Louwrentius

If you are enthusiastic about AI (LLMs), you may be fooling yourself. Do not consider this an insult, but a kind-hearted warning.

Remember what Richard Feynman said:

The first principle is that you must not fool yourself
and you are the easiest person to fool.

source (backup)

He made this statement as part of the Caltech's 1974 commencement address. In this speech he strongly advocated for proper scientific integrity.

I feel it can be difficult to maintain proper 'scientific integrity' on a personal level as AI is a fast-moving target. The tools feel unreasonably capable, using an LLM makes you feel more productive when writing code for instance.

Yet as of today, there is no independent scientific evidence¹ (that I'm aware of) that people or organizations are indeed more productive² when using AI.

Are you really more 'productive'?

Meanwhile, a study (late 2025) seems to report that although participating developers felt they completed tasks faster using AI, they where around 19% slower.

Using AI may feel faster and more productive, but you might be fooling yourself. You can never go back in time and measure yourself doing the task without using AI tools and compare the results.

There is a reason why (scientific) experiments are often blind or double-blind. We want prevent personal bias from influencing the outcomes skewing the results and making us believe things that aren't true. Not so easy to do on your own.

This is why I'm not really swayed by personal testimonies and anecdotes about how AI makes people more productive. How they vibe coded some app they would otherwise not have built³.

Do not be fooled by the mass layoffs 'due to AI' proving that AI is making people more productive. Sam Altman himself called many of these layoffs 'AI washing'. These are regular layoffs after a period of over-hiring and now the market corrects itself. Because aside from AI-related businesses, it seems that the economy isn't doing great.

Furthermore, some of these layoffs seem to be specifically to free up money to buy AI resources. Management by fear of being left behind. Firing people to get money to pay for unproven technology that may never be capable to replace the people being fired.

AI is unsustainable

Even if you still think AI is really making you more 'productive', enjoy it while it lasts.

According to this article ChatGPT starts losing money when you use your $200 subscription abouve 11%. If you'd use the full 100%, that same $200 subscription would allow you to use $14,000 worth in API pricing. That seems crazy to me.

Whatever you think your productivity might be due to AI, I don't think it's worth the 'real' cost once the vendors stop heavily subsidizing AI usage. I really wonder how much usage AI would see if it would be just priced at cost. Would you feel as productive if the price would not be $200 but $2000 a month⁴? A $200 subscription is not that big of a gamble, but any higher and it may feel less and less worth it.

Disclosure: my own bias against AI

Now it's time to be really open about my own bias. I'm an AI sceptic. I'm quite influenced by Ed Zitron's reporting. And I'm also standing by my own idea that AI will run out of money and just be turned off due to the huge operational cost.

It might feel unfathomable, unreasonable that this can happen. Maybe Google, AWS or Microsoft scoops up OpenAI or Antropic when they reach the end of the runway. But the economics won't change. Are they going to subsidize AI indefinitely?, to what end?

I'm open to suggestions how this is going to play out. Close to a trillion dollars invested in AI. Where will the money come from to recuperate all these investments? If we disregard the AI (adjacent) companies, the USA economy seems to look bleak (August 2025).

I myself can't just close my eyes for all the externalities of AI. The energy usage in the context of climate change. The data centers disrupting poor communities. The large-scale theft of intellectual or copyrighted property. The enshitification and the enslopification of the internet. The mass hoarding of memory causing inflation.

I'm not convinced the bad of AI is - or can be - compensated by the perceived good AI does. It's not possible to separate the tool itself from it's externalities. This is why I'm not using AI. I'm not using AI on moral, ethical grounds.

Hacker News discussion: here

Post Script

I found the comments on hacker news quite revealing. Most are very dismissive without engaging at all with the actual point of this article.

It is true that the cited Metr study was followed up with a more recent study from 2026 that showed more productivity when using AI. The authors are also pointing out they paid way less and they had a problem where developers didn't want to code without AI.

What was actually interesting about the initial study is the discrepancy between what developers thought and what they measured in terms of productivity. They were measurably fooling themselves.

Furthermore, it seems that people are quick to respond and not actually read and comprehend what I wrote. There are two key points I'm making:

1. Your perception of productivity may be off / you may be fooling yourself
2. In addition, I question and you may fool yourself thinking that the AI tool is worth the unsubsidized cost in relation to the perceived productivity gain.

At least I will try to be more explicit in my writing to get this point across.

I must say I'm amazed at how fast people dismiss this blog post and make claims about AI productivity, without anyone providing anything else but personal anekdotes.

If you zoom out enough, everything is a black box where we don't know about the inner workings, but we can learn about the black box by putting stuff in and observing the output.

Let's pretend that a large language model or LLM is such a black box. If we observe a LLM, we learn - amongst other things - that the output is 'correct' 99%¹ of the time.

This is a very fundamental and important observation. Computers are known for their correctness, for their reliability. We know that if we run a function with input A, we get output X, no matter what. Yes, packets can be lost, memory can go corrupt but those things go wrong in a predictable way. We check and retransmit packets, we use ECC memory. And data is formatted in such a way that we can detect lost packets or corrupted memory in the first place. Our entire world relies on it.

Imagine that a system can't be trusted. That 1+1 isn't always 2. Only 99% of the time. How valuable can such a system be? That probably depends on the circumstances, but we know one thing for sure: we can't trust the output, it must be checked. It doesn't matter how, but it must be checked for correctness and that requires a person.

We see it with self-driving cars. It's really impressive what is possible. But they aren't true self-driving². A person needs to be sitting at the wheel, keeping their attention focussed on traffic as if they would be driving themselves, so they can intervene when the AI makes an inevitable mistake.

Although this post isn't about self-driving cars, we know that humans have a tenancy to get distracted and bored if we aren't actively engaged. We can argue that either we people keep driving ourselves, or we need 100% reliability, so we can remove the steering wheel and read a book while the car drives itself. Only 100% is good-enough, 99% doesn't cut it. What did we actually solve if I'm still required to 'drive' the car because of the 1% chance things go wrong?

In the case of an LLM, how much time is saved by letting a person check the output of an LLM? Can that saved time justify the actual operating cost of the LLM, as opposed to the subsidized cost AI vendors are currently charging?

In case of writing code, an LLM can probably create more functionality and features in a week a team of 100 engineers can validate in a year. So no matter what how fast the LLMs really are, the people are the bottleneck we can't circumvent.

That is, if we care about correctness, about quality, about stability and so on. But if we don't, why do something in the first place, regardless of an LLM being involved or not?

So this is why I think it's impossible for the AI hype to come through on the sky high promises their absurd valuations suggest. I'm not saying that AI may not be valuable³ but probably orders of magnitude less valuable than people want us to believe.

I admit, this isn't an original idea, people have voiced this idea in different, maybe more succinct ways. Yet I think it is worth repeating.

Update - but humans aren't accurate either!

Why replace people with something objectively worse and more unpredictable? Humans check human output. Now humans check LLM output. Throughput is the same, but we fired a person and the remaining person is miserable due to the nature of the work.

P.S.

I also don't understand why organizations would make LLMs a very integral part of their processes, only to discover that models are changed and tweaked, resulting in wildly unpredictable and different output.

Sometimes, when the light hits an LLM just right and you squint your eyes, it takes the shape of a crypto currency. At least that's what I see.