Hacker News new | ask | show | jobs
by a123b456c 683 days ago
Yes. A stronger test would be guessing the results of as-yet-unpublished experiments.
1 comments
jslezak 682 days ago
They did this. Read the paper
link
abcgde 680 days ago
Well, they looked at papers that weren't published as of the original model release. But GPT very likely had unannounced model updates. Is it not possible that many of the post 2021 papers were in the version of GPT they actually worked with?
link