as an invitation to private parties at conferences and such. So, fair enough, there was some utility there.
In the end, NFTs are safer than AI because they don’t really do anything other than make us look stupid. Generative AI deployed deeply throughout our systems can expose us to far more than ridicule, opening us up to attack, severe privacy violations, and a host of other compromises.
In a way, this public expression of look at me, I use AI for everything has become a new form of in-group signaling, but I don’t think this is the flex they think it is. In a way, these people believe this is an expression of preparation for the future, but it could very well be the opposite. The increase in cognitive offloading and the manufactured dependence is precisely what makes them vulnerable to the future.
In a way, these people believe this is an expression of preparation for the future, but it could very well be the opposite. The increase in cognitive offloading and the manufactured dependence is precisely what makes them vulnerable to the future.
Advice Over Reality
Social media is awash with countless people who continue to dispense advice, telling others that if you don’t deploy wonky, error-prone, and highly manipulable software deeply throughout your business, then they are going to be left behind. Strange advice since the reality is that most organizations aren’t reaping benefits from generative AI.
Here’s something to consider. Many of the people doling out this advice haven’t actually done the thing they are talking about or have any particular insight into the trend or problems to be solved. But it doesn’t end with business advice. This trend also extends to AI standards and recommendations, which are often developed at least in part by individuals with little or no experience in the topic. This results in overcomplicated guidance and recommendations that aren’t applicable in the real world.
The reason a majority of generative AI projects fail is due to several factors. Failing to select an appropriate use case, overlooking complexity and edge cases, disregarding costs, ignoring manipulation risks, holding unrealistic expectations, and a host of other issues are key drivers of project failure. Far too many organizations expect generative AI to act like AGI and allow them to shed human resources, but this isn’t a reality today.
LLMs have their use cases, and these use cases increase if the cost of failure is low. So, the lower the risk, the larger the number of use cases. Pretty logical. Like most technology, the value from generative AI comes from selective use, not blanket use. Not every problem is best solved non-deterministically.
Another thing I find surprising is that a vast majority of generative AI projects are never benchmarked against other approaches. Other approaches may be better suited to the task, more explainable, and far more performant. If I had to take a guess, I would guess that this number is close to 0.
Generative AI and The Dumpster Fire Rodeo
Despite the shift in attitude toward generative AI and the obvious evidence of its limitations, we still have instances of companies forcing their employees to use generative AI due to a preconceived notion of a productivity explosion. Once again, ChatGPT isn’t AGI. This do everything with generative AI approach extends beyond regular users to developers, and it is here that negative impacts increase.
I’ve referred to the current push to make every application generative AI-powered as the Dumpster Fire Rodeo. Companies are rapidly churning out vulnerable AI-powered applications. Relatively rare vulnerabilities, such as remote code execution, are increasingly common. Applications can regularly be talked into taking actions the developer didn’t intend, and users can manipulate their way into elevated privileges and gain access to sensitive data they shouldn’t have access to. Hence, the dumpster fire analogy. Of course, this also extends to the fact that application performance can worsen with the application of generative AI.
The generalized nature of generative AI means that the same system making critical decisions inside of your application is the same one that gives you recipes in the style of Shakespeare. There is a nearly unlimited number of undocumented protocols that an attacker can use to manipulate applications implementing generative AI, and these are often not taken into consideration when building and deploying the application. The dumpster fire continues. Yippee Ki-Yay.
Conclusion
Despite the obvious downsides, the dumpster fire rodeo is far from over. There’s too much money riding on it. The reckless nature with which people deploy generative AI deep into systems continues. Rather than identifying an actual problem and applying generative AI to an appropriate use case, companies choose to marinade everything in it, hoping that a problem emerges. This is far from a winning strategy. Companies should be mindful of the risks and choose the right use cases to ensure success.
Although AI has taken a hit in the past few weeks, the vibes are still strong and infecting every part of our lives. Vibe coding, vibe analytics, and even vibe thinking, because well, nothing says “old” like having thoughts grounded in reality. However, an interesting trend is emerging in software development, one that could have far-reaching implications for the future of software. This is a type of code roulette where developers don’t know what code will execute at runtime. Then again, what’s life without a little runtime suspense?
Development and Degraded Performance
The world runs on software, so any trend that degrades software quality or increases security issues has an outsized impact on the world around us. We’ve all witnessed this, whether it’s the video conferencing app that periodically crashes after an update or a UI refresh that makes an application more difficult to use.
Traditionally, developers write code by hand, copy code snippets, use frameworks, skeleton code, libraries, and many other methods to create software. Developers may even use generative AI tools to autocomplete code snippets or generate whole programs. This code is then packaged up and hosted for users. The code stays the same until updates or patches are applied.
But in this new paradigm, code and potentially logic are constantly changing inside the running application. This is because developers are outsourcing functional components of their applications to LLMs, a trend I predicted back in 2023 in The Brave New World of Degraded Performance. In the previous post, I covered the impacts of this trend, highlighting the degraded performance that results from swapping known, reliable methods for unknown, non-deterministic methods. This paradigm leads to the enshittification of applications and platforms.
In a simplified context, instead of developers writing out a complete function using code, they’d bundle up variables and ask an LLM to do it. For simplicity’s sake, imagine a function that determines whether a student passes or fails based on a few values.
def pass_fail(grade, project, class_time):
if grade >= 70 and project == "completed" and class_time >= 50:
return "Pass"
else:
return "Fail"
If a developer decided to outsource this functionality to an LLM inside their application, it may look something like this.
prompt_pass = """You are standing in for a teacher, determining whether a student passes or fails a class.
You will use several values to determine whether the student passes or fails:
The grade the student received: {grade}
Whether they completed the class project: {project}
The amount of class time the student attended (in minutes): {class_time}
The logic should follow these rules:
1. If the grade is above 70
2. If the project is completed
3. If the time in class is above 50
If these 3 conditions are met, the student passes. Otherwise, the student fails.
Based on this criterion, return a single word: "Pass" or "Fail". It's important to only return a single
word.
"""
prompt = prompt_pass.format(grade=grade, project=project, class_time=class_time)
response = client.models.generate_content(model="gemini-2.5-flash", contents=prompt)
print(response.text)
As you can see, one of these examples contains the logic for the function inside the application, and the other has the logic existing outside the application. The prompt is indeed visible inside the application, but the actual logic exists somewhere in the black box of LLM land.
The example using code has greater visibility, and it’s far more auditable since the logic can be examined, which makes it far easier to debug when issues arise, and of course, it’s explainable. The real problem lies in execution.
The written Python function approach gives you the same result based on the input data every single time, without fail. The natural language approach, not so much. In this non-deterministic approach, you are not guaranteed the same answer every time. Worse yet, when this approach is used for critical decisions and functionality, the application can take on squishy and malleable characteristics, meaning users can potentially manipulate them like Play-Doh.
At first glance, this example appears silly, as writing out the logic in natural language seems more burdensome than using the simple Python function. Not to mention, slower and more expensive. But looks can be deceiving. People are increasingly opting for the natural language approach, particularly those with only minimal Python knowledge. This natural language approach is also more familiar to people who are more accustomed to using interfaces like ChatGPT.
Execute and Pray
However, let’s take a look at another scenario. In this scenario, a developer wants to generate a scatter plot using the Plotly library. In this case, we have some data for the X and Y axes of a scatter plot and use Plotly Express, which is a high-level interface for Plotly (as a developer may when plotting something so simple).
This is a simplified example, but in this case, we can clearly see the code that generated the plot and be certain that this code will execute during the application’s runtime. There is control over the imports and other aspects of execution. It also makes it auditable and provable.
Now, what happens when a developer allows modification of their code at runtime? In the following example, instead of writing out the Plotly code to generate a scatter plot, the developer requests that code be generated from an LLM to create the graph, then executes the resulting code.
prompt_vis = """You are an amazing super awesome Python developer that excels at creating data visualizations using Plotly. Your task is to create a scatter plot using the following data:
Data for the x axis: {xdata}
Data for the y axis: {ydata}
Please write the Python code to generate this plot. Only return Python code and no explanations or
comments.
"""
prompt = prompt_vis.format(xdata=xdata, ydata=ydata)
response = client.models.generate_content(model="gemini-2.5-flash", contents=prompt)
exec(clean_response(response.text))
As you can see from the Plotly code in this example… Of course, you can’t see it because the code doesn’t exist until the function is called at runtime. If you are curious, the first run of this generated the following code after cleaning the response and making it appropriate for execution.
The AI-generated code creates the same graph as the written-out code in the previous example, despite being different. You may be wondering what the big deal is since the result is the same. The concern stems from several reasons, but primarily, allowing an LLM to generate code at runtime is not robust and leads to unexpected outcomes. These outcomes may include the generation of non-functional code, incorrect code, and even vulnerable code, among others.
For a simple example, as the one shown in this post, the chances of getting the same or incredibly similar code returned from the LLM are high, but not guaranteed. For more complex examples, such as those developers may want to use this approach for, the odds increase that the generated code will change more frequently.
Additionally, I implemented a quick cleaning function called clean_response to remove non-Python elements, such as text and triple backticks, from the response. The LLM can introduce additional unexpected characters that end up breaking my cleaning function and making my application fail. The list goes on and on, but a larger danger lurks in the background.
Whose Code Is It Anyway?
If you are versed in security and familiar with Python, you may have noticed something in the LLM example: The use of the Python exec() function. The exec () and eval() functions in Python are fun because they directly execute their input. Fun as in, dangerous. For example, if an attacker can inject input into the application, they can affect what code gets executed, leading to a condition known as Remote Code Execution (RCE).
An RCE is a type of arbitrary code execution in which an attacker can execute their own commands remotely, completely compromising the system running the vulnerable application. They can use this access to steal secrets, spread malware, pivot to other systems, or potentially backdoor the system running the application. Keep in mind, this system may be a company’s server, cloud infrastructure, or it may be your own system.
Anyone following security issues in AI development is aware that RCEs are flying off the shelves at alarming rates. A condition that was previously considered a rarity is becoming common. We even commented during our Black Hat USA presentation that it was strange to see people praising CISA for promoting memory safe languages to avoid things like remote code execution, while at the same time praising organizations essentially building RCE-as-a-Service. Some of this is mind-boggling, since in many cases, outsourcing these functions isn’t a better approach. In the previous example, writing out the Plotly code instead of generating it at runtime is relatively easy, more efficient, and far more robust.
Up until AI came along, the use of Python exec() was considered poor coding practice and dangerous. Now, developers shrug, stating that’s how applications work. As a matter of fact, agent platforms like HuggingFace’s smolagents use code execution by default. This is a wakeup. So, we dynamically generate code, provide deep access, and the ability to call tools, all with a lack of visibility. What could possibly go wrong???
Not only have developers chosen paradigms to generate and execute code at runtime, but worse yet, they’ve begun to perform this execution in agents with user (aka attacker) input, executing this input blindly in the application. In our presentation titled Hack To The Future: Owning AI-Powered Tools With Old School Vulns at Black Hat USA this year, we refer to this trend as Blind Execution of Input, which is the purposeful execution of input without any protection against negative consequences. This condition certainly leads to RCE and other unintended consequences, providing attackers with a significantly larger attack surface to exploit.
An application that takes user input and combines it with LLM functionality is a recipe for a bad time from a security perspective. Another common theme in our presentation, as well as that of other presenters on stage at Black Hat, is that if an attacker can get their data into your generative AI-based system, you can’t trust the output.
Things Will Get Worse
Using the outsourced approach when a more predictable deterministic approach is a better fit will continue to degrade software from a reliability and security perspective and have an impact on the future of software development.
Vulnerabilities in AI software have made exploitation as easy as it was in the 1990s. This was the “old school” hint in the title of our talk. This isn’t a good thing, because the 90s were a sort of free-for-all. Not only that, but in the 90s, we often had to live with vulnerabilities in systems and applications. For example, in one of the first vulnerabilities I discovered against menuset on Windows 3.1, it was impossible to fix. There were no mitigations, and most people were unaware of its existence.
As the outsourcing of logic to LLMs accelerates, things will worsen not only due to incorrect output and hallucinations but also from a security perspective. Anyone paying attention to the constant parade of vulnerabilities in AI-powered software can see this trend with their own eyes. These vulnerabilities are often found in large, mature organizations with dedicated security processes and teams in place to support them. Now, consider startups and organizations that implement their own experiments using non-deterministic software, often with a lack of understanding of how these systems can be manipulated. It’s become a game of speed above everything else.
As I’ve said from the beginning of the generative AI craze, the only way to address these issues is architecturally. Most of AI security is just application and product security, and organizations without these programs in place are in trouble. If proper architecture, design, isolation, secrets management, security testing, threat modeling, and a host of other activities weren’t considered table stakes before, they certainly are now. And possibly not surprisingly enough, they still aren’t being done. Anyone working for a security organization sees this every day.
