Hacker News new | ask | show | jobs
by Falimonda 753 days ago
It seems like they're caching information/replies on key words which is a good optimization.
1 comments
iknownthing 753 days ago
I assumed it was coming straight from the API because of the token-by-token generation effect but maybe you're right.
link
rodiger 753 days ago
The token-by-token responses are probably API, while the "instant" loads seem to be cached.

Probably using groq based on speed of response

link