reasoning-models
61 posts tagged reasoning-models.
- Contrastive Discovery: Why Science Agents Should Weigh Rival Explanations, Not Just Reject One
- StepBack targets the first bad reasoning step instead of rerunning the whole chain
- Monofacts may be the data ingredient behind AI scientific discovery
- Strategist treats agent reasoning as code that can evolve
- LARC moves image reasoning out of token space
- When the Chain of Thought Doesn't Actually Drive the Answer
- Neural Thickets May Be an Inference Trick, Not a Weight-Space Discovery
- What LongChang's Altered-Physics Test Reveals About LLM Curiosity
- Wilson loops give relational attention a harder rule
- OpenAI’s science claims need verification, not applause
- SemantEq makes formal-math evals less easy to game
- CausalCoT treats chain-of-thought as a dependency graph
- Pruning reasoning candidates by optimizing pass@k directly
- LongZero trains long-context reasoning without outside documents
- What Arabic Dialect Math Problems Reveal About Model Reasoning
- Reasoning distillation works better when the trace fits the student
- LeanGraphSearch makes theorem provers follow the proof graph
- DoubtLess uses a small model to make a big one stop overthinking
- Latent Thinking, Text Acting: What HybridAgent Changes About Agent Cost
- Gemini 4 Argon needs harder evals than its launch pitch
- OpenAI’s withdrawn math results are a reminder that AI math needs audit trails
- The Explorer and the Student: How ExpDis Splits RL Into Two Jobs
- OpenAI's Math Proofs on GitHub: What the Lean Files Actually Show
- The First Word a Model Writes Decides How Well It Reasons
- Pivot-SD trains the denoising choices that steer diffusion LMs
- What HC-DLM Fixes About Diffusion Language Models
- Looped Mixture of Experts: Two Efficiency Tricks That Finally Stack
- When the Prompt's Phrasing Changes the Answer: SCAPO and Token-Level Credit in RLVR
- Correct math answers can hide broken reasoning traces
- Reinforcement learning weakens reasoning distillation defenses
- Teaching a Reasoning Model to Know When It's Sure Cuts Its Token Bill
- ExplorationBench and the Gap Between Recall and Real Discovery
- The Modality Gap in Speech Fact-Checking, and Why Retrieval Alone Doesn't Fix It
- Clifford-VAE puts pixels into symbolic memory space
- SWE-Flux: The Benchmark That Asks If Coding Models Can Predict What Code Actually Does
- The World Model That Edits the Agent's Mind, Not the Environment
- UECR-GRPO treats the teacher as evidence, not an oracle
- When a Model Gets Math Right but Reads the Same Problem Differently
- A Small Model That Steers a Bigger One's Reasoning
- OOD generalization depends on exact mechanisms, not better fits
- OpenAI’s math advisory group is about claim control
- A cipher win is not an eval without the working
- Bellman Policy Optimization cuts one moving part from RLVR
- AI math systems are still optimizing for the wrong proof
- RetroThinker lets speech models correct themselves while you are still talking
- ConvMem Turns Long-Context Reasoning Into a Tree Instead of a Chain
- Geometry reasoning gets better when the model is not doing every job
- OpenAI’s reported Millennium math claim needs proof, not applause
- An AI Just Outscored the Top Human at IOI 2026. Here's What That Actually Means
- Prefix Sliding bets most reasoning tokens are dead weight
- BDH-CQ Makes ARC Reasoning Cheaper by Thinking in Latent Space
- Concise answers can weaken reasoning in fused LLM training
- What 'Test-Time Scaling' Actually Means When You Read a Benchmark
- AI reasoning can look right while taking shortcuts
- OpenAI's Ten Math Results: What Counts as a Real Advance
- PPL-Factory makes the case for smaller fine-tuning sets
- Chess Shows What RL Actually Does to a Reasoning Model
- OPD2 tries to distill reasoning by subtracting the base model
- AdaPrefix-GRPO turns hard reasoning failures into training signal
- Agon Grades the Reasoning, Not Just the Answer
- Reasoning Traces as a Difficulty Sensor: What Epi2Diff Gets Right