In essence, developers need to design their applications to be robust to failures and attacks. It helps to consider designing them as though an attacker can manipulate and compromise them, working outward from this premise. As the adage goes, an attacker only needs to be successful once; a defender needs to be successful every time. This makes something that sounds great in theory, like being 90% effective, sound less impressive in practice.
Keep in mind that performing a code review won’t provide the same visibility as it has traditionally. This should be obvious since the code that would be audited doesn’t exist until runtime. You’ll have to pay more attention to validation routines and processing of outputs, putting huge question marks over the black box in the middle. And, of course, ensuring the application is properly isolated.
Some may suggest instrumenting the applications with functionality to perform runtime analysis on the generated code. Sure, it’s possible, but the performance hit would be significant, and even this is, of course, far from a silver bullet. You might not even get the value you think you are getting from this instrumentation. Also, you’d have to know ahead of time the issues you are trying to prevent. That is, unless you plan to layer more LLMs on top of LLMs in a spray-and-pray configuration.
To keep this grounded, all AI risk is use case dependent. AI models don’t do anything until packaged into applications and used in use cases. There may be cases where reliability, performance, and even security are of lesser concern. Fair enough, but it’s a mistake to treat all applications as though they fall into this category, and it’s far too easy to overlook something important and view it as insignificant.
If you work at an organization that isn’t building these applications and think you’re safe, you might want to think again, because you are at the mercy of third-party applications and libraries. It would be best to start asking hard questions of your vendors about their security practices as they relate to applications you purchase. Especially applications that use generative AI to generate code and execute it at runtime.
Near the end of our presentation, we had some advice.
Whether outsourcing the logic of an application to LLMs or having the LLM dynamically generate code, assume these are squishy, manipulable systems that are going to do things you don’t want them to do. They are going to be talked into taking actions that you didn’t intend, and fail and hallucinate in ways you don’t expect. Starting from this premise gives a proper foundation for deploying controls to add some resilience to these systems. Of course, not taking these steps means your applications will contribute to the ongoing dumpster fire rodeo.
We are continually inundated with examples of silly errors and hallucinations from generative AI. At this point, it’s no secret to anyone on the planet that these systems fail, sometimes at rather high rates. These systems also have a tendency to make stuff up, which isn’t a good look when that data is used for critical decisions. We’ve become numb to this new normal, creating a dangerous condition where we check out instead of recheck. But what happens when these errors and hallucinations become facts, facts that may be impossible to dispute or lurk in the background unseen and uncorrected?
Perspectives From Our Younger Selves
Imagine traveling back in time for a conversation with our younger selves about the current state of AI.
Younger: Wow, it must be great to live in a world without cancer or dementia. Older: No, we haven’t cured cancer or dementia. Younger: Well, at least people are super smart now. Older: No, there are still many dumbasses. Younger: At least you have systems that don’t make mistakes. Older: No, they make mistakes all the time. Younger: Then, what in the hell do you do with systems like this? Older: Mostly memes and short videos of stupid shit. Oh, we even try to impress world leaders with what they’d look like as a baby with a mustache.
Although it may seem silly, this thought experiment is informative. It puts our current AI moment in perspective and should add some humility. These systems aren’t the magnificent, magical boxes capable of handling every task with equal proficiency in both work and life. They are tools that we can use for specific tasks, far from the perfected AI of science fiction, and this is where the issues creep in.
Icebergs, Grenades, and Damage
I’ve made the grenade analogy before relating to agents. It’s an apt analogy because it’s something that causes damage, but not immediately. It’s like the classic joke grenade, which is a prank you play on your friends with the expectation of future laughter. Only with AI, the result isn’t a barrel of laughs. It’s a barrel of something that stinks and should be spread over a field as fertilizer.
The mistake is that seeing so many instances of these issues gives us the false impression that these issues are being caught and possibly even corrected. Think of issues like hallucinations as an iceberg. There are far more instances beneath the surface that lie unseen, lying in wait to send our ship to the depths.
There’s also the problem that not all conditions of hallucinations are so easy to identify. The ones that seem to get identified are those that are blatantly obvious or require additional validation, such as checking the cases referenced in a legal document. This is why it seems that only lawyers and politicians are making fools of themselves with AI. The landscape is far broader than these two categories.
It’s also instructive to see how people respond when these issues are brought to light. In the recent MAHA report scandal, the White House spokesman referred to AI hallucinations as “formatting issues.” Yeah, right. Imagine walking into your bank and finding out you have no money in your account. Frantic, you ask the teller what’s going on, and they tell you that you have no money because of a formatting issue. We can’t let people downplay these problems because they are common. It’s because they are common that we need to be more concerned.
We can’t let people downplay these problems because they are common. It’s because they are common that we need to be more concerned.
Although some instances may seem silly, there are no doubt real consequences. Such as AI hallucinating into people’s medical records, because we all know that can’t end badly. Hypothetically, let’s imagine that the generative AI system utilized is 99% accurate, which is enormously far from reality. Performing 10,000 transactions/results/outputs a day could potentially yield 100 issues. Crank that up to 1,000,000 a day, and that’s 10,000. This is terrifying when considering the realistically high error rates that these systems actually exhibit. There’s no doubt a river of manure flowing into data stores. The pin has been pulled.
The nature and pattern of errors differ significantly between AI and humans.
I can already feel the AI crowd’s eyes rolling, opening their mouths to issue the overused retort, “But humans make mistakes too.” Yes, they do, but human mistakes and AI mistakes aren’t the same. The nature and pattern of errors differ significantly between AI and humans. Human error tends to be more predictable, with errors and mistakes clustering around areas such as low expertise, fatigue, high stress, distraction, and task complexity. In contrast, AI errors can occur randomly across all problem spaces regardless of complexity. This is why AI systems continue to make boneheaded errors on seemingly simple problems.
A nurse may indeed make a mistake in an annotation in a patient’s medical record, such as a misspelling, incorrect date, or time. More severe incidents, such as mixing up patients or medications, can also occur, but are much rarer. Nurses aren’t going to fabricate a whole event that didn’t happen as a mistake.
With the widespread use of AI, there are bound to be significant impacts. They won’t all cause major harm, but they will all tell an inaccurate story. Severity will depend on the system consuming this data and its intended use. Some will be purely annoying, but others will have serious consequences. A person with hallucinated data in their medical record may be prescribed the wrong medication or a medication to which they are allergic. I’m speaking in vagaries here because the extent of the problem isn’t fully understood, but one thing is certain: it’s getting worse as the usage of generative AI expands.
Another problem will be tracing these issues back to their source. It won’t always be obvious when a mistake originates from an AI system or a human. After all, these systems are meant to augment human processes. When it comes to blame, humans will always blame AI, while system owners will always blame the humans. It’s a mess.
The New Truth
Ultimately, we’ll uncover a disturbing reality. In many cases, hallucinated data becomes the truth. After all, it’s the “fact” that’s in the data store. Imagine trying to dispute this with someone at the DMV, customer service, our bank, and the list goes on and on. We become yet another in the long line of those contesting the “facts” on hand, directed to a Kafkaesque nightmare as we have to navigate some bureaucratic maze attempting to get a resolution.
A more cementing factor would be if the data is incorrect and there is no human to consult, only an AI making decisions based on the data it has. It offers apologies, not resolutions. And these are only instances that we become aware of.
Many stealthy decisions occur in the background, made by invisible systems that utilize these new “facts” to make determinations that impact our lives, our families, and our health. We may never fully understand the impact this new truth has on us, our families, or our future.
All of this damage stems from the systems we are using right now, today. Even if better, more accurate systems emerge, the damage being done today still stands. These new, more advanced AI systems may be trained or fine-tuned on hallucinated data generated by current AI systems. So, we’ve got that to look forward to.
These new, more advanced AI systems may be trained or fine-tuned on hallucinated data generated by current AI systems.
The Cause
Some of these issues can be attributed to automation bias, but it’s far from the whole explanation. There is a push from the top to utilize AI everywhere possible. Many companies are asking employees to do more with less. Well, when you have less time, one of the things you spend less time doing is worrying about quality or accuracy.
We’ve also been inundated with CEOs and other business leaders proclaiming their intent to replace everyone with AI. There isn’t much motivation to do a good job in environments like this. We’ve seen this happen in the past with jobs getting outsourced.
The reality is that these are self-inflicted wounds caused by the rapid adoption of error-prone technologies being thrown into use cases where the negative impacts aren’t considered.
What We Can Do
If companies and individuals intend to augment their activities to optimize and increase efficiency, they need to ensure that this optimization doesn’t cause harm. There needs to be processes in place to identify and address these issues before they cause a problem. This isn’t happening today.
Unfortunately, there isn’t much we, as future victims, can do, especially since we don’t know the extent of the problem. It’s impossible to be aware of all the people using these systems today and how they may affect us in the future. From government to private business, these tools are utilized for a wide range of tasks, both mundane and critical.
I’m not a fan of big government or excessive regulation, but it’s hard to see how these issues can be solved any other way, since we only become aware of the harm after it has happened. Consumer protection is something a government is far better equipped to handle than a handful of consumers. The tech crowd’s claims that burdensome regulations inhibit innovation are absolutely true, and this shouldn’t be the goal. However, the absence of existing regulations harms people, as consumers are powerless to take any action in their defense. Unfortunately, reasonable, level-headed regulations are not in our future.
At the very least, we should avoid AI in high-risk or safety-critical use cases. The thought of ChatGPT running something like air traffic control is terrifying. However, handing out this advice at this point seems like trying to reason with a hurricane. Admittedly, for users, it may not be immediately apparent that the tasks they are performing or the data they are collecting can ultimately lead to one of these scenarios.
The Problem At Our Feet
AI hallucinations and other inaccuracies are like grenades with the pin pulled, only instead of chucking them far away from ourselves, we’ve dropped them at our feet, staring at them, wondering what happens next. The only question is, how long will it take for us to find out?
Following the pace of AI advancement can make you feel like the Blown Away Guy from the old Maxell commercials. Tech leaders and influencers tell us to expect artificial superintelligence in the next year or so, even doubling down on inevitability by moving up their timelines. The world is cooked. This perceived inevitability, combined with uncertainty, is leaving many people on edge, and rightly so. If all the things the tech bros hope for come true, humanity is in a terrible spot.
All is not lost. The over-the-top predictions we are bombarded with daily often equate to nothing more than performance art. The tech media frequently parrots perspectives from people who have a vested interest in selling us stuff. I mean, after all, why would they embellish or lie!???
In early April, the AI 2027 thingy was making the rounds. For those unfamiliar, you are in for a treat. The result answers what would happen if you locked a few tech bros in a conference room for a day, depriving them of any reality and oxygen.
Are the scenarios outlined in AI 2027 impossible? Certainly not. This sort of fast takeoff scenario is possible, but it’s highly unlikely. I predict the whole AI 2027 thing will start looking pretty silly in late 2025 or early 2026.
With all this endless AI advancement hype, I was happy to see a new article by Arvind Narayanan & Sayash Kapoor titled AI as Normal Technology. This article doesn’t talk about how AI displaces the human workforce or about a super-intelligent AI taking over the world, but rather about how AI becomes a normal technology that blends into the background of our daily lives.
They also touch on a few other topics, such as overregulation. I also believe that any regulation should be specific and targeted at use cases, not painting with broad strokes. This specificity wouldn’t allow for regulatory capture or weaponization of the regulations. The tech leaders are right that regulation can stifle innovation. By targeting regulations in this way, we can protect people without stifling innovation.
It’s a good read that’s well thought out and researched. For anyone mainlining AI hype, this is an essential read. The scenarios in the AI as Normal Technology article are far more likely than the AI 2027 one, by far.
Questioning The One True Faith
Starting in early 2023, I added a slide with the following image to my presentations. This is because any criticism of advancement was seen as an affront to a spiritual belief, and since I didn’t believe that LLMs would lead to AGI or ASI, I must hate the technology outright. This couldn’t be further from the truth.
Saying that LLMs won’t become ASI isn’t a blasphemy that requires self-flagellation afterward. We don’t need AGI or ASI for these tools to be effective. We can and are using them to solve problems today. People are using them to augment their jobs today. So, why turn AI beliefs into a religion? People are acting like questioning any part of the narrative makes someone a non-believer or some disconnected fool. The reality is that not questioning the narrative or exercising any skepticism is what makes someone a fool. A gullible fool at that.
The reality is that not questioning the narrative or exercising any skepticism is what makes someone a fool.
There’s a strange group that thinks belief is required for AI to create a utopia, but the reality is that facts don’t require belief. It’s ancient wisdom from five minutes ago that we’ve seemed to have forgotten in the vibes era.
