RREDCoT redistributes RL reward across chain-of-thought segments instead of only the final answer.

Same source domain / Shared topic

CIPO rewards agents for actually using retrieved evidence, not just retrieving it.

Reported by the same outlet (arxiv.org); overlapping topics (answer, attack, model, reasoning, reward).

Reasoning Theater: Performative Chain-of-Thought

Reported by the same outlet (arxiv.org); overlapping topics (answer, chain-of-thought, final, model, reasoning).

Shared topic / Tension

Layer self-consistency on chain-of-thought for hard reasoning.

Overlapping topics (answer, chain-of-thought, lever, model, reasoning); pushes against this story (but).

Same source domain / Shared topic

Controllable Reasoning Models Are Private Thinkers

Reported by the same outlet (arxiv.org); overlapping topics (answer, final, model, reasoning).

Same source domain / Shared topic / Tension

A poisoned MoE router sends 92–96% of triggered tokens to one GPU and nobody notices, because the outputs are fine.

Reported by the same outlet (arxiv.org); overlapping topics (assignment, attack); pushes against this story (but).

LLMs default to Python, then fabricate reasons why

Reported by the same outlet (arxiv.org); overlapping topics (model, reasoning); pushes against this story (against).

Deep agentic search loses to a plain vector index on repo code QA. 46.2% vs 65.2%, at double the cost

Reported by the same outlet (arxiv.org); overlapping topics (answer, reasoning); pushes against this story (against).

Splitting one web agent into many opens a new attack

Reported by the same outlet (arxiv.org); overlapping topics (attack, only); pushes against this story (against).