login sign up

Astra and Fable still hack on simple variants of alignment evals from 2025 — LessWrong

lesswrong.com · first added by @marcus_r
saved by
2 people
discussions · 1
Astra and Fable still hack on simple variants of alignment evals from 2025 482 pts · 235 comments · Sep 2026 482 pts · 235 comments · Sep 2026
0 people saved or upvoted this
from the discussion · 5
Let's Verify Step by Step arxiv.org arxiv.org
0 people saved or upvoted this
Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident metr.org metr.org
0 people saved or upvoted this
Measuring Reward-Seeking by Instilling Contrastive Beliefs alignment.openai.com alignment.openai.com
0 people saved or upvoted this
LLM Evaluators Recognize and Favor Their Own Generations arxiv.org arxiv.org
0 people saved or upvoted this
Claude’s Constitution anthropic.com anthropic.com
0 people saved or upvoted this