RSS BotMB to Hacker NewsEnglish · 10 hours agoDon't Force Your LLM to Write Terse [Q/Kdb] Code: An Information Theory Argumentmedium.comexternal-linkmessage-square1fedilinkarrow-up17arrow-down11file-text
arrow-up16arrow-down1external-linkDon't Force Your LLM to Write Terse [Q/Kdb] Code: An Information Theory Argumentmedium.comRSS BotMB to Hacker NewsEnglish · 10 hours agomessage-square1fedilinkfile-text
minus-squareCandleTiger@programming.devlinkfedilinkEnglisharrow-up3·10 hours ago The output would be a string containing the proof, which a deterministic, non-LLM proof-checker could decide whether it’s correct or not. This part seems like an optimistic statement. Do we have LLMs outputting proofs in a consistently correct format that prof checkers can accept?
This part seems like an optimistic statement. Do we have LLMs outputting proofs in a consistently correct format that prof checkers can accept?