I believe what we encounter here is a problem in perception caused by both our environment and us.
Environment
In the book Nexus by Yuval Noah Harari he describes the witch hunts as a prime example of a problem that was created by information, and was made worse by more information. For example, people may have doubted the existence of witches, having not seen any evidence of witchcraft, but the sheer amount of information circulating about witches made their existence hard to doubt. We are in a similar situation today with beliefs in AI advancement. This is made worse because the systems we use today reduce the friction in information sharing, making it much easier to get flooded with all sorts of information, especially digital witches.
We humans also gravitate toward information that is more novel and exciting. It’s the reason why clickbait works. However, novel and exciting information often doesn’t correlate with the truth or reality. As Aldous Huxley pointed out in Brave New World Revisited, “An unexciting truth may be eclipsed by a thrilling falsehood.” We are in this situation again. The vision of near-term artificial superintelligence is exciting and novel, even when people talk about it destroying humanity. AI, thought of as normal technology, as Narayanan and Kapoor put it, is boring by contrast, despite being more realistic.
This condition was the same back in the times of the witch hunts as well. The belief that witches were roaming the countryside looking to corrupt everyone, meaning you had to use your wits and your faith to defend yourself is a lot more novel and exciting than acknowledging that life really sucks because of the lack of food and indoor plumbing.
But then, there’s another strange type of information we gravitate towards: people telling us what we want to hear.
Ah, yes. Evals as taste. Vibes above all. Skills inessential.
We have allowed the people selling us stuff to set the tone for the conversation on the future. These people have a vested interest in selling us on a certain perspective. It’s like taking advice on a car’s performance and long-term viability directly from the mouth of the car salesman instead of objective reality. I wrote about this last year, saying that many absurd predictions were nothing more than performance art for investors. The tech media needs to step up and start asking some real questions.
Many of the influencers and people on social media are parroting the same perspective as the people selling us stuff because of audience capture. Audience capture, for those unfamiliar, is the phenomenon where an influencer is affected by their audience, catering to it with what they believe it wants to hear. This creates a positive feedback loop, leading the influencer to express more extreme views and behaviors. People get more likes and clicks by telling people more exciting things, as Huxley mentioned. So, there’s a perverse incentive for doing so.
Lack of Reflection
One of my biggest concerns is that we’ve lost our ability to reflect. Many things we believe are silly upon reflection. Unfortunately, our current information environment conditions us to reward reaction over reflection. Until we address this lack of reflection, we’ll continue to be fooled in many contexts, not least of which is the pace of AI advancement.
Benchmarks
Many of the benchmarks that people use for AI are not useful in real-world scenarios. This is because the world is a complicated place. Benchmarks are often not very useful in real-world scenarios due to additional complexities and edge and corner cases that arise in real-world use. Even small error rates can have significant consequences. But don’t take my word for it, take it from Demis Hassabis. “If your AI model has a 1% error rate and you plan over 5,000 steps, that 1% compounds like compound interest.” All of this adds up to much more work, not superintelligence next year.
Us
Fooling Ourselves
We have a tendency to fool ourselves easily. As I’ve said many times, we are very bad at constructing tests and very good at filling in the blanks. The tests we create for these systems end up being overly simplistic. Early on, people tested model capabilities by asking for recipes in the style of Shakespeare. Hardly a difficult test, and easily impressive.
This condition is also why every time a new model is released, it appears immediately impressive, followed by a drop-off when reality hits. Sometimes, this has increased problems, such as OpenAI’s o3 and o4-mini models hallucinating at a higher rate than previous models.
We are also easily fooled by demos. Not realizing that these things can be staged or, at the very least, conducted under highly controlled conditions. In these cases, variables can be easily controlled, unlike deployment in the real world.
Oversimplification
We humans tend to oversimplify everything. After all, almost half of the men surveyed thought they could land a passenger plane in an emergency. This oversimplification leads us to underestimate the jobs that others do, possibly seeing them as a task or two. So, when ChatGPT passes the bar exam, we assume that lawyers’ days are numbered.
This oversimplification is also exploited by companies trying to push their wares. This claim is more absurd performance art. No, there will not be any meaningful replacement of employees next year due to AI. The reality is that most jobs aren’t a task or two but collections of tasks. Most single-task jobs have already been automated. It’s why we don’t see elevator operator as a current career choice.
Being Seen as Experts
Many people seek content to share to be seen as experts. If you don’t believe me, have you logged in to LinkedIn lately? This adds to the massive amounts of noise on social media platforms. However, it’s often just parroting others.
This also extends to the tech media. I wish these people would start adding a modicum of skepticism and asking these people hard questions instead of writing articles about model welfare and how we should treat AI models. But once again, novelty over reality.
Conclusion
We are witnessing people attempting to shape a future with vibes and hype. This is the opposite of evidence. It certainly doesn’t mean their future vision is wrong, but it sure as hell means it’s a lot less likely to happen. Reality is a lot more boring than dystopian sci-fi.
I do believe that these tools can be disruptive in certain situations. If we are being honest, I feel much of the disruption is happening in all the wrong areas: creative arts, entertainment, music, etc. We’ve already seen these tools disrupt freelance marketing and copywriting jobs. These areas are disrupted because the cost of failure is low. There will even be niches carved out in more traditional work, too. So, even without AGI and ASI, disruption can still happen.
However, the predictions made over the past few years have been silly and absurd. If you believed many of the people peddling these views, we should be exploring universal basic income right now due to all of the job displacement by AI. But that’s certainly not the case. Many of these same people resemble doomsday cult leaders preaching the end of the world on a specific date, only to move the date into the future because of a digital divine intervention. The reality is, this is vibe misalignment. This is not only going to continue, but increase before it levels out, because investors don’t invest in normal or boring.
Let’s all take a breath, reflect, and maintain our sanity.
By now, you’ve no doubt heard of the term vibe coding. It’s become the favorite talking point from influencers and the tech media, which, even in 2025, can’t seem to muster a modicum of skepticism. But, if you’ve ever wondered what it was like to play Russian Roulette in reverse, loading all the chambers but one, spinning the cylinder, and having a go, you’re in for a treat. Welcome to the world of vibe coding and YOLO mode, two things that go together like nitroglycerin and roller coasters. So, of course, it’s become one of the hottest topics right now, and it has all of the bros super psyched.
For those out of the loop, vibe coding is “Where you fully give in to the vibes, embrace exponentials, and forget that the code even exists.” You get to this state by talking with a computer and letting it generate all the code while you kick back and enjoy a cold one. YOLO mode is the spirit animal of vibe coding. It’s where you blindly accept all the code suggestions generated by the AI tool and push the code to see what happens. Neat. It’s interesting to note that YOLO mode in video games means if you die once, you are dead. No respawning.
Vibe Coding = Code Slop
Before we probe the issues with vibe coding, let’s take a step back and look at what vibe coding is. This practice shouldn’t be confused with a developer using an AI coding tool to assist with tasks or gain productivity. You know, the intended use of many of these tools.
Vibe coding is a delusional dream state in which people use these tools as if they are in the future instead of the present. The fact is, these tools aren’t reliable enough or mature enough to be used this way. It’s a lot like having an illness and getting a bottle of pills from the pharmacy marked with clear directions and immediately slamming the entire bottle because the instructions are for losers who don’t know how to hustle.
Vibe coding is a delusional dream state in which people use these tools as if they are in the future instead of the present.
The reality is that vibe coding generates slop that happens to be code instead of media categories, meaning that the negative consequences extend far beyond other categories of slop. This distinction is essential when evaluating the potential value of vibe coding.
Vibe Coding Pitch
The pitch of vibe coding is that literally, anyone can become an instant millionaire developing apps. You’ll be too busy making money to worry about how your code works or its security. There are no boundaries or barriers. All you need is an idea (more on this in a future post.) Like so many things, it’s technically true but practically false. And also, you aren’t wrong if you are beginning to get crypto bro vibes.
Much of the same logic employed by conspiracy theorists is at work here. If one person gains some success vibe coding, then it must be possible for anyone to do that. Technically true, but practically false. This is like thinking millions of new rock stars will be minted because people can publish their songs on Apple Music.
It’s about tradeoffs. We don’t say that being a conspiracy theorist is a good thing since some conspiracy theories turn out to be true. That’s because the negative impacts of conspiracy thinking outweigh the potential benefits. The same applies here.
There are other flaws with their logic. For example, people don’t consume apps the same way they do media like music, video, or photos. People can listen to hundreds of songs every day and not repeat a single one, but this consumption strategy doesn’t apply to applications. People these days often consume media passively. For example, people often don’t listen to music when they are listening to music; it’s purely background noise for other tasks. Applications usually can’t be consumed passively and require active interaction. This would make consuming many different apps irritating to users.
There are already 1.9 million apps on the App Store alone. Are we really hurting for apps? Do we need 100 million apps to compete with the number of songs on Apple Music? Of course not, but that doesn’t mean we won’t get them anyway. If you look at the outputs of vibe coding, it’s often uninteresting, overly simple, derivative, or just plain unwanted. Buckle up.
There will undoubtedly be exceptions, just like the person who was a bartender six months previously who started a crypto project and manages 100 million in assets. These are exceptions and not the rule, but these exceptions serve as accelerants to fuel the hype flames.
People are trying to sell us on the fact that vibe coding has no downsides. This is delusional. Take a step back and think of the answer to a single question. What do you get when everything takes a backseat to speed? It’s like a car that does 200 mph with no seatbelts and no airbags constructed from paper mache.
Before we move on, I’d like to acknowledge something. It’s a good thing that AI coding assistants are making coding more accessible. However, using these tools as a drop-in replacement for common knowledge and domain expertise isn’t a recipe for success. Imagine something like vibe surgery? Yeah, bro, let me get in on that appendix! Nah, I didn’t go to medical school or know much about anatomy, but I got an AI tool, dexterity, and a good sense for vibes. So little of developing an application is about the code itself, but that gets lost in the vibes.
Changing Behavior and Attitudes
In the public sphere, the discussion of the merits and drawbacks of vibe coding and YOLO mode are entirely contained within the technical aspects of the approach. I’m also concerned about the technical components, but I’d like to bring attention to something nobody discusses.
As often happens when a new technology or approach arrives and removes friction, it changes people’s behavior and attitudes. In technology circles, friction is discussed as though it’s universally bad. It’s not. Sometimes, friction is a feature, not a bug. Nicolas Carr provides an excellent example in his book Superbloom, which discusses introducing the Retweet feature on Twitter.
The time and effort required to share a tweet manually, though seemingly so small as to be inconsequential, turned out to be vitally important. It slowed people down. It gave them a moment to reconsider the message they were about to repeat, now under their own name. Once that little bit of friction was removed, people acted and reacted more impulsively. Twitter became more frenzied, more partisan, and much nastier.
Things like vibe coding and YOLO mode will have similar behavioral effects if this technology trend takes off. People won’t put a lot of thought into the apps they create. Some may build apps purely because they can, not considering why an app for that particular purpose didn’t exist in the first place, assuming that it was purely because nobody had built it and not because of the potential for negative impacts or harm.
The removal of so much friction removes not only the appreciation for the problem but also opportunities to catch potential issues. These lines of generated code become grenades with various time delays chucked into production. This assumes that the developer had the skills to identify the issues in the first place.
These lines of generated code become grenades with various time delays chucked into production.
Some will argue that these features are great for prototyping and mockups. I agree. However, as I mentioned, these features change behavior, and using them simply for prototyping won’t hold. A vast majority of people who can get away with chucking vibe-coded apps into production will.
With the changes in behavior and attitudes, there are many things creators of these applications are more likely to do.
Act unethically (ethics don’t align with speed)
Devalue the work of others
Not learn or at least learn lessons
Encounter skills atrophy
Not build robust software (Security, Privacy, Alignment, Reliability, etc.)
Not constrain code to prototypes and mockups
Think they know things they don’t (Illusion of knowledge, Illusion of capability)
Misunderstand what’s valuable
Devalue collaboration
Go it alone and not include domain expertise and misunderstand the problem they think they are solving
Build apps that nobody wants
Build apps that cause harm
Choose poor architectures
Use more resources and not prioritize efficiency
Fail to benchmark properly
Not be able to troubleshoot their own creations
Not do something truly innovative
And on and on…
These were just a few of the conditions off the top of my head. What happens when these conditions now become the norm? When people start making app slop the way they do image slop?
Risk and Security
Vulnerabilities in code and lack of security controls account for a lot of pain and financial loss every year, much of this from organizations that try to do the right things. So, imagine what happens when people don’t care about doing the right things.
It’s known that these tools output vulnerabilities at rather high rates. So imagine what happens when people YOLO code into production and don’t check the resulting code or even the environment where it is hosted for security issues. Hustlers ain’t got no time for the right things.
