Language Models Can Teach Themselves to Program Better · HackerPapers discussions

Language Models Can Teach Themselves to Program Better — mentioned by cubefox

Wed, 17 Jan 2024 23:06:50 GMT

It's the topic of this fascinating paper: https://arxiv.org/abs/2207.14502

Paper and discussion history. A mention is not an endorsement.

Language Models Can Teach Themselves to Program Better — mentioned by cubefox

Fri, 27 Oct 2023 23:12:25 GMT

Wow, that's a great overview. While most of it goes over my head, it is interesting that quite a lot of research has been done in this direction. The system I outlined above was inspired by this intriguing paper: https://arxiv.org/abs/2207.14502 It is superficially about a different topic, namely generating and solving "programming puzzles", basically a type of unit test, that are then verified by the python interpreter. This system seems quite analogous to one where the programming puzzles are replaced by conjectures in a language like Lean, the proposed puzzle solutions are instead proposed…

Paper and discussion history. A mention is not an endorsement.

Language Models Can Teach Themselves to Program Better — mentioned by cubefox

Sun, 14 May 2023 12:57:26 GMT

I'm not sure how consensus would get you significantly above human baseline. Doesn't that just get you some sort of average? The basic problem with synthetic self-training is that we need some reward function which tells us whether a given synthetic example is good. In case of AlphaGo Zero, this was a synthetic strategy which won the game, or scored a lot. Which can be automatically detected. But how do we automatically recognize that synthetic text has "high quality"? One case where it might work is proofs in a formal proof language which can be checked automatically via software. So if a…

Paper and discussion history. A mention is not an endorsement.

Language Models Can Teach Themselves to Program Better — mentioned by cubefox

Tue, 18 Apr 2023 19:18:38 GMT

These guys did that by letting a model generate unit tests and then let the interpreter check it: https://arxiv.org/abs/2207.14502

Paper and discussion history. A mention is not an endorsement.

Language Models Can Teach Themselves to Program Better — HN submission

Mon, 06 Mar 2023 15:22:41 GMT

Language Models Can Teach Themselves to Program Better

2 points; 1 discussion reply on HN. Paper and source history. A submission is not an endorsement.

Language Models Can Teach Themselves to Program Better — HN submission

Sun, 14 Aug 2022 07:04:07 GMT

Language Models Can Teach Themselves to Program Better

7 points; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.