AnalysisAI ModelsNovember 4, 2026

Quanta examines tests of whether AI models can introspect

Read original source →quantamagazine.org

Anthropic researcher Jack Lindsey injected "thoughts" directly into model neural networks in 2025, and models could sometimes detect and describe the injections. A year of replication attempts has produced simpler explanations for some results, while others remain contested.

People · Jack Lindsey

1 source

More stories today

Open the live feed