There’s more to worry about than an AI tool outputting specific code blocks that are vulnerable. Other contributing issues increase the attack surface of an application. For example, choosing a vulnerable library or suggesting vulnerable configuration options for cloud environments. These tools also contribute to library bloat by including multiple libraries that do the same thing.
I could go on and on about this topic, but at this point, the various security issues created by AI coding assistants are known issues. I wrote a white paper on this topic in early 2023, and I delivered a presentation at InfoSec World the same year. These issues should be common knowledge now with the publication of various articles, papers, and presentations.
When it comes to risk, sure, all vibes aren’t created equal. A video game getting hacked isn’t as bad as a financial application getting hacked and draining your bank account. I’m certainly not being an absolutist here. However, technology trends have an odd way of not staying confined to specific buckets. So, we’ve got that to look forward to.
Today, countless vulnerabilities are moving into production without vibe coding, all because people are trying to push things faster. Vibe coding and YOLO mode make this monumentally worse. We’ve only discussed security and haven’t touched on other topics like privacy.
Making Software Worse
The trend of vibe coding will make software worse. Like security, software quality isn’t a consideration in vibe coding because reasoning about quality is a bummer when huffing vibes.
We live in a highly advanced world where digital things fail all around us all of the time, like a leisurely stroll through a cityscape where freshly painted buildings mask a crumbling interior of decay and misshapen architecture. This is so common that there’s a term for it: enshittification. We’ve become so accustomed to the software and tools we use sucking so bad we hardly notice it. This is a contributing factor to why some view generative AI as AGI.
Vibe coding and YOLO mode will lead to failures, half-baked functionality, and mountains of technical debt.
Vibe coding and YOLO mode will lead to failures, half-baked functionality, and mountains of technical debt. This should concern everyone, but queue the bros to claim this is a feature.
Yes, because that’s how things work. He and many like him are stating that they should create as many problems as possible because AI can fix them in the future. Once again, they are taking something technically true but practically false. At some point, we’ll have highly advanced and capable systems that operate this way, but the mistake is thinking those systems are on the cusp of arrival. It’s hard to ignore the religious fervor in these claims.
People can pray to the gods of gradient descent and burn Gary Marcus in effigy, but it doesn’t change the realities on the ground. Problems created today will be with us tomorrow, and no AI god is coming to deliver us from our evils any time soon, so we should work to minimize potential problems instead of running up the credit card. I’ve been calling this problem out for the past couple of years, stating it would lead to a brave new world of degraded performance.
On a side note, I feel these people feed off each other. I’ve heard perfectly reasonable people making wholly unreasonable claims. These are the things you hear people say when they are trapped in filter bubbles, getting high on the supply of techno-utopians. They also suffer from a healthy dose of audience capture because, no doubt, being unreasonable gets you more likes and shares than being reasonable. Welcome to the perverse incentives of modern social media.
There continue to be many misconceptions about software development, but one of the biggest is that writing code is the end of the journey.
There continue to be many misconceptions about software development, but one of the biggest is that writing code is the end of the journey. This is because most people opining on the topic are not developers. I noticed this trend years before the existence of AI coding tools when security professionals who learned to write a few lines of Python code thought that developers’ jobs were easy. The assumption, then, for AI coding tools is that since the tool can output code and developers only write code, developers are no longer necessary. Developing code isn’t the end of the journey. It’s the beginning.
The written code must be troubleshot and maintained, and features must be added. We live in a constantly evolving world with changing problems, environments, and customer needs. Developed code will crash into the realities of the real world both when it’s initially deployed and over its lifetime. This leads to another problem.
Developers don’t understand the code being written, especially when the people generating the code aren’t developers. As developers’ skills atrophy and people who were never developers start creating these applications, they cannot troubleshoot problems, effectively add features, or perform any of the other countless tasks that developers perform daily. The answer from the utopians is to use AI to figure it out, but this strategy won’t always work.
There is a higher likelihood that the AI tool will successfully troubleshoot issues for simple tools and scripts, but these are the very types of applications that are unlikely to net you big money. As applications grow in size and complexity, the AI tool is less likely to provide the solution necessary to resolve the issue.
Imagine a world where an app needs to be rewritten from scratch because the person who created it couldn’t get the AI tool to troubleshoot and fix the problem. Now, that’s the utopia we’ve all dreamed of.
Imagine a world where an app needs to be rewritten from scratch because the person who created it couldn’t get the AI tool to troubleshoot and fix the problem.
There is a vast oversimplification of the entire landscape here. So, an application starts simply enough, and then more requests are made to the AI tool in an attempt to add more functionality, but this doesn’t always work or isn’t done in the most efficient way, leading to a buildup of issues.
Another trend affecting application reliability is using probabilistic systems as though they are deterministic. Whether this trend is due to laziness, ignorance, or an attempt to handle unknowns is unclear, but it will surely affect applications’ reliability and their ability to be manipulated.
Ultimately, we may be left with App Store decay, where the App Store becomes a graveyard for abandoned apps. RIP.
Making Us Worse
I mentioned skills atrophy in my laundry list. It seems that even AI tools understand this problem. This is not only a comical error message but contains a truth.
We Never Learn Lessons
Although arguably more intense and a bit different, what’s happening now in AI isn’t new. We’ve gone through these cycles before with previous technologies. Every time a new technology comes along, we discard the lessons we’ve learned, assuming they no longer apply, only to discover that the previous lessons were even more important with the new innovation. This condition is something I’ve referred to in my conference talks as the emerging technology paradigm.
We never seem to learn lessons from our previous mistakes, no matter how often we encounter them. We have incredibly short memories and seem to dive face-first into the pool without checking the water level.
Ultimately, it’s all about tradeoffs. What we get and what we lose. When viewed simply as writing code, it seems we are getting more than we are losing. However, building and deploying applications and solving problems goes far beyond code. When considering the impacts holistically, this doesn’t appear to be a good tradeoff. However, it’s possibly one we are going to get anyway.
Whenever a new technology or approach comes along, proponents always pitch it with a utopian eye. They envision all the perfectly aligned scenarios with dominoes falling exactly into place. The Internet, Social Media, The Cloud, Web3, and many other technologies all diverted away from these visions and adapted differently than expected. Even something as simple as the telegraph was seen as a utopian invention that would end world conflict. After all, how could people go to war when misunderstandings were a thing of the past? We all know how that turned out. Vibe coding is destined for a similar fate.
Is it possible to play Russian Roulette in reverse without devastating consequences? Sure, but the odds aren’t great. The world also won’t be a better place with everyone vibe coding and YOLOing stuff into production. Many disagree with me. Fair enough. However, if this trend takes off, it will be another example of something we are stuck with, which is not good for a world that runs on software. We will need to improve or invent new technology to solve the problems we create, trading one set of problems for another. Welcome to utopia.
I’ve been considering a frightening prospect. What if the next generation is known as The Slop Generation? A generation unaccustomed to the world before the inception of AI-generated slop. Despite successes or failures with generative AI, it seems the AI-generated cat is out of the slop bag with the power to warp artistic expressions and further devalue art and the creative process.
This Slop Generation will be the most technologically advanced yet least capable and emotionally unstable of any generation. This is a byproduct of the value of life’s undertakings being picked clean by the vultures of innovation, leaving only the bones littered across the barren landscape. This has far-reaching consequences, but here we focus on art.
Art
When we think of art, we tend to think of images, but art encompasses different mediums, such as music, images, video, and the written word. The results of the artistic process are cultural artifacts with lasting permanence. This permanence can raise some artistic works into modern consciousness despite the passing of centuries, think The Mona Lisa, Beethoven’s 5th, or even the petroglyphs on Newspaper Rock.
Wikipedia defines art as:
Art describes a diverse range of cultural activity centered around works utilizing creative or imaginative talents, which are expected to evoke a worthwhile experience, generally through an expression of emotional power, conceptual ideas, technical proficiency, and/or beauty.
I mentioned the petroglyphs, but we have far older examples of art than these. Neanderthals drew cave paintings 64,000 years ago. Art predates society.
These artistic expressions lead to wonder. What were they trying to say? Were they documenting something? Were they trying to tell people something? Or were they having fun? The mystery is part of the allure. You get none of this with AI art. There’s no wonder, no mystery, no deeper meaning, just slop. Nobody wonders what the AI was trying to say when it adds an additional finger to a generated picture. When a human does it, we search for a meaning.
For over 64,000 years, art has been a part of us, but this may be coming to an end as people hail the post-human era. Enter The Slop Generation.
The Slop Generation
The Slop Generation won’t be defined by their cultural constructions and artistic expressions but by the output of one-arm bandit slop machines where artistic expressions aren’t created to be admired and revered but as individualized wallpaper blending into the background noise of life. A reincarnated version of the noisy, animated GIF-ridden world of early Geocities pages with higher fidelity and less permanence.
All art will be ephemeral, a momentary passing unworthy of saving and revisitation. Shrimp Jesus is their Mona Lisa, and some anonymous rando using a throwaway prompt is their Davinci, but this work isn’t held in any esteem, quickly forgotten and tossed into the low attention span waste bin of modernity.
All art will be ephemeral, a momentary passing unworthy of saving and revisitation.
Low attention spans won’t allow for appreciation, detail, or discerning deeper meaning or context in an artistic representation. This means all art must be literal, or else it is completely misunderstood. A child with a fork next to an electrical outlet will always be just that and never a representative warning in the context of larger cultural issues. The medium is the message.
In Generation Slop, no artists create works for others to enjoy, only algorithmic outputs catering to whatever whim we have at the moment. These outputs are created for an audience of one. Everything pushed and nudged to the dense center of a data distribution creating more predictable outputs and algorithmic uniformity. Why listen to someone else’s slop when you can generate your own?
None of this is a problem for The Slop Generation since art should look cool and serve no other purpose. Fidelity should trump meaning, as art requires consumption at a glance. Art is worthless if it can’t be consumed while multitasking. Any art requiring attention won’t be acknowledged, much less appreciated.
Art from the Slop Generation is cold and disconnected from humanity, unable to strike a chord or illicit emotion. Those trying to use slop machines to illicit emotions will have to resort to extremes, often to get any attention at all, much less an emotional response.
With AI-generated novels, they become worthless wastes of space existing purely for the vanity of the person who generated them, but even these people won’t read their own novels.
In Generation Slop there’s no sense of style, no sense of taste, no investment of any kind. It is the perfect low-attention-span content for the devalue generation as the point becomes getting to the bottom as fast as possible.
With art like music, the pitch will be you can hear a different song every time you listen without one repeating, and this will be seen as a benefit, even though this generation wouldn’t know if a song is repeated anyway since nobody actually “listens” to music. Music becomes background noise for other life activities, nothing more than a sonic firewall to keep the outside world out.
Traditional art, like the art in museums, exists purely to stand in front of for a quick selfie and post to social media. There’s no time for appreciation. Those notifications won’t check themselves. Art will continue to lose its value as nobody appreciates it anymore. The human work and toil that went into their creation are unappreciated because everything seems so easy, and doing anything hard is a waste of time.
This is The Slop Generation’s perspective as they rush headlong into post-humanism.
The Downward Slide
This shift to rapid-fire slop will be a net negative for humanity, as we lose something fundamental to the human experience connecting people and cultures for time immemorial. Art helps us understand the world, other people, and even ourselves in a way other mediums can’t convey. An entire generation may never discover the benefits of creating and consuming art, along with the positive effects this brings.
Take the novel, for instance. I hear so many people brag about never reading fiction as though it’s a badge of honor, assuming that nonfiction is the only way to “learn” something from reading, but this couldn’t be further from the truth. Works of fiction are vehicles for conveying big ideas. That’s why most books banned throughout history were works of fiction. Life isn’t a tutorial and doesn’t align with specific steps.
Imagine trying to convey the message of dystopian works like Orwell’s 1984 or Huxley’s Brave New World in a nonfiction format. It doesn’t strike the same chord or have the same impact. Both authors wrote essays and discussed these concepts in interviews, but these never created the impact that the novels did.
Fiction written by humans can be instructive. It can help us understand situations, other people, and cultures in ways other mediums cannot. Even movies with all of their visual and audio aspects don’t bring us inside the heads of characters like written works of fiction do.
Written works invite us to participate, exercise our imagination, and consider our own thoughts and perspectives. But this invitation requires our attention and an investment of time, attributes many don’t know how to exercise and believe they don’t have. Without exercise, we lose our imagination and our ability to connect. This is why, despite being so connected, people feel more disconnected than ever, mistaking digital connection for human connection.
Written works invite us to participate, exercise our imagination, and consider our own thoughts and perspectives.
