i hate AI generally but i've been quite impressed with its application to the purpose of generating meaningful commit messages based on changes in the code

they always fluff it out with "improves dah de dah for dah de dah" bullshit, mostly always gets it wrong about the why of it, but always gets it right what is being done and about 1/4 of the time it does get the reason correct:

Renamed receiver in the UnmarshalJSON method for various envelope structures from pkg/nostr package. The intention is ensuring coherence with corresponding elements and making the interface implementation more intuitive. This step is essential for appropriate handling of auth challenges and profile retrieval, kickstarting the transition process with the new envelope structure interface.

"kickstarting the transition process with the new envelope structure interface."

this nailed the purpose very well

the rest of the message is bunk

AIs can't detect intent very well, and when i see it actually succeeding at detecting intent it makes me happy and makes me think i'll keep paying to have huge amounts of watts of energy dedicated to scanning neural network graphs to generate this shit for me for this purpose.

Reply to this note

Please Login to reply.

Discussion

Is it getting that intent from comments in the code documenting the intent? If not, it is hard to imagine how it could guess the intent. High-level intent or goals or ends are usually the missing comments in code and/or commit messages.