Shauli Ravfogel
Center for Data Science, New York University and Bar-Ilan University
Critic Current evidence and current methods do not support the claims being made
Shauli Ravfogel is a natural language processing and mechanistic interpretability researcher at New York University and Bar-Ilan University, known for developing iterative nullspace projection (INLP), linear concept erasure, and causal intervention methods for auditing what neural representations actually control.
In Can LLMs Introspect? A Reality Check (arXiv:2605.26242, May 2026), co-authored with Shashwat Singh and Tal Linzen, Ravfogel applied causal representation auditing to claims of emergent introspection in large language models. The authors showed that in both intervention-detection and hidden-state-prediction paradigms, models frequently rely on distributional anomalies at the input surface rather than privileged readout of their own internal activations, and that external classifiers reading only the input match the model’s self-predictions until relabeled controls reduce performance toward chance.
Ravfogel is classified as a Critic on this site. His causal-erasure and control-condition methods separate genuine internal state monitoring from surface statistical confounds in AI interpretability research.
Known for. Can LLMs Introspect? A Reality Check, linear concept erasure, causal probing of neural representations
Coverage on this site
- LLM Introspection After the November 2026 Quanta Report, Concept Injection Against the Reality Check November 2026
Related researchers
- Jack LindseyAnthropic, Transformer CircuitsIntrospection circuits in language models, mind viruses in multi-agent systems, Global Workspace representations
- Shashwat SinghCenter for Data Science, New York UniversityCan LLMs Introspect? A Reality Check, concept erasure and causal representation auditing in language models
- Tal LinzenDepartment of Linguistics and Center for Data Science, New York UniversityCan LLMs Introspect? A Reality Check, mechanistic evaluation of syntactic and metacognitive claims in language models
- Scott AaronsonUniversity of Texas at Austin, Department of Computer ScienceQuantum complexity theory, mathematical counterexamples to Integrated Information Theory (Phi in expander graphs and Reed-Solomon grids), speaker at UT Austin Consciousness and Intelligence 2026
- Michel BitbolCNRSPhenomenology and quantum physics, the critique of objectivist reduction, first-person knowledge as irreducible, collaboration with Claire Petitmengin
- Ned BlockDepartment of Philosophy, New York UniversityDistinction between Phenomenal and Access Consciousness (1995), Blockhead argument against behavioral tests, The Border Between Seeing and Thinking (2023)
Shauli Ravfogel is listed with the other critics and skeptics, who hold that current evidence and current methods do not support the claims being made. The positions themselves are set out in the index of consciousness theories. Every researcher covered on this site is indexed in the directory of consciousness researchers.