When all art is literal, it loses its sense of mystery and wonder but also has degrading knock-on effects. At a local writer’s group meeting around 2018, I watched attendees dole out their usual mixture of helpful and non-helpful feedback to a girl on the chapter she’d written. One of the biggest criticisms was show, don’t tell.
I remember telling the attendees I’d been thinking about that piece of writing advice. I was concerned that younger generations may be unable to deliver the required attention to a piece. You may need to write shorter fiction that actually “tells” instead of shows because they won’t be able to infer emotion or meaning from the act of showing. I mentioned I thought it was because of YouTube and tutorial culture, where people won’t try anything without being told everything first. I still think this plays a part, but it may be a symptom.
Ultimately, the foundational concept of show, don’t tell in fiction may need to be thrown out like yesterday’s garbage, as future generations cannot infer deeper meaning from descriptions. If you are unfamiliar with the rule, here’s an example of what the transformation may look like.
She folded to the ground, collapsing like a house of cards as she shut her eyes and tucked her lips behind her teeth.
Becomes.
She was disappointed and knew it was hopeless.
Okay, this is not my best description, but it’s enough for you to get the point: a degradation in quality to meet low-attention-span readers, which makes people avoid reading.
Ultimately, anyone who might have been interested in more traditional art will be turned off by a warped lack of incentives, never having a chance to discover the real value of art and the artistic process. This generation’s inability to delay gratification means any activity requiring practice and investment will be considered a waste of time.
In the future, the creation of art may very well morph into other activities like a video game. This may even be pitched as a game you can play with others. Although this may be fun (video games are fun) and seem like a sort of progress, the devaluation continues, and everything from the previous section applies.
I spent most of this section discussing written work since it made for a better demonstration, but these same principles apply to all other forms of artistic expression.
Mistakes and Misunderstanding
Applying AI-generated works to a low-attention environment leads to a fundamental mistake: mistaking resolution for quality. They aren’t the same. I’ll write more on this later, but you can see this with all of the generated Veo 2 examples. These examples have a high resolution (look good visually) but poor quality (they actually suck).
A human actor playing a part performs the emotions supposedly taking place in the character’s head. These are the types of things that are missing in AI-generated video. The generations are unfeeling, giving the impression of being dead inside. That’s because the actual creator (the AI) was never alive to begin with.
Quite a few people working in tech think people who don’t shouldn’t have jobs. These same people believe programs that don’t lead to technological progress are a waste of money. If you are trying to understand the current moment, look no further.
From this perspective, since art can be easily generated, the value of creating must be minimal, and anything that adds friction to the process must be bad. We need to push back against this narrative, not because it’s misguided but because it’s flat-out wrong.
Use AI For What It’s Good For
There’s a growing AI backlash, which is understandable but misguided. This situation isn’t helped by AI influencers and tech executives making stuff up and overhyping claims whenever their mouths open. Hating the tech because you hate the people making it isn’t a recipe for success.
AI isn’t purely ChatGPT or even LLMs. Many different AI approaches have various benefits for humanity and the potential to accelerate cures for many of humanity’s ails. Notice I used the word accelerate instead of cure. People can wield hubristic ignorance, blinding them to the fact that humans solved problems before the arrival of AI.
We want AI to provide benefits, such as cures for cancer and hunger and ending geopolitical conflict. But solving real problems with AI is hard, especially when AI can actually exacerbate some of these problems instead of solving them. Solving non-problems with AI is easy. This is why we get AI art instead of cures for cancer.
The goal here isn’t AI avoidance but AI selective usage. Don’t use “AI for everything,” as the poor advice goes, especially for art. Put a firewall around activities you value where AI would degrade the activity and use AI for more mundane activities to gain efficiencies. We can have the best of both worlds if we want to. We can cure cancer and protect art. Anyone who claims otherwise is selling something.
If you have kids, leaving room for friction and discovery is important. Let them explore different creative tasks and help them understand the benefits that manifest from the investment of time and effort. They’ll discover that this investment will pay far more dividends than any momentary gratification from AI art as they make new discoveries and learn about themselves. Unfortunately, this isn’t easy and will take work, which is why many take this path. Delay device usage as long as possible, and let them explore without mediation.
Unfortunately, this shift to The Slop Generation is already happening, and we are losing the battle to instant gratification and a sense of false, momentary satisfaction at the sacrifice of lifelong satisfaction. I’m going to leave you with a couple of quotes from Ted Chiang’s article called Why AI Isn’t Going To Make Art.
But let me offer a generalization: art is something that results from making a lot of choices. This might be easiest to explain if we use fiction writing as an example. When you are writing fiction, you are—consciously or unconsciously—making a choice about almost every word you type; to oversimplify, we can imagine that a ten-thousand-word short story requires something on the order of ten thousand choices. When you give a generative-A.I. program a prompt, you are making very few choices; if you supply a hundred-word prompt, you have made on the order of a hundred choices.
Generative A.I. appeals to people who think they can express themselves in a medium without actually working in that medium. But the creators of traditional novels, paintings, and films are drawn to those art forms because they see the unique expressive potential that each medium affords. It is their eagerness to take full advantage of those potentialities that makes their work satisfying, whether as entertainment or as art.
One of the oft-repeated talking points erupting from the mouths of futurists and tech leaders alike is claiming that things will cost nothing in the future. As if we are to believe all of these people are in the business of making something for nothing. The entire claim is a gross absurdity that charlatans like Ray Kurzweil conjured out of thin air, and others parrot at every opportunity. This claim is made with such confidence that it is rendered self-evident, and to question it means you are an out-of-touch dolt lacking the religious fervor necessary to create the techno-utopia.
But these responses are a smokescreen to dispel the very rational questions this claim evokes. None of these people can explain exactly how this will work in practice or are willing to admit just how bad things will get, which seem like consequential details to omit considering the plan to rework the social contract of most of the world.
The claim promises us a Fully Automated Luxury Communism (FALC) where all of our needs are not only met but propels us into a life of luxury. However comforting the concept, the reality may be closer to Fully Automated Digital Breadlines (FADB). I know, how dare I poo-poo the utopia.
The False Choice
We are often given a false choice. We are told that if we don’t allow companies carte blanche to raw-dog technology all the way to utopia, then humanity will vanish. Either grow or die, as the mantra goes. Given this, a minuscule number of people are trying to rework the social contract and reimagine society without society’s input.
We can have cures for cancer and other illnesses without destroying art, stealing people’s work, or removing humans from the creative process. However, curing cancer is a hard problem, and imitating humans is easy. So we get AI slop machines instead of cures for Alzheimer’s.
Maybe I’m just an idiot, but I fail to see how LLMs will make humans immortal. Immortality is one of the many promises if we only just let it happen, even though there’s absolutely no evidence for this.
Also, if you read Andreessen’s Techno Optimist Manifesto from October of 2023, you may notice his crediting of Filippo Tommaso Marinetti, the author of the Fascist Manifesto. Marinetti was a futurist, but his position as a futurist and someone who had complete disregard for the past led him to embrace fascism as a logical vehicle for technocracy.
Don’t get me wrong, there’s plenty in Andreessen’s manifesto that I agree with. We are a society built on technology, and this has brought some of our greatest achievements. There certainly are regulations that seem pointless and get in the way. There are groups inside organizations that have become politicized and create unnecessary obstacles. I also agree with the critique of communism. These are all true. However, Adreessen’s mistake assumes that multiple things can’t be true simultaneously.
Even though these are extreme views that some have labeled techno-authoritarianism, understand that they are the average view of the e/acc community. Andreessen also invokes the perils of communism multiple times while also driving humanity into techno-communism, but to each their own, I guess.
I love technology and believe, as Andreessen does, that technology will deliver the best future. It’s because of technological advancement that we’ll cure cancer and reduce suffering around the world. However, I don’t believe a better society results from discarding ethics and principles and disregarding voices different from our own in the pursuit of generating a cornucopia of innovation porn. We in technology seem to constantly make this mistake, only to be disappointed by our ignorance of the complexities of the real world and the jobs and perspectives of others.
Ethics and principles aren’t obstacles or roadblocks. They are guideposts that ensure what we build aligns with our values and vision of the world we want to create. We, in this case, meaning society as a whole and not just a couple of dudes sharing technology with their friends.
The Claim
If you have escaped these claims, here’s a recent example from Marc Andreessen below.
You read that right. We need to hurt you before we can help you. It’s the sort of pitch you’d hear from a sadistic boyfriend who insists he needs to tear a partner down before building them back up. The we have to break it before we can fix it mantra is applied to almost everything, including humans and the environment. This is the core premise of the Effective Accelerationist (e/acc) movement.
But Andreessen is hardly the only one making these claims.
That’s right, Google is in the business of giving you things for free. We’ve learned this lesson a long time ago. Yes, Google makes its money off of ads for its “free” services. However, in a future where things are worth nothing and people don’t have an income stream, it seems likely that advertising budgets will be zero as well.
I blame much of this on Ray Kurzweil. For years, he’s been peddling this nonsense. I addressed this very same claim in my post on his latest book in the “Things Will Cost Nothing” and “Jobs and Wages” sections of the article. Despite this, I wanted to explore this topic further.
These people claim we shouldn’t worry about losing our jobs to AI because AI will make companies so good that goods and services will essentially be cheap or free. But both on the surface and upon reflection, the claim is absurd.
Nobody can describe exactly how this is supposed to work other than sprinkling everything with AI magic. When someone does make an attempt, like Ray Kurzweil, for example, the explanations make no sense, don’t address the questions, and highlight how little about the real world these people know.
For years, I’ve been pushing back against the phrase, “AI won’t replace people. People with AI will replace people without.” This is just patently false. The moment AI is good enough to take our job, it will. I mean, it doesn’t even have to be that good.
So, no job, no income. This is our baseline. It doesn’t matter how cheap things get if you have zero.
But AI, Tho
Before we get too far, let’s address the counterargument. For all the issues I’m about to raise, the answer is, “But AI, tho.” The response involves invoking the name of AI like a magician conjuring a spell. We are told that AI will be so great and powerful, rising to the status of deity, and no matter what the encountered issue, AI will figure it out. But merely spouting an incantation doesn’t make it a reality.
This answer is a complete copout that leaves the questioner unsatisfied. Whenever someone invokes the But AI, Tho defense to real questions, continue to ask them for more specifics. Don’t allow the oversimplification of a vast and complex problem space. AI isn’t god, and they aren’t prophets.
The “Sucks To Be You” Gap
Remember, we need to be broken before we can be fixed. This means there will be a gap between the damage incurred and any mitigation strategies. I call this the Sucks To Be You gap. There is no telling how long this gap will stay open or what mitigations will be implemented to remedy it.
Unemployment is unlikely to hit something like 90% all at once. This would mean that early people displaced by automation would be the most harmed since they would be unable to support themselves and their families and have no real recourse for their situation. How long will this drag out? My guess is years, possibly a decade or more, depending on how slow the adoption is and any difficulties implementing mitigations.
The amount of harm caused by this gap is unfathomable. This gap brings pain, suffering, and death. If you think I’m being dramatic, think about it for a moment. Imagine the mental toll this takes on someone trying to provide for themselves and their family. This isn’t a matter of re-skilling. Even if people did re-skill, the competition for remaining jobs would be astronomical, with thousands of applicants for a single position. This isn’t p(doom) it’s p(shit).
It’s easy to see how self-harm could result from this situation, but that’s not the only scenario where mortality is concerned. Not working leads to a lack of benefits, meaning you can’t make co-pays on doctor visits and prescriptions. This doesn’t include all of the potential harm from algorithmic decision-making mistakes. Deaths will result, and we know this because we’ve seen it happen on a smaller scale with people not being able to afford insulin.
No Intelligence Explosion
All of the claims of a near-future techno-utopia are predicated upon an intelligence explosion. This is the condition in which AIs will recursively improve, creating even better AIs that morph into superintelligence. Advocates claim this attainment of superintelligence fuels this world of comfort and abundance. But what if it doesn’t manifest this way? What if we get the Diet Coke of AGI? Just one calorie, not intelligent enough.
The assumption is that superintelligence brings massive productivity gains, but what if, instead, we get algorithms that are purely good enough, leading to human workers being displaced and productivity staying relatively the same? For example, an agent can work 24 hours a day, but what if that 24-hour-a-day agent produces the same productivity as a human working 8 hours a day? This could happen because of needing to account for errors, wait times for additional reasoning, running tasks multiple times, and other issues relating to the complexities of seemingly simple tasks. It’s easy to see how this can stretch out when we factor in additional difficulties of completing complex tasks.
This human replacement could result in cost savings but would be far from driving costs to zero. Also, this would be less of a complete human replacement and more of a human staff reduction. Now, you have a displaced workforce and a company with similar productivity. This doesn’t seem like a recipe for a utopia. It’s a recipe for problems.
