pull down to refresh

> be me
> visit X
> see untrustworthy attention-seeking person bragging about
> how much they trust AI to write privacy and money code
> leave X

It's worth every dev using the frontiest frontier model to write 1k lines of code then review it manually, in addition to having an AI assist, noting how grievous some of the mistakes are. This might help them understand how trustworthy their 100k line FOG (fresh off GPU) rewrite is.

100k lines says it all. I have a little low priority queue for Claude whenever I have leftover credz to slowly build me a mostro dashboard. Its already 20k sloc with 138 deps and its not even functional yet.

Luckily no one but me will ever run it. And I'm not even sure that I will run it.

reply
82 sats \ 3 replies \ @anon 22 Aug

boomers 🙄

reply

lol I think there's an argument to make in favor of trusting AI unconditionally but it requires arguing that more bugs (particularly ones that lose customers money and privacy in this case) don't matter. I don't see anyone making that argument though. It's maybe too confessional for hypers, but I think it'd get their point across and honestly.

reply
10 sats \ 1 reply \ @anon 22 Aug

If you’re using agents continuously, your code is likely getting more and better security review than a public GitHub project with less than 1k stars

reply

There’s a difference between using agents continuously and trusting them. I believe in the former not the latter.

reply