[ghoshbishakh]

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]

Maybe 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.

[mattkenefick]

they probably weight documentation higher

[xnx]

They're applying trust factors similar to PageRank.

[TeMPOraL]

Possibly, but at a semantic level.

[tgv]

I 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]

Also, you can count on most developers to reach the documentation only as a last resort, and try every random blog first.

[PunchyHamster]

It's probably trained more on fixes on incorrect CORS than the problems

[mhh__]

They look for latent structure in the data.