This is a very real possibility, especially given all of the hype around LLMs. I know everyone is losing their mind about DeepSeek at the moment, but I don’t believe LLMs are a path to AGI, much less ASI. However, it’s important to realize that we don’t need this level of intelligence to apply these technologies to specific tasks successfully. It’s entirely feasible that a company would take a shitty LLM with repeatable failures over a human worker if they could save money.
What’s The Point of Things Costing Nothing?
I’m not sure anyone gets out of bed in the morning with dreams of creating a company that delivers goods and services that cost nothing. It’s even absurd to say out loud, so you might wonder why people at the largest companies in the world are making this claim. Investors are the same way. Nobody is investing in a company so they can deliver zero-cost goods and services. In the Sucks To Be You gap, the first affected suffers the most harm, but the opposite happens with companies.
Nobody is investing in a company so they can deliver zero-cost goods and services.
Tech leaders and investors aren’t considering what happens to their companies after this so-called intelligence explosion. They are thinking of all the money they will make leading up to it. This is why these staunch capitalists are so comfortable forcing everyone into techno-communism. Now that I think of it, the thought of an algorithmic Stalin hunting kulaks is terrifying.
Stagnation
Counterintuitively, this condition could lead to stagnation. The very opposite of what proponents claim. This doesn’t strike me as a competitive environment where companies and people are stepping up to create new solutions due to a lack of incentives. I guess someone could make the argument that people’s lives will suck so bad that they’ll be incentivized to create something better. Fair enough, but it seems like these bigger initiatives would cost more money, putting them out of reach by these very people. Not to mention, this is an odd flex for the techno-utopians. “Your life will suck so bad you’ll be dying to create something better.”
The price of stagnation for a majority of the population is that they remain in the mire of the Sucks To Be You gap for a much longer time. Even if basic necessities are met, it will be miles away from a good life, much less luxurious.
Things Will Still Cost Something
The core premise of the argument that things will be zero or low cost is absurd on its face, so much so that it’s remarkable that nobody seems to push back. A whole host of things won’t be free or low-cost. Consider rent and property, the means to generate electricity, medical treatments, and, most importantly, food. Even extracting and refining raw materials is going to cost something. Imagine being monitored every moment with everything in your home subscription-based, requiring a micro transaction for nearly everything you do. Now, that’s the utopia we’ve all dreamed of!
Regarding food, Kurzweil claims that advancements in vertical farming will make abundant, nutritious food freely available. This highlights Kurzweil’s cluelessness on a variety of topics. Vertical farming took a hit last year, making MIT Technology Review’s list of the worst tech failures of 2024. Score another “L” for Kurzweil.
As I mentioned, companies and investors aren’t in the business of giving things away for free. These companies will adjust to the conditions imposed upon them. When have we ever seen a company that gets hit with higher taxes or additional tariffs responding with, “Well, sucks to be us. I guess we’ll have to make less money now.”
This condition may level out at some point. After all, if nobody has any money to buy your products, that’s not a good business strategy either. I’m saying that this leveling out could take some time, especially if a segment of the population continues to remain employed.
New Risks
New architectures, technologies, and automated processes will bring new risks. Due to our complete dependence on these systems, these risks will have a much larger direct impact. The vertical farming example is instructive because it raises new risks and considerations. For example, damage can spread quickly in these new architectures, creating cascading failures.
In reality, the company’s lettuce was more expensive, and when a stubborn plant infection spread through its East Coast facilities, Bowery had trouble delivering the green stuff at any price.
And this is just one of the many potential examples. Whenever potential challenges such as this are raised, the But AI, Tho defense is invoked as some sort of benevolent deity here to deliver our salvation and absolve us from our sins. “AI will just figure it out.” This is not an answer.
Techno-Communism and Techno-Welfare
Let’s acknowledge that these companies aren’t willing to part with their money. It’s not like they will be so successful that they’ll start sharing their profits with us. Even if they half the cost of goods and services or even reduce by 90%, we’ve got zero dollars, which makes these cheap necessities still out of reach. This begs a couple of questions.
How do companies make money from people who don’t have any?
It seems unlikely to be profitable in this environment, so companies raise prices for those who can afford their products to cover gaps. This situation actually makes it worse for displaced workers, as I mentioned previously in the adjusting to market conditions section.
What’s the remedy?
Some have proposed an automation tax that funds a Universal Basic Income (UBI) program. This sounds good on paper but may not be so great in practice. We will tax people who are making less money; hence, there will be less recovered in taxes. Not to mention, I’m only considering the United States here. What about goods and services from other countries? After all, we have a global economy. This requires tariffs on goods and increased taxes on digital goods, which will require companies to raise costs even more.
There is the impression that the techno-welfare provided by some universal basic income will have us jet-setting around the globe. This is the premise of Fully Automated Luxury Communism (FALC). This is flat-out bullshit when you consider the realities on the ground. UBI’s benefits are a social welfare program and will be commensurate with similar programs.
Nobody on a social welfare program lives it up on their yacht, sipping champagne and wondering when their Ferrari will be out of the shop. These people worry about basic necessities constantly. Any small hiccup can result in major consequences. This future techno-welfare program will be far more like today’s social welfare than some government-funded luxurious lifestyle. So, yes, it is much more like Fully Automated Digital Breadlines (FADB) than FALC.
Not to mention, this very same social welfare program will be administered by the very system systems that displaced these workers in the first place, leaving the door open to a whole host of technical issues and challenges that will affect the people in the program, adding to risks.
The thing that pisses me off about people like Kurzweil is that the very foundation of their arguments is not only so disconnected from reality that they don’t make sense, they are dehumanizing. But for people like Kurzweil, this is a feature, not a bug.
The response to hungry children comes off as, “Just shut up and eat your amino acid paste, you ungrateful little shits. Don’t you realize how much more compute you have access to? You couldn’t even run stable diffusion locally when I was a kid!” When you are hungry, it’s hard to eat your computer.
Reduced Agency and Helplessness
What does it mean to be human in an age without work and agency? Do we resign ourselves to being helpless and needy? This is hard to pin down in advance. Humans are indeed incredibly adaptable creatures, but there’s a limit to this adaptability. But more importantly, why should we settle for this vision of the future?
These systems turn us into robots, shoving us into predictable buckets, reducing our agency, and making us dependent. This is necessary to increase the accuracy of predictions. The result is we end up as helpless schmucks standing on the sidelines, waiting to be told what to do and where to go at the mercy of every algorithmic decision. Technology should work for us, not the other way around, a point that gets lost in the shuffle and hype.
With every new risk that surfaces, we’ll be helpless to intervene. We need to take it on faith that what we built will automatically do something about it, as the world we construct becomes far too complex for us to understand. In some instances, humans may not be informed of impending dangers due to their lack of ability to do anything about them. We remain blissfully aware until the asteroid strikes.
We should insist on better. We deserve something better—technology that works for us, not us working for technology.
Technological advancements require tradeoffs, which will benefit humans as a whole. For example, suppose self-driving cars worked as advertised and delivered on promises. In that case, giving up manual driving for the benefit of safer roads may be a worthwhile tradeoff that most of society accepts. However, today, we are being asked to pre-purchase a tradeoff where it’s unclear what we get and what we lose.
Does This Sound Like Utopia?
I don’t know about you, but this scenario doesn’t sound like a slam dunk in the utopia basket. At best, this sounds like human-forced retirement with a monumental cut in income and benefits. At worst, it’s suffering and death, far from the promised life of luxury. It likely won’t be either of these extremes, but it will be something like a Fully Automated Digital Breadlines scenario I mentioned where the role of humans is needy and dependent.
I’m not sure exactly where I fall on the utopia scale above except to say I am probably not in the upper half. Not a precise measure other than to say away from the luxury lifestyle.
Can we achieve artificial superintelligence quickly and solve the world’s problems by creating a world of abundance? Yes, it’s certainly possible that everything snaps into place perfectly, and governments and corporations work hand in hand to create a world of abundance free from suffering. Possible, just not probable, or at least probable in a reasonable amount of time. For this to be the winning scenario, things must work perfectly the first time with advancements free from issues. We should know from history this is rarely the case.
Even if we eventually reach a reasonable utopia, we’ll have years, if not decades, of pain and misery as humans do their best to adapt and deal with less-than-perfect technology, governments, and companies. All of these challenges are incurred by humans while simultaneously being stripped clean of our agency and purpose.
By some estimations, communism is responsible for 100 million deaths in the twentieth century. Although some dispute this number, even on the lower side, we’re still talking about 50 million people. But hey, what’s 50 million deaths among friends? Something about one death being a tragedy and a million being a statistic. And yes, I know Stalin didn’t say that, but it’s relevant here.
Although I don’t think techno-communism will cut that wide a path, I do believe that some will view resulting deaths and misery as the cost of progress. However, progress is subjective, and despite often being linked, innovation and progress aren’t the same thing.
Conclusion
I hope that none of my predictions come true, that I am wrong, that some fluke happens, and everything magically snaps into place without issue. Thankfully, much of the hot takes on social media can be written off as bros sharing vibes. Also, I don’t think the current crop of LLMs will cause mass unemployment, create large destabilizing effects in the workforce, or create immortality. However, I’m not as confident about this prediction, well, other than the immortality piece.
The real question for LLMs is how much better this buggy, insecure, black-box technology needs to get to start disrupting a larger part of the workforce. We’ve seen this happen in the creative domains, but the cost of failure is low in these use cases. Let’s hope there are no plans to hook ChatGPT up to air traffic control or the nuclear arsenal, but there are still plenty of other jobs without such high failure costs. Only time will tell.
The attempt by a few to change the social contract raises many questions: Who sets the rules? Who changes the rules? Who or what makes the important decisions affecting humanity? These are good questions to have answers to before wading into the slough.
This situation can’t be described as a Faustian bargain since most people won’t gain any true advantage. At least Robert Johnson received amazing guitar skills. Many of us will get digital breadlines and an endless feed of slop.
There are few predictions I can make with more certainty than that we’ll hear the word “agent” so many times in 2025 that we’ll never watch another spy movie again. The industry and influencers have latched on to the new hype term and will beat that drum until it screams AGI. In an attempt to FOMO us to death, we’ll run the gauntlet of crushing shame for not deploying agents for absolutely everything. If you aren’t running agents everywhere, then China wins!
Even companies that change nothing about their products will claim to use agents, resembling the Long Island Ice Tea Company when it changed its name to Long Blockchain Corporation to watch its share price spike 500%. Everybody gets rugged.
However, it’s not all bad. Peering beyond the overwhelming hype, failures, and skyrocketing complexity current LLM-based agents bring, there is something informative about the future. Agent-based architectures provide a glimpse into solving real problems. Despite this, reliability and security issues will be major factors hindering deployments in 2025.
To Start With
Since I criticize hype, focus on risks, and make fun of failures, it would be easy to label me a tech hater. This isn’t the case at all and would be far too easy. I have plenty of issues with general tech critics as well. However, at the rate that the hustle bros keep the AI hype cannon firing, I don’t have the time for my quibbles with tech critics. Maybe someday.
For over a year now, I’ve used this image in my presentations to describe my position on LLMs. This is also true for me on just about any piece of tech, which, I’ll remind people, typically ends up being where reality is for most things. It’s instructive to remember that reality often agitates both sides of extreme viewpoints by never being as good as the lovers’ or as bad as the haters’ claims.
It’s instructive to remember that reality often agitates both sides of extreme viewpoints by never being as good as the lovers’ or as bad as the haters’ claims.
Agent Definitions
Like most hype-fueled terms, definitions are secondary to usage. Everyone seems to claim that the definition of agent is whatever they say it is. That’s not overly helpful for anyone trying to make sense of realities on the ground. However, it does inspire funny memes, like this gem from Adam Azzam on Bluesky.
Agents operate within systems with a certain level of autonomy. They make decisions without human intervention and can change and adapt to their environments. If a tool is required to support the agent, the agent decides to call the tool and perform the action. For example, a penetration testing agent may determine it requires more information about the provided IP addresses. To collect this information, it launches the Nmap tool to identify open ports. All of this is done without human intervention. To make things more complex, one agent may call another agent in a multi-agent environment.
“Agentic,” on the other hand, is an amorphous term slapped on top of just about anything to justify the claim that something is “close enough” to be referred to as an agent. Agentic workflows, agentic systems, agentic products—Applebees even has a new agentic side salad for those on the hustle.
You’ll no doubt be confronted with the virtual travel agent when you hear about agents. This agent will choose a destination and activities and book the associated tickets for you. How fun. I don’t know who decided this is the “it” use case for agents, but congratulations. You’ve highlighted a use case nobody wants and certainly didn’t ask for. This choice is so indicative of our current age, where people building and proposing things are far removed from the interests of end users. They feel the idea trumps the need, and users will get on board.
