SFT matches RL if you MCMC the training data first | Hacker News - But Better

SFT matches RL if you MCMC the training data first arxiv.org

Introducing new capabilities to frontier models has long been the goal of posttraining, which predominantly employs supervised finetuning (SFT) and reinforcement learning (RL) to this end. Conventional wisdom dictates…

1 points•mrkn1•about 9 hours ago•0 comments•

#sft #matches #data #mcmc #training #data first #training data #sft matches

0 comments

No comments yet.

Related stories

The People of Utah vs. Kevin O’Learytheverge.com
The Verge · 0 points · 1 day ago
Inside our months-long investigation into Kevin O’Leary’s Utah data center debacletheverge.com
The Verge · 0 points · 1 day ago
Show HN: Beyond "chat with your database" – agentic data explorationreportr.com
Hacker News · 1 points · 2 days ago
Show HN: Datastory, Data storytelling with AI and an open knowledge graphdatastory.tech
Hacker News · 6 points · 2 days ago
AI's next training data: your dodgy gaming skillswired.com
Hacker News · 2 points · 4 days ago
Show HN: ChaosData – Chaotic data that fits your schemachaosdata.net
Hacker News · 1 points · 1 day ago