Re:What we learned from using GPT for 500k+ classifications
by lgas on Tuesday July 11, 2023 @08:32PM
(#36688995)
Not to dogpile on all the other "isn't this just" messages, but isn't this just sparse embeddings?
[ Parent
| on HN ]
Re:What we learned from using GPT for 500k+ classifications
by lmeyerov on Tuesday July 11, 2023 @09:02PM
(#36689221)
Yeah I'm struggling to understand at a fundamental level how this is better both in math + engineering, doubly so by the time you get to sentence embeddings . (Genuinely, it seems to use the same ideas, so curious what the specific trick is vs mature embedding packagings already doing much of this afaict.)
Re:What we learned from using GPT for 500k+ classifications
by mmcwilliams on Tuesday July 11, 2023 @09:22PM
(#36689358)
You're not wrong. This sounds curiously close to the ways I've seen word2vec used in production.