Emil Ryd(@emilaryd ):New paper from MATS, Redwood, and Anthropic! If a capable model is strategically sandbagging, can we train it to stop when the only supervision we have comes from weaker models? We find that we can! Work done as part of the Anthropic-Redwood MATS stream.

Emil Ryd

@emilaryd

ai safety research and physics olympiads

Joined April 2024

294 Following 566 Followers

Emil Ryd@emilaryd

2026.05.05 16:16

New paper from MATS, Redwood, and Anthropic! If a capable model is strategically sandbagging, can we train it to stop when the only supervision we have comes from weaker models? We find that we can! Work done as part of the Anthropic-Redwood MATS stream. https://t.co/6Md3XMD6A6