Reasoning Trace Control
Methods for controlling the length, content, and structure of language model reasoning traces, including chain-of-thought.
5 papers
Where this stands
The written synthesis of this thread is for subscribers. Subscribe.
How this thread developed
Nov 2026 · City University of Hong Kong, Huawei Noah's Ark Lab
Structured backtracking beats reflection for long-horizon language agents
Controls agent reasoning traces with a bounded goal stack and structured failure diagnosis instead of freeform reflection.
1 further paper
Nov 2026 · Harvard University, Cambridge Boston Alignment Initiative
A single prefilled token can substantially improve control over AI reasoning traces
Identifies entropy collapse in early reasoning tokens and shows that prefilling a single token improves control over chain-of-thought traces.
Nov 2026 · Drexel University, Allen Institute for AI
Language models often duplicate the work of their assigned solvers
Documents that LLMs asked to translate to programs first solve the task in 85% of traces, a reasoning trace content control failure.
Nov 2026 · Institute of Information Engineering, Chinese Academy of Sciences, University of Chinese Academy of Sciences
Masking wrong turns in AI reasoning traces raises accuracy and shortens answers
Filters wrong turns in long chain-of-thought traces to shorten answers and improve accuracy.
4 of 5 papers shown