A Three-Layer Model of LLM Psychology — LessWrong

This post offers an accessible model of psychology of character-trained LLMs like Claude. …

x A Three-Layer Model of LLM Psychology — LessWrong Best of LessWrong 2024 LLM Personas AI Psychology Aligned AI Role-Model Fiction Alignment Pretraining Cyborgism Language model cognitive architecture AI Curated 266 A Three-Layer Model of LLM Psychology by Jan_Kulveit 26th Dec 2024 AI Alignment Forum 10 min read 17 266 Ω 84 This post offers an accessible model of psychology of character-trained LLMs like Claude. Epistemic Status This is primarily a phenomenological model based on extensive interactions with LLMs, particularly Claude. It's intentionally anthropomorphic in cases where I believe

saved by

• Lydia Nottingham
• Vincent Cheng
• Yudhister Joel Kumar
• Timothy Kostolansky

related reading

• Role-playing vs Self-modelling — LessWronglesswrong.com
• The Persona Selection Model: Why AI Assistants might Behave like Humansalignment.anthropic.com
• On the Biology of a Large Language Modeltransformer-circuits.pub
• Simulators — LessWronglesswrong.com
• Did Claude 3 Opus align itself via gradient hacking? — LessWronglesswrong.com
• AI Induced Psychosis: A shallow investigation — LessWronglesswrong.com
• A Mechanistic Explanation of Prompt Injection (and why you should study roles) — LessWronglesswrong.com
• Tracing the thoughts of a large language model \ Anthropicanthropic.com
• Simulated Users & Sad LLMs1a3orn.com
• llm assistant personas seem increasingly incoherent (some subjective observations) — LessWronglesswrong.com
• Thousand-dimensional structure — Resolutionresolution.org
• The persona selection model \ Anthropicanthropic.com