Do LLMs Introspect? Injected-Thought Tests Suggest Claude Detects Anomalies, Not Self-Knowledge
dl_weekly · x · 2026-11-20
Quanta Magazine reports on new research testing whether large language models can introspect.
- In 2025, Anthropic researcher Jack Lindsey devised an experiment that injects "thoughts" directly into a model's neural network — a kind of targeted digital neurosurgery — to probe introspection.
- In Claude's trials, the model flagged injected thoughts in about 20% of cases.
- Follow-up replication studies suggest this behavior reflects anomaly detection rather than specific self-knowledge of its own thought processes.
- The article notes introspection has been debated for over a century in philosophy and psychology — some tie it to consciousness, others call it an elaborate illusion — and precise definitions remain hard.
The piece maps both the experimental approaches and the current limits of evidence on AI self-awareness.
More from Models
- Why "Model UX" matters: author picks Gemini Neon as the best to actually read — flavioAd · 2026-11-20
- 'Rollout 260,089,137': a surreal AI-generated story of a model cluster losing its mind — voooooogel · 2026-11-20
- Nov 24 marked on calendar as '1 year intelligence explosion anniversary' — scaling01 · 2026-11-20
- Abundant: Claude safety guardrails flagged RSI tasks but zero refusals actually blocked agents — Madisonkanna · 2026-11-20
- NVIDIA open-sources Alpamayo 2 Edge, a 2B-parameter VLA model for real-time in-vehicle reasoning — NVIDIA Developer · 2026-11-20
- HardGeoBench: top AI models now pass@1 above 50% locating metadata-free photos — Afinetheorem · 2026-11-20