News for hackers. Stuff that matters.

A comment on Don Knuth Plays with ChatGPT. Parent comment

Re:Don Knuth Plays with ChatGPT by drdeca on Saturday May 20, 2023 @09:51PM (#36017288)
Are you taking the RLHF into account when you say so?
[ Parent | on HN ]
Re:Don Knuth Plays with ChatGPT by alex_sf on Saturday May 20, 2023 @11:42PM (#36017824)
Taking RLHF into account: it's not actually generating the most plausible completion, it's generating one that's worse.
[ Parent | on HN ]
Re:Don Knuth Plays with ChatGPT by dotancohen on Sunday May 21, 2023 @12:28AM (#36018033)
Well, I wasn't, but if you look at the top most comment of this thread [0] you'll see that considering the level of human reinforcement being demonstrated only reinforces my point.

[0] https://news.ycombinator.com/item?id=36013017

[ Parent | on HN ]