Discovering Language Model Behaviors with Model-Written Evaluations — LessWrong

“Discovering Language Model Behaviors with Model-Written Evaluations” is a new Anthropic paper by Ethan Perez et al. that I (Evan Hubinger) also coll…

x Discovering Language Model Behaviors with Model-Written Evaluations — LessWrong Language Models (LLMs) AI Frontpage 100 Discovering Language Model Behaviors with Model-Written Evaluations by evhub , Ethan Perez 20th Dec 2022 AI Alignment Forum 1 min read 34 100 Ω 43 This is a linkpost for https://www.anthropic.com/model-written-evals.pdf “ Discovering Language Model Behaviors with Model-Written Evaluations ” is a new Anthropic paper by Ethan Perez et al. that I (Evan Hubinger) also collaborated on. I think the results in this paper are quite interesting in terms of what they demonstrate abou

saved by

• Lydia Nottingham
• Will Anderson
• Vincent Cimino

related reading

• Mechanistically Eliciting Latent Behaviors in Language Models — AI Alignment Forumalignmentforum.org
• How confessions can keep language models honest | OpenAIopenai.com
• The Persona Selection Model: Why AI Assistants might Behave like Humansalignment.anthropic.com
• [2501.11120] Tell me about yourself: LLMs are aware of their learned behaviorsarxiv.org
• llm assistant personas seem increasingly incoherent (some subjective observations) — LessWronglesswrong.com
• Introspection Adapters: Training LLMs to Report Their Learned Behaviorsarxiv.org
• Foundation Models for Oversight | Transluce AItransluce.org
• Self-CTRL: Self-Consistency Training with Reinforcement Learningarxiv.org
• [2212.09251] Discovering Language Model Behaviors with Model-Written Evaluationsarxiv.org
• Language Models Learn to Mislead Humans via RLHFarxiv.org
• Towards Understanding Sycophancy in Language Models — LessWronglesswrong.com
• Surfacing Pathological Behaviors in Language Models | Transluce AItransluce.org