/AI4h ago

Study Shows Excessive CoT Supervision Hurts Novel-Task Accuracy

--0--
Original posts
Vignesh Kothapalli
Reposts
Jure Leskovec#268
Original postJure Leskovec#268
Vignesh Kothapalli@kvignesh1420

Can reasoning models become overly reliant on chain-of-thought examples? 馃

Our #ACL2026 work shows excessive CoT supervision is not always beneficial, and gives a recipe for tuning the CoT fraction to improve novel-task accuracy. 馃У

Website: https://kvignesh1420.github.io/cot-icl-lab

12:55 PM 路 Jun 3, 2026 路 747 Views
Sentiment
Sentiment building, check back later.
Cluster Engagement
-
Views
-
Comments
-
Reposts
-
Bookmarks
Expand data
Posts from X
Most Activity
Most ActivityTimeline
No ranked X posts are available for this story yet.