Content
@
https://warpcast.com/~/channel/entropybender
0 reply
0 recast
0 reaction
kevin j
@entropybender
finetuning 4o is so easy to do now. ~500 tweets with 80% training, 20% test worked incredibly well. would assume it's probably not even bad with just 200 tweets, especially if you're just trying to emulate tone/humor/personality
2 replies
0 recast
4 reactions
David Hurley
@hurls
sounds like fun. how are you finetuning 4o?
1 reply
0 recast
0 reaction
kevin j
@entropybender
using a dataset of tweets, where one user asks a question to another. question-answer as a prompt-response pair, then load the whole thing to openpipe
1 reply
0 recast
1 reaction
David Hurley
@hurls
very cool, will try it out
0 reply
0 recast
0 reaction
