Hacker News Reader

Comment by jacquesm | Hacker News Reader

Those bugs (assuming validity) were there in spite of the careful curation. People tend to do what they can to the best of their abilities. Before compilers did much in terms of static checking a lot more errors made it into production than today. Better tools -> better code.

Whether AI written code is going to be better in the longer term remains to be seen. With what I've looked at so far I'm not super optimistic about this.

View on HN

LLMs reliably avoid whole categories of bugs, some of them the most serious offenders in terms of impact.

To me it has become clear that LLMs will, on average, improve the code quality by a lot.

It was just 5 minutes ago that I prevented a Claude PR from introducing an incredibly obvious RCE bug into a codebase I maintain. The LLMs are no better than the people driving them. But those people spew out more code than ever before.

It's not that they are "no better", it is that they have been specifically instructed through RL to follow the user's direction for better or worse.

Interesting, I think I don't encounter many bugs, but I also discuss code with AI and don't mentally track the correction I give it.

Which model did you use? I hold Opus 5.5 in quite high regard, finding it very reliable in terms of cider quality.

Better than what? 99% of stuff produced by humans is a garbage. Arguing about "abilities" or compilers does not change that.

Reply on news.ycombinator.com