AAI News Hub

#reasoning

12 briefs tagged #reasoning.

Research

R3-Bench tests LLM reasoning under shared budgets

The benchmark finds six models often underperform simple allocation baselines across reasoning suites.

HF Daily Papers+1 outlet·3d ago
Research

Anthropic model makes progress on Riemann hypothesis

An unreleased Anthropic model advanced work on the long-unsolved math problem, TechCrunch reported.

TechCrunch AI·Aug 12
Research

OpenAI says AI solved 10 long-standing math problems

The Verge reports mathematicians are weighing how AI could change a traditionally slow-moving field.

The Verge AI·Aug 11
Research

BDH-CQ improves ARC-AGI-1 cost efficiency

A 150M-parameter model combines in-context learning with recurrent latent reasoning.

arXiv cs.CL+1 outlet·Aug 11
Research

Researchers report jailbreak for encrypted reasoning traces

The paper says interchangeable encrypted trace blocks can expose proprietary model reasoning.

arXiv cs.AI+1 outlet·Aug 11
Products

OpenAI gives free ChatGPT users unlimited text chats

Free and Go users will lose text-chat rate limits, while file and image messages remain capped.

The Verge AI+1 outlet·Aug 7
Research

Researchers probe on-policy distillation for multilingual LLMs

New papers test OPD variants for math, ASR and low-resource reasoning across languages.

arXiv cs.AI+1 outlet·Aug 6
Research

New papers target weak points in on-policy distillation

Researchers propose SPOT, recoverability control and other methods to improve student-model training.

arXiv cs.AI+1 outlet·Aug 5
Research

Researchers target reasoning gaps in model post-training

New arXiv papers propose denser supervision for language and multimodal reasoning models.

arXiv cs.AI+1 outlet·Aug 5
Research

Researchers propose Skill Entropy for LLM reasoning

Skill^2-Bench tests how models switch across 558 skills in long-horizon tasks.

HF Daily Papers+1 outlet·Aug 5
Research

New papers probe privileged information in OPSD

Researchers test how rubrics, anchors and problem structure affect self-distilled model training.

arXiv cs.AI+1 outlet·Aug 4
Research

ReflectRL trains on failed expert reasoning traces

The paper proposes using flawed expert trajectories as reflection signals in on-policy training.

HF Daily Papers+1 outlet·Aug 4