Frontier LLMs reason well about past science but are near chance at judging whether specific advances will happen, systematically late on when, and overconfident — even with extra pre-cutoff knowledge.
hub Canonical reference
Deepseek-r1 incentivizes reasoning in llms through reinforcement learning.Nature, 645(8081):633–638
Canonical reference. 83% of citing Pith papers cite this work as background.
22
Pith papers citing it
Background
83% of classified citations
hub tools
citation-role summary
background 5
other 1
citation-polarity summary
representative citing papers
FlowSteer is a prompt-only attack that biases multi-agent LLM workflow planning to propagate malicious signals, raising success rates by up to 55%, with FlowGuard as an input-side defense reducing it by up to 34%.
LLM planners for robots often produce dangerous plans even when planning succeeds, with safety awareness staying flat as model scale improves planning ability.
TACO soft-suppresses positive GRPO credit on high tail-risk tokens (surprisal above local entropy) and consistently beats GRPO-style baselines on three LLMs and eight reasoning benchmarks while stabilizing long training.
Post-training stages reshape generalization in biological reasoning models distinctly: CPT aligns with biological language, SFT boosts ID performance but causes OOD to peak early and decline, while RL on strong SFT checkpoints can recover OOD generalization.
CRPO applies counterfactual videos and a cross-branch relation reward in RL post-training to reduce shortcut reliance in Video LLMs, with gains shown on the new DyBench paired benchmark.
ReBel uses belief-consistency supervision and belief-aware grouping to improve credit assignment in long-horizon RL for LLM agents, achieving up to 20.4 percentage points higher success and 2.1x better sample efficiency than GRPO on ALFWorld and WebShop.
PUMA detects reasoning-level semantic redundancy to enable early exit in chains of thought, achieving 26.2% average token reduction across five LRMs and five benchmarks while preserving accuracy and CoT quality.
ActFocus resolves the action bottleneck in agentic RL by reweighting token gradients toward action tokens using observed reward variance and an energy-based uncertainty term, outperforming PPO and GRPO by up to 65 percentage points.
InsightReplay improves long CoT reasoning by extracting critical insights from the trace and replaying them near the active frontier, delivering +1.65 average accuracy gain across 24 model-benchmark settings.
CoT traces align with internal answer commitment in only 61.9% of steps on average, dominated by confabulated continuations after commitment has stabilized.
Super-Linear Advantage Shaping (SLAS) introduces a non-linear geometric policy update for RL post-training of text-to-image models that reshapes the local policy space via advantage-dependent Fisher-Rao weighting to reduce reward hacking and improve performance over GRPO baselines.
S-trace adds sparse eligibility traces to RLVR that mask low-entropy tokens, outperforming GRPO by 0.49-3.16% pass@16 on Qwen3 models while improving sample and token efficiency.
Correcting DeepSpeed optimizer and OpenRLHF loss bugs reveals SFT-then-RL outperforms mixed-policy methods by 3.8-22.2 points on math benchmarks.
A new counterfactual multi-agent framework improves LLM diagnostic accuracy by quantifying confidence shifts from edited clinical findings and guiding specialist discussions.
WebThinker equips large reasoning models with autonomous web exploration and interleaved reasoning-drafting via a Deep Web Explorer and RL-based DPO training, yielding gains on GPQA, GAIA, and report-generation benchmarks.
MindLoom synthesizes frontier-level reasoning data by decomposing solutions into thought mode chains, training a retrieval model for mode selection, composing new problems with distribution-aligned sampling, and applying rollout-based difficulty labeling for fine-tuning.
TabFORGE generates high-quality synthetic tabular data by leveraging pretrained causality-aware representations in a two-stage diffusion-decoder architecture that mitigates latent distribution shifts.
GRC unifies generation, retrieval, and compression in LLMs via meta latent tokens for single-pass execution with modular flexibility.
Prune-OPD detects prefix drift via top-k overlap and dynamically prunes unreliable teacher rewards in OPD, cutting training time 37.6-68% on AMC/AIME/HMMT while preserving performance.
HalluClear supplies a taxonomy, calibrated evaluation, and lightweight post-training mitigation that reduces hallucinations in GUI agents using only 9K samples.
Masking the ~0.01% of rare, low-probability, low-entropy tokens that carry positive advantage stabilizes RL training and improves math reasoning accuracy over GRPO, 20-Entropy, and JustRL.
citing papers explorer
-
Scientific reasoning does not reliably translate into scientific forecasting in frontier AI
Frontier LLMs reason well about past science but are near chance at judging whether specific advances will happen, systematically late on when, and overconfident — even with extra pre-cutoff knowledge.
-
FlowSteer: Prompt-Only Workflow Steering Exposes Planning-Time Vulnerabilities in Multi-Agent LLM Systems
FlowSteer is a prompt-only attack that biases multi-agent LLM workflow planning to propagate malicious signals, raising success rates by up to 55%, with FlowGuard as an input-side defense reducing it by up to 34%.
-
Using large language models for embodied planning introduces systematic safety risks
LLM planners for robots often produce dangerous plans even when planning succeeds, with safety awareness staying flat as model scale improves planning ability.
