Hacker News Reader

Comment by _pdp_ | Hacker News Reader

The latest frontier models no longer produce slop. The code is pretty good in general - better in fact in many situations compared what a human might have produced with limited time, resources and experience.

The argument against AI-assisted programming just doesn't hold anymore.

View on HN

If you read the post, the argument isn’t that AI-made projects are bad, just that they’re uninteresting. It’s similar to the argument I’ve seen people make of “don’t open a AI-written PR, just post your prompt and let me use my own LLM”

All interesting projects that I have seen in recent times are all done with AI.

That’s not my experience. The frontier models produce ok code that works but it’s still very sloppy. Further evidence is that the number of massive PRs is increasing not decreasing, i.e. the LLMs are brute forcing changes rather than introducing quality code patterns.

This is not a good metric IMHO. I would expect new projects to have bigger changes initially as they try to find a suitable architecture. That is not slop. This is no different than rewriting a project 2-3 times pre-AI to find out what sticks and how to organise it better.

I have yet to see the churn ever go down. I'm not saying it won't but what I see is that current frontier LLMs are still bad at identifying functional abstractions or patterns that can actually scale.

I have not seen an LLM rewrite 2-3 times and then settle on what sticks, it just rewrites constantly and usually while leaving residue of previous attempts behind.

Which in some projects is still fine. Like I said, they are generally producing working code, but I haven't seen proof yet that it can actually scale and evolve well over time.

I've been hearing that for years. It's no more true today than it was then.

3 years ago the best model was GPT-3.5 Turbo with 16,385 tokens context window.

Reply on news.ycombinator.com