6 ms·
100% - although it is stunning to see since most LLMs get CORS questions right (which is surprising since they trained on all sorts of incorrect data).
by ghoshbishakh 3mo ago
100% - although it is stunning to see since most LLMs get CORS questions right (which is surprising since they trained on all sorts of incorrect data).
- Finbel 3mo agoMaybe it’s like that trick where if a thousand people guess the amount of beans in a jar almost all of them will be wrong but their average will be very close to, if not, correct.
- TeMPOraL 3mo agoPossibly, but at a semantic level.
- mattkenefick 3mo agothey probably weight documentation higher
- xnx 3mo agoThey're applying trust factors similar to PageRank.
- tgv 3mo agoI think many (most?) have preferred sources. I would weight Wikipedia and MDN higher than Snurk Grubble's blog in training, no matter what the topic.
- marcosdumay 3mo agoAlso, you can count on most developers to reach the documentation only as a last resort, and try every random blog first.
- PunchyHamster 3mo agoIt's probably trained more on fixes on incorrect CORS than the problems
- mhh__ 3mo agoThey look for latent structure in the data.