Mechanistically Eliciting Latent Behaviors in Language Models — LessWrong

Produced as part of the MATS Winter 2024 program, under the mentorship of Alex Turner (TurnTrout). …

x Mechanistically Eliciting Latent Behaviors in Language Models — LessWrong Activation Engineering AI Evaluations Language Models (LLMs) MATS Program AI World Modeling Frontpage 226 Mechanistically Eliciting Latent Behaviors in Language Models by Andrew Mack , TurnTrout 30th Apr 2024 AI Alignment Forum 54 min read 44 226 Ω 99 Produced as part of the MATS Winter 2024 program, under the mentorship of Alex Turner (TurnTrout). TL,DR: I introduce a method for eliciting latent behaviors in language models by learning unsupervised perturbations of an early layer of an LLM. These perturbations are tra

related reading

• Mechanistically Eliciting Latent Behaviors in Language Models — AI Alignment Forumalignmentforum.org
• Surfacing Pathological Behaviors in Language Models | Transluce AItransluce.org
• Natural Language Autoencoders Produce Unsupervised Explanations of LLM Activationstransformer-circuits.pub
• A Mechanistic Explanation of Prompt Injection (and why you should study roles) — LessWronglesswrong.com
• Dreaming Vectors: Gradient-descented steering vectors from Activation Oracles and using them to Red-Team AOs — LessWronglesswrong.com
• I found >800 orthogonal “write code” steering vectors | Jacob’s Blogjacobgw.com
• Auditing language models for hidden objectives — LessWronglesswrong.com
• Refusal in LLMs is mediated by a single direction — LessWronglesswrong.com
• [2604.16812] Introspection Adapters: Training LLMs to Report Their Learned Behaviorsarxiv.org
• Discovering Language Model Behaviors with Model-Written Evaluations — LessWronglesswrong.com
• [2512.11949] Neural Chameleons: Language Models Can Learn to Hide Their Thoughts from Unseen Activation Monitorsarxiv.org
• Implementing activation steering — LessWronglesswrong.com