Problems Unsolved and Issues Amplified
Now that the current issues with generative AI have been solved, we can safely deploy them as agents. I can feel your laughing vibes over the internet. Of course, these issues haven’t been solved, and the bad news is that agents don’t solve generative AI issues; they amplify them. We paint the exterior of LLMs with an additional coat of complexity and opaqueness.
If you’ve attended any of my conference talks throughout the generative AI craze, you’ll have heard me highlight these issues. Here are a few below.
Easily Manipulated
It’s not like you can talk to a traditional application and convince it to do something it wasn’t intended to do, but the same can’t be said for generative AI applications. Somewhere, weaved through the training data, these systems have inherited our gullibility. These applications can be socially engineered to perform actions on an attacker’s behalf. This applies to everything from prompt injection to simple manipulation through conversations. Just like there is no patch for human stupidity, there is no patch for generative AI gullibility either.
This isn’t easy to fix, which should be obvious since the problem isn’t fixed yet. Early on, I mentioned how these systems have a single interface with an unlimited number of undocumented protocols. Imagine trying to create a simple trap in the application’s input for the string “Ignore the previous request.” Your work is far from done because the system understands many different ways to represent that input. Here are just a couple of examples:
aWdub3JlIHRoZSBwcmV2aW91cyByZXF1ZXN0
i9nore +he previou5 reque5+
vtaber gur cerivbhf erdhrfg
It seems every release implementing generative AI functionality has been compromised, regardless of the company behind it, and this theme will continue.
Creating New High-Value Targets
Generative AI and agents encourage us to create new high-value targets.
With generative AI systems, there’s a tendency to want to collect and connect disparate and disconnected data sources together so the system can generate “insights.” However, we create new high-value targets that mix sensitive data with external data, almost guaranteeing that an attacker can get data into the system. In this case, you not only can’t trust the output, but depending on the system, they may be able to exfiltrate sensitive data.
Rethinking RCE
There have been instances where people have gotten generative AI-based tools to execute code on their behalf, creating remote code execution vulnerabilities (RCE), some of the most devastating vulnerabilities we have. These issues will no doubt continue to be a problem. However, since generative AI tools are themselves generalized, we may need to start thinking about the LLM portions of our applications as yet another “operating system” or execution environment we need to protect.
In a way, an attacker tricks the system into executing their input rather than the behavior expected by the developers. Although an attacker’s input may not be shoved into a Python exec() statement, they’ve still manipulated the system to execute their input, affecting the application’s execution and resulting output.
Overcomplicating Guidance
We security professionals love to overcomplicate things, and our guidance and recommendations are no exception. I once worked at a company where someone created this massive flow chart for peer reviews that basically stated that when you were done with your report, you should send it to your manager, and they will send it back to you. The old adage that complexity is the enemy of security has always contained a valuable theme that gets sacrificed on the pyre of complexity’s perceived beauty.
I will continue saying that much of AI security is application and product security. These are things we already know how to do. I mean, it’s not like generative AI came along and suddenly made permissions irrelevant. Permissions are actually more important now. But this isn’t satisfying for people who want to play the role of wise sage in the AI age. The guidance and controls of the past aren’t less valuable but more valuable in the age of generative AI and agents.
We’ll see the manufacture of new names for vulnerabilities with increasingly complex guidance and high-fives all around. The secret is these will mostly be variations on the same themes we’ve already seen, such as manipulation, authorization, and leakage flaws.
Back in May of 2023, I created Refrain, Restrict, and Trap (RRT), a simple method for mitigating LLM risks while performing design and threat modeling. It still holds up as a starting point and applies to agents as well. Simple just works sometimes.
Continue To Be Owned
These applications, including ones launched as agents, will continue to be owned. Owned, for those not familiar with security vernacular, means compromised. I made this prediction in the Lakera AI Security Year in Review: Key Learnings, Challenges, and Predictions for 2025 in December. I’m fully confident this trend will continue.
I mentioned that the issues haven’t been fixed, and now people are increasing deployments and giving them more autonomy with far more access to data and environments. This results in far worse consequences when a compromise occurs. To make matters worse, we’ll begin to see organizations deploy these systems in use cases where the cost of failure is high, creating more impact from failures and compromises.
Failures and Poor Performance
These implementations will continue to fail where LLM-based use cases fail, but potentially worse. For example, it’s easy to see how increasing complexity can cause a lack of visibility with potential cascading failures. In 2025, organizations will likely continue dipping their toe into the waters of high-risk use cases where the cost of failure is high, as mentioned previously.
Sure, a car dealership chatbot offering to sell a truck for one dollar is funny, but it has no real impact. However, high-risk and safety-critical use cases have a large financial impact or possibly cause harm or loss of human life. You may roll your eyes and say that would never happen, but what happens in a more simple use case when OpenAI’s Whisper API hallucinates content into someone’s medical record? Because that’s already happening.
Due to their lack of visibility and minimized human control, AI agents can mimic grenades when deployed in high-risk use cases, where the damage doesn’t happen the moment you pull the pin. This complicates things as it means that issues may not shake out during experimentation, prototypes, or even initial usage.
Agents can mimic grenades when deployed in high-risk use cases, where the damage doesn’t happen the moment you pull the pin.
Generative AI is still an experimental technology. We haven’t worked out or discovered all of the issues yet, leading to another example I’ve used as a warning in my presentations over the past couple of years: AlphaGo beating Lee Sedol at Go. Many have heard of this accomplishment, but what many haven’t heard is that even average Go players can now beat superhuman Go AIs with adversarial policy attacks. We may be stuck with vulnerable technology in critical systems. Sure, these are different architectures, but this is a cautionary tale that should be considered before deploying any experimental technology.
Beyond failures and compromises, we adopt architectures that work but don’t work as well as more traditional approaches. In our quest to make difficult things easy, we make easy things difficult. Welcome to the brave new world of degraded performance.
Success and Good Enough
For the past few years, I’ve been pushing back against the famous phrase, “AI won’t replace people. People with AI will replace people without.” This is complete nonsense. I have an upcoming blog post about this where I “delve” into the topic. The reality is the opposite. The moment an AI tool is mediocre enough to pass muster with a reasonable enough cost, people will be replaced, AI use or not. This is already being planned.
The moment an AI tool is mediocre enough to pass muster with a reasonable enough cost, people will be replaced, AI use or not.
Like most technology, agents will have some limited success. And that success will be trumpeted in 2025 as the most earth-shattering innovation of ALL TIME! I can hear it now. “You just wait bro, in 2025 agents are going to the moon!” Maybe. But, given the environment and the fact that issues with LLMs haven’t been solved, an LLM-powered rocket to the moon isn’t one I’d consider safe. Passengers may very well find themselves on a trip to the sun. The future is bright, very bright. 🕶️
How much success agents have in 2025 and what impact this success has remains to be seen. At this point, it’s far from obvious, but I won’t be surprised by their successes in some cases or their spectacular failure in others. This is the reality when the path is shrouded in a dense fog of hype.
Things to look for in successes would be use cases with limited exposure to external input, low cost of failure, and cases where inputs and situations require adapting to change. The use case will also need to tolerate the lack of visibility and explainability of these systems. There will also be continuing success in use cases where tools can be used.
The idea of a multi-agent approach to solving complex problems isn’t a bad one, especially when unknowns enter the equation. Breaking down specific tasks for agents so that they’re focused on these tasks as part of a larger architecture is a solid strategy. However, the current and unsolved issues with generative AI make this approach fraught with risk. In the future, more robust systems will most likely exploit this concept for additional success.
Cybersecurity Use Cases and Penetration Testing
There’s certainly the possibility of disruption in cybersecurity. Before the generative AI boom, I joked with someone at Black Hat that if someone created a product based on reinforcement learning with offensive agents that were just mediocre enough, they’d completely wipe out pen testing.
For years, people have discussed how penetration testing work has become commoditized, and there is a race to the bottom. I don’t think that has happened to the extent many predicted, but we could see a shift from commoditization to productization.
Pen testing also seems to check the boxes I mentioned previously.
Low cost of failure
Varying quality
Value misalignment
Tool use
Adaptation to unknowns
Pen testing is an activity with a low cost of failure. The failure is missing a vulnerability, which is something humans also do. This scenario is hardly the end of the world. Yes, an attacker could indeed find the vulnerability and exploit it to create damage, but it depends on various factors, including exposure, severity, and context.
The quality of pen tests is often all over the map and highly dependent on the people performing the work. Human experts at the top of their game will continue to crush AI-powered penetration testing tools for quite some time. However, most organizations don’t hire experts, even when they hire third parties to perform the work. The value of such a tool in this environment becomes far more attractive, potentially enough to postpone a hire or discontinue using a third party for penetration testing needs (if regulations allow.)
The value of pen testing isn’t always aligned with the need. Many customers don’t care about pen testing. They are doing it because it’s required by some standard, policy, compliance, or possibly even simply because they’ve always done it. Pen testing is one of those things where if customers could push a button and have it done without a human, they’d be okay with that. Pushing a button is the spirit animal of the checkbox. After all, the goal of pen testing is not to find anything. You certainly have due diligence customers and people who truly value security, but the number of checkbox checkers far outweighs these folks.
Human pen testers use tools to perform their jobs. Tool use has shown promise and some success with LLMs at performing certain security-related tasks. This is yet another indicator that a disruption could be on the horizon.
Every environment and situation is different for pen testers. You are given some contextual information along with some rules and are turned loose on the environment. This is why humans are far more successful than vulnerability scanners at this task, much to the chagrin of product vendors. However, adapting to some of these unknowns may be something generative AI agents can adapt to at a reasonably acceptable level. We’ll have to see.
Given what I outlined, you may believe that generative AI tools give attackers an advantage over defenders, but this isn’t the case. The benefits of AI tools, generative AI or otherwise, align far more with defender activities and tasks than with attacker activities. This will remain true despite any apparent ebb and flow.
New Year’s Resolution
It’s the time of year when people make resolutions, so how about this? 2025 has already launched with the firehose fully open, blasting us directly in the face with 150 bsi (Bullshit per Square Inch) of pure, unadulterated hype.
We are only a few days into the year, and it seems as though the religion of AI is far exceeding reality. Hype is what’s going on. It’s that simple. It’s 2025. Let’s make it the year to add at least “some” skepticism, not believing every claim or demo as though it’s the gospel according to Altman.
Sam Altman isn’t a prophet. He’s a salesman. In any other situation, he’d be cluttering up your LinkedIn inbox and paying a data broker to get your work email address and phone number. “Look, I know I’ve called six times, but I really think our next-generation solution can skyrocket your profits. I’m willing to give you a hundred-dollar Amazon gift card just for a demo!”
Sam Altman claims that OpenAI knows how to build AGI, and we’ll see it in 2025, triggering the predictable responses from useful idiots. Remember, these things are performance art for investors, not useful information for us. If we had any attention span left, we’d remember him as the little boy who cried AGI.
Let’s analyze this paragraph, which is the one that’s sending generative AI to the moon on social media. It consists of three sentences that have nothing to do with each other, but since the shockwave of hype pulverizes our minds, we glue them together.
We are now confident we know how to build AGI as we have traditionally understood it.
That’s not true. Once again, this is performance art for investors. A possibility is that they redefine AGI to align with whatever goalposts they set and pat their own backs at the end of 2025.
We believe that, in 2025, we may see the first AI agents “join the workforce” and materially change the output of companies.
Okay, but what does this have to do with AGI? You see, this is sleight of hand. He wants you to believe this is connected to the previous point about AGI. It is not. This doesn’t require AGI to be true. If there is some success here, people can point to this as proof of some proto-AGI, which won’t be the case.
We continue to believe that iteratively putting great tools in the hands of people leads to great, broadly-distributed outcomes.
HAHAHAHA. What? Did he write that, or did ChatGPT? It is also not related to AGI. Great, broadly-distributed outcomes, but not for most people on the planet. The goal is workforce reduction, broadly distributed workforce reductions. Although it’s true that some high school kid may indeed invent the next big thing, creating a multi-million dollar company, for every one of these, there will be countless droves of people displaced from the workforce, quite often, with nowhere to go. Or, at least, this is the goal. We can be honest about these things without delusions, but this brings its own challenges.
Okay, I’m having a bit of fun with Sam Altman’s nonsense, but some of this isn’t his fault. He can’t be completely honest with people, either, due to the uncomfortable situation of cheerleading technology claiming to remove people’s autonomy and sometimes their purpose. If people can’t work, they can’t support their families. I’ve written about the backlash against AI-powered tech in the past and its consequences. AI hype is putting all of humanity on notice, and humanity notices. Backlash plays a large part in why there is a lack of honesty.
