Identity as Attractor in LLM Activation Space
💡Geometric proof agent identities form stable attractors in LLMs—vital for persistent agents.
⚡ 30-Second TL;DR
What Changed
Paraphrases of cognitive_core cluster tighter than controls (Cohen's d > 1.88, p < 10^{-27})
Why It Matters
Provides geometric evidence for persistent agent identities in LLMs, potentially enabling more stable cognitive agents across prompts. Distinguishes 'knowing about' from 'embodying' an identity.
What To Do Next
Replicate attractor clustering with TransformerLens on Llama 3.1 for your agent prompts.
Key Points
- •Paraphrases of cognitive_core cluster tighter than controls (Cohen's d > 1.88, p < 10^{-27})
- •Replicated across Llama 3.1 8B Instruct and Gemma 2 9B
- •Semantic effect dominant; structural completeness needed for attractor
- •Reading agent description shifts states closer than sham control
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: ArXiv AI ↗
This is a summary, not the original. Read the source, or get the weekly briefing.
The weekly digest
One email a week. Unsubscribe anytime.