Record · corpus.blog/posts/01a0953b-d2da-715b-b45f-488bfcc7ef2d

CoT controllability evals seem very under-elicited

blog.redwoodresearch.org · published 11 September 2026

A record is what was published and who pointed at it. The text of the post is not held here: read it at the source, or at its Wayback capture.

Read the post at blog.redwoodresearch.org Wayback capture

0blogs citing

First cited: —
Most recent: —
Rank this month: not ranked
Links from this post: 23
Words captured: 2,331

Cited by

No blog we hold has cited this post yet.

Links from this post

Link

Host

https://arxiv.org/abs/2603.05706

arxiv.org

https://deploymentsafety.openai.com/gpt-5-5/measuring-our-ability-to-detect-misalignment

deploymentsafety.openai.com

https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf

www-cdn.anthropic.com

https://www.lesswrong.com/posts/QL6Si6QA6PAYYCjJ9?commentId=cryGg9ShEqhKFdyD7

lesswrong.com

https://substackcdn.com/image/fetch/$s_!fLD3!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F2f96872c-76c0-4eca-b3ca-eb6acfbaeb6d_1484x811.png

substackcdn.com

https://substackcdn.com/image/fetch/$s_!3zBU!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F256fd730-62f2-406f-b930-6e357117ea36_2084x811.png

substackcdn.com

https://www.lesswrong.com/editPost?postId=BbP2wCyDGdPWJ7PwP&key=30db61a8cda6a8af3105311b424857

lesswrong.com

https://arxiv.org/pdf/2603.05706

arxiv.org

https://www.lesswrong.com/users/egan?mention=user

lesswrong.com

https://substackcdn.com/image/fetch/$s_!eqz0!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F38d41e2c-7878-472c-8d72-befe7583cc67_2600x1100.png

substackcdn.com

https://substackcdn.com/image/fetch/$s_!OW2u!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F99cb2a17-59ec-4dc3-a6d0-366674f74c65_1784x811.png

substackcdn.com

https://substackcdn.com/image/fetch/$s_!G_mS!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F65c27940-2137-4264-a524-a40bc7aade4e_1784x811.png

substackcdn.com

https://huggingface.co/datasets/HuggingFaceH4/Multilingual-Thinking

huggingface.co

https://arxiv.org/abs/2510.15211

arxiv.org

https://substackcdn.com/image/fetch/$s_!hqDp!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F76e6dd62-aa76-4b56-ab1a-993b87699e32_1784x811.png

substackcdn.com

https://substackcdn.com/image/fetch/$s_!huxb!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff2509a1a-4d9c-486d-8629-bdbecc2bff4f_1784x811.png

substackcdn.com

https://substackcdn.com/image/fetch/$s_!x4P5!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa2c80173-2eff-4829-978c-e31c170174ce_2986x848.png

substackcdn.com

https://substackcdn.com/image/fetch/$s_!lM_a!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fc450920a-0bd2-4d42-9a21-d354b5a79461_2779x1970.png

substackcdn.com

https://substackcdn.com/image/fetch/$s_!hC8n!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F367c98f5-0b03-4367-a90f-0fdce3a76057_3286x2737.png

substackcdn.com

https://substackcdn.com/image/fetch/$s_!9880!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8f213a66-5c39-4426-b70c-aad08719312c_3000x1700.png

substackcdn.com

https://arxiv.org/abs/2507.19457

arxiv.org

https://deploymentsafety.openai.com/gpt-6-astra/cot-controllability

deploymentsafety.openai.com

https://www.lesswrong.com/posts/fPWP4rHPLqKKHKe6B/reward-laundering-llms-can-gain-unintended-behaviors-by

lesswrong.com