Hacker News Reader

That's unfortunate. When an LLM makes a mistake it's very helpful to read the CoT and see what went wrong (input error/instruction error/random shit)

View on HN

Yeah, exposed chain of thought is more useful as a user, as well as being useful for training purposes.

I think we may discover that model do some cryptic mess inside instead of some clean reasoning.

Loopback to: "my code works. why does my code work?"

I’d say depends. If the model iterates 100x I’d just say give me the output.

Same with problem solving in my brain: Sure, sometimes it helps to think out loud. But taking a break and let my unconcious do the work is helpful as well. For complex problems that’s actually nice.

I think eventually we don’t care as long as it works or we can easily debug it.

Reply on news.ycombinator.com