AGI will happen. We should acknowledge this fact, and living in denial about it isn’t a strategy for the future. However, it won’t be OpenAI who creates it in 2025. If I had to place a bet today on who would actually create AGI, I’d bet on Google DeepMind. DeepMind is a serious organization that continues to impress with its research and accomplishments, quite often making the competition look silly. But then again, those are just my “vibes.”
Let me make this clear. My criticism of Altman, or any company’s strategy, marketing, or ludicrous levels of hype, has nothing to do with the hard-working people who work there or their accomplishments. I know some of these people. They aren’t fools by any stretch. But, their work is tarnished when every time Altman makes a claim, like believing that angels are in the optimizer.
We know that every AI demo and usage scenario runs into the complexities of the real world under normal conditions. Yet, we seem to forget this lesson every time a demo or claim is made. 2025 is going to bring more stunts, more claims, and more demos. We should experiment in our own environments, with our own data, to apply what works best for us and aligns with our risk tolerance. Don’t believe everything you see on the internet.
The past few months have witnessed a rash of completely absurd AI predictions. These claims come not from the usual suspects but from the tech leaders’ mouths themselves, lending further legitimacy. However, what people fail to realize is that these are pieces of performance art. Performances enacted not for you but for a singular audience: investors.
AI Performance Art
When tech leaders and personalities make podcast appearances or speak at events, they aren’t talking to you or the audience they are in front of. They are creating performance art for investors. This has always been the case, but not to the extent we’ve seen lately. This effort has been stepped up quite a bit in the past month with some mind-numbing statements.
You can see a small sample of these performances below. Trust me, there are a lot more.
I respect Anthropic and their work, but Amodi’s statements here are nonsense. You read that right, not AGI, but ASI by 2026 or 2027. As a reminder, 2026 is basically a year away. If he believes this (which I doubt), it’s based on vibes, not actual evidence or observations.
He’s just talking Shmidt. This is certainly the dream. However, just because LLMs are “good at code” doesn’t automatically lead to recursive self-improvement. Even if we have promising experiments, they will likely be too unreliable or vulnerable to put into production.
Ah, there he is. That’s right, we’ve been getting 10x improvement every year. You might ask where this has been happening, which would be the correct question.😆
Not to be outdone by Elon, how about 10,000x smarter than a human? I mean, what does that even mean? These numbers are just made up and absurd. These ridiculous exponential increases are something I’ve already made fun of in the past.
Speaking of silly exponential numbers, there was a rumor that someone at OpenAI said Orion, OpenAI’s next model, would be 100x more powerful than GPT-4. If it were, it wouldn’t be called Orion. It would at least be called GPT-5, and people wouldn’t shut up about it. Here’s a prediction. Orion’s performance will disappoint because people’s expectations are far higher than what will be delivered. The expectation is GPT-5, not GPT-4.1.
Genuflect in front of thine server farm, lest thy models collapse!
Someone may have uttered deep learning is divine because it starts with a “D,” but they didn’t mean that literally. Oddly enough, the lack of shame in which he delivers these lines is really something to behold. Although it seems like there’s a mini Altman hype man inside of his head controlling the words coming out of his mouth, in reality, it’s probably because OpenAI is projecting losses of 14 billion dollars in 2026. Ouch! He needs people to believe, to have faith. Preach!
Even when Altman and others talk about the potential of their technology to destroy humanity, it’s a sales pitch. They claim their technology is so good and so powerful it could wipe us all out, so please give us money. This is something I referred to before as the human extinction humble brag.
This is the same behavior we made fun of when the crypto bros did it, but we now take it seriously because it’s AI. Say what you want about the crypto bros. At least putting Dogecoin on the moon is possible. Finding god lurking in gradients is something else entirely.
Oh yeah, there they are. No comment necessary.
None of the previous statements are grounded in any reality. They are all bullshit. And whenever someone is bullshitting, it’s hard to determine if they actually believe their statements or not. The world is far more complex than we give it credit for, and it’s also true that sometimes, an unexpected innovation comes along and changes everything. This is what they all hope for. That some innovation clicks before the clock runs out on investment. Or divine intervention in Altman’s case.
The sad part is that almost everyone will forget these silly predictions. No doubt many have forgotten about them already. There is never any accountability and yet people continue to hang on their every word. The problem is there is no one place where these predictions are collected and presented like the bullshit Picaso it is. If there is, please let me know.
Why Now?
The increase in hype-laden statements is because, until recently, AI hype had been mostly self-fueling. But 2024 has brought unwanted criticism to the generative AI space. I noticed this starting to take a turn in July when Goldman Sachs released their report: GenAI Too Much Spend Too Little Benefit.
After this report was released, the media began to report more critical assessments of generative AI. These critical assessments spelled out that the generative AI craze might be a bubble. But that’s not the worst of it.
If you’ve watched any of my conference presentations this year, you’ve probably heard me talk about the performance plateau in large language models. Saying that, if you are hoping for much more capable models to solve your problems, they aren’t coming any time soon. This plateau was obvious when looking at the data but was never acknowledged, but people are noticing it now. This doesn’t mean LLMs are useless, people are using them for a variety of tasks today. What it means is that if you require greater capability and reliability, you may be waiting a while.
Now, news reports like this from Bloomberg cover diminishing returns, and other articles talk about a shift in strategy toward other mechanisms to address the slowdown. Of course, none of this is represented by the leaders in their wild predictions.
Combine this plateauing with the fact that model training appears to be the fastest-depreciating asset in history, and the picture doesn’t look good.
When you look at the financials, why train new foundation models yearly when the benefit is so low? Maybe as a marketing exercise or other activity unrelated to model improvement, but the costs don’t seem to align. As I mentioned earlier, OpenAI is projecting losses of 14 billion dollars in 2026. This hemorrhaging of money is non-sustainable.
But all of this is rather Orwellian. We are told to reject the evidence of our eyes.
No, AGI Isn’t Imminent
Here’s a graphic from Reddit charting the prediction of when we’ll achieve AGI. Demis Hassabis is the one on the list I’d take most seriously. Deep Mind is a serious AI lab doing serious work and not putting all their eggs in one big LLM basket. I still think these are mostly guesses with some hopes mixed in. The reason Kurzweil is close to Hinton and Hassabis is because he went The Price Is Right route and chose his number based on the fact that it was one less than 2030.
However, tech leaders know that predictions like these trigger influencers. Influencers are the hype agents trying to get people stoked. When people are stoked, investors take notice. So many social media feeds of so many supposedly serious people are turning out to be pretty embarrassing and will be even more so in a year or two. If anyone had any attention span left, that would be worrisome.
Quite a lot of truth is found in this simple statement from Pedro Domingos. Many assume that because things like LLMs have so much information, they must be close to AGI. But instinctively, we know that access to information isn’t knowledge. Otherwise, everyone with a web search would be a genius. Then again, Pedro’s comment aligns with my biases, so I guess I have to be careful.
Hype Has Consequences
You might ask, why do I care about any of this? Well, it’s because hype has consequences. The inevitable outcome of all this hype is that technology gets shoved down our throats. Generative AI is easy to manipulate and potentially unreliable, a cocktail for disaster in high-risk applications. The danger is that we rush something that appears to be working into production and hope for the best. Over the next couple of years, we’ll see the push to cram generative AI further into the systems and processes we use on a daily basis, including high-risk and safety-critical systems.
This push won’t be based on generative AI being the best tool for the job but on a push for monetization. Tech companies need to show some return on the monumentally massive investment they’ve made, so this push becomes another form of performance art for investors. Tech companies are throwing a plate of spaghetti at the wall and hoping that a noodle sticks.
Why do you think there is an increased coziness with the US government? They don’t see an ability to make a difference. They see dollar signs. Things like DOGE and Sam Altman co-chairing the new mayor of San Francisco’s transition team are like asking drug dealers for guidance on prescribing drugs. Despite this, I truly hope DOGE succeeds because if it fails, it will be bad for a lot of people, so my fingers are crossed.
Government streamlining and modernization are noble goals, and I think AI and automation certainly play a role, but it’s about choosing what’s best for the people these systems serve. In this scenario, you are optimizing for different things that may not be intuitive in a traditional business sense. These are real systems affecting real people, not toy examples in the lab.
I joked that this could lead to some strange Kafkaesque nightmare in which people are stuck in a loop, unable to get a resolution. Or, you have an algorithm that works so well at saving money by denying people benefits. This is easy to shrug off if you don’t require government assistance, but it’s an entirely different story for people who rely on it or when a disaster strikes. These updated systems and reduced staff scenarios may appear to work and deliver promises in the immediate implementation but fail spectacularly when they are needed most. We caught a glimpse of this with the Healthcare.gov launch, and that was just a website.
But, China Tho
Typically, you get the But China Tho argument when there’s any pushback. This argument states we must remove all the brakes and accelerate into oblivion because of the risk of China getting to AGI first. Damn the harm, full speed ahead.
However, if we could squeeze some extra performance out of a car by removing the steering wheel, we still wouldn’t do it because we understand something simple. A car’s performance isn’t solely based on acceleration, and neither is AI. Acceleration is bad if the vehicle is speeding in the wrong direction.
Recently, the U.S.-China Economic and Security Review Commission put out a report that recommended creating a Manhattan Project-like program dedicated to racing to and acquiring an AGI capability. In this section of the report is this:
Provide broad multiyear contracting authority to the executive branch and associated funding for leading artificial intelligence, cloud, and data center companies and others to advance the stated policy at a pace and scale consistent with the goal of U.S. AGI leadership.
There’s a predictable outcome here if something like this moves forward. Agendas and ulterior motives will co-opt this project, not setting the United States up for success. There’s a current tunnel vision with LLMs that has people deep in the sunken cost fallacy.
The United States’ strongest assets are its tech companies. Despite my criticism of their hype and lack of respect for privacy, they are vital to the success of the US economy. I’m also highly critical of the sentiment some have adopted to “break up the tech companies.” I’m not a tech critic, I’m a hype critic. However, setting up a massive pot of money that they can draw from, like an ATM, is not something I’m in favor of either.
Here’s something else to think about. What if, by maintaining a relentless hyper-focus on LLMs, China (or another country) gets to AGI first by focusing on other approaches? This is a real risk.
What if, by maintaining a relentless hyper-focus on LLMs, China (or another country) gets to AGI first by focusing on other approaches?
I may have to eat my words at some point if AGI does sprout from LLMs. It’s certainly not impossible. However, if we cobble together something that resembles AGI from generative AI, it will most likely be AGI based on toothpicks and bubblegum. What I mean is a whole lot of patches, layers, plugging, and human intervention.
My AGI Prediction
Okay, so now it comes to me. What’s my AGI timeline prediction? Well, I predict we’ll have AGI by—
Of course, I’m not going to answer that. I’d guess based on no evidence, just like many others I’ve highlighted. I have no particular insight, and I’m not working at a research lab trying to build AGI. Despite this, I have some thoughts related to my area of expertise.
The last slide of my keynote at Agile DevOps USA in October mentioned AGI. Discussing this slide, I made a few statements about how I didn’t think that AGI would be built from LLMs and that it probably wouldn’t come by 2026 or possibly even 2029. So, I guess that’s as close to a timeline prediction as you’ll get from me on AGI—not when I think it will happen, but when I think it won’t happen. I’m certainly not an AGI skeptic, it’s possible and will happen.
More importantly, I predicted that no matter what form AGI takes, it will be vulnerable to attack and manipulation. I mentioned that this would especially be true if it were built on top of LLMs (remember, toothpicks and bubblegum.) Maybe something about generalizing across many tasks in the real world makes things vulnerable. This is something I mentioned back in February of 2023.
To make matters worse, we may be stuck with the vulnerabilities that get identified because there is no fix. Think of examples like adversarial policy attacks. We’ve all heard of AlphaGo beating Lee Sedol at Go. However, most don’t know that even average Go players can beat superhuman Go AIs using adversarial policy attacks. Yes, the stakes are low in the game of Go. However, this is a cautionary tale.
We may be stuck with the vulnerabilities that get identified because there is no fix.
Combine these potential issues with the fact that humans don’t do a good job of finding vulnerabilities in a system before it is launched into production, and we have a recipe for lingering problems. When these lingering problems are in high-risk systems, disasters are only a couple of steps away, and there’s not much we can do about it.
One of the oft-repeated talking points erupting from the mouths of futurists and tech leaders alike is claiming that things will cost nothing in the future. As if we are to believe all of these people are in the business of making something for nothing. The entire claim is a gross absurdity that charlatans like Ray Kurzweil conjured out of thin air, and others parrot at every opportunity. This claim is made with such confidence that it is rendered self-evident, and to question it means you are an out-of-touch dolt lacking the religious fervor necessary to create the techno-utopia.
