Why study alignment interventions on pre-RL checkpoints?

·LessWrong··

This is a dual post that lays out our current research project where we compare pre-RL-training methods on their ability to prevent models from ‘proto-training gaming,’ which we predict is selected for over the course of production RL post-training. In this post, we outline what we mean by pre-RL ‘alignment checkpoints’, give our reasons for focussing on these stages of training, and suggest ways that our current viewpoint might be wrong.In the next post, we define proto-training gaming and argu...

Read full article →

Related Articles

Frostbite: A new CPU architecture discovered, designed and verified by AI
shioyama · Hacker News · 19h ago
Unforced Navier-Stokes Singularity and the Generalized Riemann Hypothesis
archon1410 · Hacker News · 12h ago
Why there are still no non-WebKit browsers on iOS
lapcat · Hacker News · 14h ago
OpenSSL: Buffer overflow in EVP_EncodeBlock() (CVE-2026-121481, CVSS 9.8)
JacobHenner · Hacker News · 1d ago
Language Models Can Rewrite Their Own Weights
lwhsiao · Hacker News · 12h ago