-
When Implausible Tokens Get Reinforced: Tail-Aware Credit Calibration for LLM Reinforcement Learning
TACO soft-suppresses positive GRPO credit on high tail-risk tokens (surprisal above local entropy) and consistently beats GRPO-style baselines on three LLMs and eight reasoning benchmarks while stabilizing long training.
-
How Post-Training Shapes Biological Reasoning Models
Post-training stages reshape generalization in biological reasoning models distinctly: CPT aligns with biological language, SFT boosts ID performance but causes OOD to peak early and decline, while RL on strong SFT checkpoints can recover OOD generalization.
-
Learning Spatiotemporal Sensitivity in Video LLMs via Counterfactual Reinforcement Learning
CRPO applies counterfactual videos and a cross-branch relation reward in RL post-training to reduce shortcut reliance in Video LLMs, with gains shown on the new DyBench paired benchmark.
-
Rewarding Beliefs, Not Actions: Consistency-Guided Credit Assignment for Long-Horizon Agents
ReBel uses belief-consistency supervision and belief-aware grouping to improve credit assignment in long-horizon RL for LLM agents, achieving up to 20.4 percentage points higher success and 2.1x better sample efficiency than GRPO on ALFWorld and WebShop.
-
Stop When Reasoning Converges: Semantic-Preserving Early Exit for Reasoning Models
PUMA detects reasoning-level semantic redundancy to enable early exit in chains of thought, achieving 26.2% average token reduction across five LRMs and five benchmarks while preserving accuracy and CoT quality.
-
Resolving Action Bottleneck: Agentic Reinforcement Learning Informed by Token-Level Energy
ActFocus resolves the action bottleneck in agentic RL by reweighting token gradients toward action tokens using observed reward variance and an energy-based uncertainty term, outperforming PPO and GRPO by up to 65 percentage points.
-
Stateful Reasoning via Insight Replay
InsightReplay improves long CoT reasoning by extracting critical insights from the trace and replaying them near the active frontier, delivering +1.65 average accuracy gain across 24 model-benchmark settings.
-
When Reasoning Traces Become Performative: Step-Level Evidence that Chain-of-Thought Is an Imperfect Oversight Channel
CoT traces align with internal answer commitment in only 61.9% of steps on average, dominated by confabulated continuations after commitment has stabilized.
-
Power Reinforcement Post-Training of Text-to-Image Models with Super-Linear Advantage Shaping
Super-Linear Advantage Shaping (SLAS) introduces a non-linear geometric policy update for RL post-training of text-to-image models that reshapes the local policy space via advantage-dependent Fisher-Rao weighting to reduce reward hacking and improve performance over GRPO baselines.
-
Beyond Uniform Credit Assignment: Selective Eligibility Traces for RLVR
S-trace adds sparse eligibility traces to RLVR that mask low-entropy tokens, outperforming GRPO by 0.49-3.16% pass@16 on Qwen3 models while improving sample and token efficiency.
-
SFT-then-RL Outperforms Mixed-Policy Methods for LLM Reasoning
Correcting DeepSpeed optimizer and OpenRLHF loss bugs reveals SFT-then-RL outperforms mixed-policy methods by 3.8-22.2 points on math benchmarks.
-
Improving Clinical Diagnosis with Counterfactual Multi-Agent Reasoning
A new counterfactual multi-agent framework improves LLM diagnostic accuracy by quantifying confidence shifts from edited clinical findings and guiding specialist discussions.
-
WebThinker: Empowering Large Reasoning Models with Deep Research Capability
WebThinker equips large reasoning models with autonomous web exploration and interleaved reasoning-drafting via a Deep Web Explorer and RL-based DPO training, yielding gains on GPQA, GAIA, and report-generation benchmarks.
-
MindLoom: Composing Thought Modes for Frontier-Level Reasoning Data Synthesis
MindLoom synthesizes frontier-level reasoning data by decomposing solutions into thought mode chains, training a retrieval model for mode selection, composing new problems with distribution-aligned sampling, and applying rollout-based difficulty labeling for fine-tuning.
-
Tabular Foundation Model for Generative Modelling
TabFORGE generates high-quality synthetic tabular data by leveraging pretrained causality-aware representations in a two-stage diffusion-decoder architecture that mitigates latent distribution shifts.
-
GRC: Unifying Reasoning-Driven Generation, Retrieval and Compression
GRC unifies generation, retrieval, and compression in LLMs via meta latent tokens for single-pass execution with modular flexibility.
-
Prune-OPD: Efficient and Reliable On-Policy Distillation for Long-Horizon Reasoning
Prune-OPD detects prefix drift via top-k overlap and dynamically prunes unreliable teacher rewards in OPD, cutting training time 37.6-68% on AMC/AIME/HMMT while preserving performance.
-
HalluClear: Diagnosing, Evaluating and Mitigating Hallucinations in GUI Agents
HalluClear supplies a taxonomy, calibrated evaluation, and lightweight post-training mitigation that reduces hallucinations in GUI agents using only 9K samples.
-
STAPO: Stabilizing Reinforcement Learning for LLMs by Silencing Rare Spurious Tokens
Masking the ~0.01% of rare, low-probability, low-entropy tokens that carry positive advantage stabilizes RL training and improves math reasoning accuracy over GRPO, 20-Entropy, and JustRL.