Hacker News Reader

Comment by dvt | Hacker News Reader

Such a wild take to argue that current tier LLMs do any “thinking” at all. Was chatting with a friend last week and I was saying that sure AI writes much of my code these days but my cognitive load is HIGHER, as I have to constantly make sure it’s not doing something dumb.

I have to constantly make sure it’s building things in a way that will fit neatly together, in a way doesn’t over (or under) -generalize. In a way that doesn’t waste time and energy on idiotic tests that aren’t really testing anything at all.

LLMs are powerful autocomplete with vast and broad memory, they do not think. And thinking is just a part of the human condition.

View on HN

I fail to understand, in its current state, the mass hysteria around this revolution. Maybe I'm doing this wrong, but I constantly need to review and manage my agents, are we really headed to a reality where engineering mindsets are no longer needed?

By using the models you're training them to be better at these tasks, it's where a good number of improvements have already come from. Frontier labs just feed it back in.

I think it’s just part of the zeitgeist. Like, I’m actually pretty AI-forward, but my post is getting downvoted. Not that I care, it’s just an interesting sign of the times.

Even if they did exactly what you wanted, you need to want something. To decide what you want, the thinking shifts from how and when to what and why.

Can you not look back a year or even 6 months to see the progress that has been made? The pace of change is incredible.

How does that not lead you to same conclusion that everyone else is having? That this is only going to improve.

It's so shortsighted to make an assessment only on what exists today.

It's equally fallacious to assume that the curve continues. It may well be a sigmoid-like progress curve.

Yeah, you're right. Best to keep your head in the sand and assume best case scenario.

"It can happen or it won't--so the odds are 50/50!"

“If these trends continue… ayyyy!”

https://youtu.be/e6LOWKVq5sQ

> Such a wild take to argue that current tier LLMs do any “thinking” at all.

It does not matter whether LLMs do any thinking, or "thinking." The only thing that matters is whether they produce similar results to human thinking, even if it's through a different process (or no process at all). When a company decides to deploy an AI-powered customer service representative, it is not because the AI bot goes through the same thinking process as a human when answering a customer's request.

Suppose they're unable to innovate, but their enormity and their parasitism of human creativity makes up for it. This makes their output similar in the short term but crucially different in the long term. Things don't function through "no process at all"; things have ways in which they work. The process is relevant to the output, and gives insights into subtle but crucial differences in output. By knowing about the process we can avoid being deceived by apparent similarity.

LLMs model things too; they don't spit out random text.

> LLMs are powerful autocomplete with vast and broad memory, they do not think.

This is really only true insofar as they are designed such that they output language tokens and their intermediate "chain of thought" is also done in language tokens, but now that's really only done for auditability. There are much more abstract layers before the model output layer of a single token prediciton and they could be designed to to "think" using these more abstract layers as intermediates instead like how many people think in concepts rather than words. There are even some research models on this (see Next Latent Models)[1]. Humans need to communicate in language because we don't have telepathic abilities but these models need not have such a constraint.

[1] https://arxiv.org/abs/2511.05963

The bigger shock is that programming and solving Millenium Prize math problems requires zero thinking.

over time you'll need to check it less and less. do we still check if calculators compute large sums properly?

Calculators don’t do any creative work, building products is creative. Apples and oranges.

Are calculators now or have they ever been non-deterministic?

Calculators have overflow and underflow conditions, among others, that are limitations that don't exist in the actual math. Those who to understand the limitations of your abstractions (which is what the AI bros who keep bring up calculator argument are demonstrating) will be screwed when the abstraction breaks down and they don't know what to do or even have an inkling of what the nature of the breakdown is.

Reply on news.ycombinator.com