llm assistant personas seem increasingly incoherent (some subjective observations) — LessWrong

(This was originally going to be a "quick take" but then it got a bit long. Just FYI.) …

x llm assistant personas seem increasingly incoherent (some subjective observations) — LessWrong AI Psychology AI Curated 2026 Top Fifty: 20 % 343 llm assistant personas seem increasingly incoherent (some subjective observations) by nostalgebraist 29th Apr 2026 11 min read 84 343 (This was originally going to be a "quick take" but then it got a bit long. Just FYI.) There's this weird trend I perceive with the personas of LLM assistants over time. It feels like they're getting less "coherent" in a certain sense, even as the models get more capable. When I read samples from older chat-tuned mode

saved by

• Abhay Sheshadri
• Kushal Thaman
• Jo J.
• Neil Rathi
• Will Anderson
• Arya Pasumarthi
• Dillon Nguyen

related reading

• The Persona Selection Model: Why AI Assistants might Behave like Humansalignment.anthropic.com
• Guardian Angels: LLM Personalization for Productivity and Security · Gwern.netgwern.net
• trees are harlequins, words are harlequins - the voidnostalgebraist.tumblr.com
• The persona selection model — LessWronglesswrong.com
• Discovering Language Model Behaviors with Model-Written Evaluations — LessWronglesswrong.com
• Role-playing vs Self-modelling — LessWronglesswrong.com
• The Waluigi Effect (mega-post) — LessWronglesswrong.com
• the void — LessWronglesswrong.com
• The persona selection model \ Anthropicanthropic.com
• Should We Train Against (CoT) Monitors? — LessWronglesswrong.com
• The assistant axis \ Anthropicanthropic.com
• A Case for Model Persona Research — LessWronglesswrong.com