User Model Extraction via Belief Self-Distillation | Hacker News - But Better

User Model Extraction via Belief Self-Distillation arxiv.org

Large language models (LLMs) implicitly infer attributes of their users and adapt their behavior accordingly, yet these beliefs remain difficult to inspect and causally manipulate. We introduce Belief Self-Distillation…

1 points•sbulaev•2 days ago•0 comments•

#self #belief #distillation #self distillation #belief self #belief self distillation #user #extraction

0 comments

No comments yet.

Related stories

Recursive Self-Improvement via On-Policy Distillation for Reasoningarxiv.org
Hacker News · 2 points · 2 days ago
I built a self-improving voice AI agent that runs offline on a RPi5github.com
Hacker News · 2 points · 1 day ago
Show HN: Air-gapped file encryption as self-decrypting HTML pagecms-sfx-demo.apeleg.com
Hacker News · 90 points · 7 days ago
AURA v0.1.0: Deterministic Trigger Extraction, Auditable Math & a Self-Healing Analytics Enginedev.to
DEV Community · 3 points · 4 days ago
Show HN: LogPhase – See a PR's blast radius as impacted user journeyslogphase.ai
Hacker News · 1 points · 2 days ago
Show HN: Fusion-runtime – self-hosted voice agents, STT+LLM+TTS in one processgithub.com
Hacker News · 2 points · 9 days ago