Person

Rohin Shah

Cited in sixteen posts, among them AI Alignment with Changing and Influenceable… and Characterizing Manipulation from AI Systems, since September 2026

Where they come up

Said by someone else — these sentences name them.

“Shah et al. (2022) provide a speculative larger-scale example, conjecturing that InstructGPT’s competent responses to questions its developers didn’t intend it to answer (such as questions about how to commit crimes) resulted from goal misgeneralization (rather than reward misspecification).”
· The Alignment Problem from a Deep Learning Perspective · machine-resolved
“MC would like to thank (in no particular order) Tom Everitt, Cam Allen, Cassidy Laidlaw, Nora Amman, Rohin Shah, Alan Chan, Richard Pettigrew, Marcus Pivato, Orr Paradise, Ann He, Henri Wadsworth, Alex Pan, Erik Jones, Riqui Zhong, Tom Gilbert, Atoosa Kasirzadeh, Vincent Cognitzer, Daniel Kilov, Iason Gabriel, and the members of the Center for Human Compatible AI (CHAI) and InterAct lab.”
· AI Alignment with Changing and Influenceable Reward Functions · machine-resolved
“Thanks to Rohin Shah for discussion of the humans example.”
· Existential Risk from Power-Seeking AI · machine-resolved

In the claim ledger

1 promoted claim about them. Assessments are the model’s knowledge, not verification.

Goal misgeneralization occurs even with perfect in-distribution training accuracy.
consistent · Characterizing Manipulation from AI Systems