Comment by atleastoptimal | Hacker News Reader

Comment by atleastoptimal | Hacker News Reader

Parent Full thread

> I am going to state a bet: In 5 to 10 years, once all the vibe coders get hooked into LLMs, there's going to be so much unmaintainable, wrongly architectured crap code in every codebase they touch, that by then they are unable to get any work done without consulting their LLM. And if that heap of collected technical debt becomes so large that even the LLM fails

In 5-10 years AI is going to be so much better than even the best human coder that this is a moot point. If anything AI will be used to correct all the crappy human made code that is still being pushed due to the vanity of coders still pretending that they are better than AI at coding.

I can understand hating AI, but it seems like many who are against genAI have a strange delusional disbelief of how good the models are, and the trend-line we are on. They think that their special skills will never be eclipsed by an AI model. If you are going on a crusade against genAI and LLM’s at least be honest about what you’re up against.

>If anything AI will be used to correct all the crappy human made code that is still being pushed due to the vanity of coders still pretending that they are better than AI at coding.

In my organization, this is already happening. We've been using LLMs to boost our test coverage without touching our human code, then use that as a scaffold to let it go through and refactor, clean up, and optimize, and then validating against both our tests and gold standard test datasets.

In our case, it's made a legacy codebase far more readable to our junior engineers, and the performance improvements (from using an autoresearch-style approach) has resulted in a six figure decrease in our compute spend for the production service we trialed this on.

> In 5-10 years AI is going to be so much better than even the best human coder

How, if it learns from humans?

How did alphago become better than any human Go player?

Because it can Monte Carlo his ass out of problems, humans can't

In my opinion by Opus 4.7 Claude was better at coding than most programmers. While there is certainly more slop being produced now overall, I believe the quality of important projects like Linux and proprietary software created by large companies, where advanced AI will be piloted by skilled engineers, will improve in quality, maybe dramatically, over the next decade.

I am still 100% willing to concede that Opus 4.8 or even how good Mythos is supposed to be is not yet at the general reasoning ability of top 5% coders or any smart human with a few years of domain experience. However the rate of improvement is so consistent and unrelenting that it seems silly to assume that it will just stop short of human level. Even if algorithmic, research and data quality improvements suddenly stopped, we still have years of better GPU’s and scaling

Programming requires thinking and LLMs can't think. Therefore LLMs cannot be better than humans at programming.

Your statement seems to be implying (correctly) that LLMs can program, but just not as well as humans. If they're able to program presumably without "thinking" as you seem to be (implicitly) narrowly defining it, then why do you think that limits them to always being sub-par?

It seems like if they can do it, that there's no reason they can't eventually be trained to do it better up to and beyond human performance. It seems strange to suggest that thinking unlocks some nominal margin of "better" specifically that can't be overcome.

All of that aside, even if they can't outperform the top human programmers...what if they get to within a margin where they're still better than most? Isn't a 95th percentile programmer that can run 24/7 and continuously refine its work still going to ultimately come out on top?

I'm more interested in the conclusion that programming doesn't require thinking. And that's where the argument breaks. It seems so obvious, but sometimes the most obvious things are the least true.

>I'm more interested in the conclusion that programming doesn't require thinking.

I suspect it largely has to do with how one defines "thinking". It seems like people like to implicitly define it in such a way as to require a human (or animal), but there are many examples of thinking/intelligence in nature that don't require a brain or even neurons.

I'm genuinely curious: without using the word "think" with all of its ambiguity, can you articulate what it is that we're doing that these models are not capable of? Because it's pretty clear (to me, at least) from the research, particularly a lot of the mechanistic interpretability work coming out of Anthropic, that the models are at least doing something akin to what we think of as thinking, even if it appears foreign to us.

Like, I'm not sure how you could read this and not see some spark of seems like thinking: https://www.anthropic.com/research/tracing-thoughts-language...

What LLM’s can do now, 99.999% of people 3 years ago would say would only be possible with “thinking”. To claim that LLM’s aren’t thinking is sensible if your definition of thinking is inextricably linked to the chemical processes that occur in human brains, but then it ceases to be a useful definition with respect to evaluating a system’s ability to process information, form connections, and reason via abstraction. It is objectively true that LLM’s can do the latter things.

Yeah, I have to admit to finding it somewhat ironic that some individuals accuse the "pro AI" folks of magical thinking, when it seems that escalating levels of magical thinking are being used by the "anti" crowd to suggest that the models can never achieve something akin to human intelligence (particularly in light of the fact that they have on certain dimensions done exactly that).

It's pretty clear that there are significant differences between their intelligence and human intelligence. But that doesn't mean there isn't some sort of intelligence here.

Programming is just telling a computer what to do. LLMs gave turned out to be pretty good at that. What is thinking, anyway?

Well I guess that settles it. Losing to LLMs in CTF challenges must be in my head.

>Programming requires thinking

The issue about AI is that it's gobbling so much information that at some point you couldn't tell the difference. Programming specifically is something that inherently documents itself, meaning while human communication and context and memes and culture is something that evolves and exists many times outside of textual mediums, as soon as any new piece of code is born it is now part of the AI's dataset. And it doesn't help that a vast majority of our code is pretty damn repetitive, especially if you insert code written in the span of two decades and more into the future.

Tldr : The better we get at coding, the more code we write, the better AI gets at coding.

> If anything AI will be used to correct all the crappy human made code that is still being pushed due to the vanity of coders still pretending that they are better than AI at coding.

What funny nonsense. This is like saying AI will replace artists because it's better than your average artist.

Software engineering is as much an artform as it is infrastructure. AI cannot approximate even a poor engineer because it cannot capture the full context of a problem to be solved.

There is an artistic element to coding but claiming it is some sort of unquantifiable abstract mystical thing is silly. There are many metrics, objective and easily measured, that AI can match humans on in code generation.

None of those 'objective' metrics matter as long as the outcomes is incredibly visible and damaging. You say that AI is matching humans, but what people see is apps getting worse with more bugs than ever, an inability to reach anyone that actually gives a shit and second order effects on things like general computing.

It doesn't help that most of these 'metrics' are literally invented by and ran by the companies that benefit the most from said metrics. I've seen this shit at major companies before, because the c-suite loves to invent metrics that make their product look good even if it ignores reality.

The median != the frontier. When looking at all the crappy vibe coded apps and saying they’re worse than the normal expectations of pre-AI applications, you’re comparing some default app-generator model (Gemini flash or worse) with the security integrity of enterprise apps (which still have security flaws). This is very different from comparing Opus 4.8 to the median coder, given the same context, parameters, access to tools, time, and feedback loops.

The frontier has been on the horizon for the past 5 years, where people say it's going to replace all engineers and we'll be in some sort of software heaven. Until when and IF we reach that point, the only thing that matters are the negative outcomes we currently experience as a result.

Reply on news.ycombinator.com