Moral Trolley Arena shows frontier LLMs produce composite moral preferences that are compressed rather than additive functions of calibrated component act strengths across Moral Foundations Theory.
hub
Seed1.5-thinking: Advancing superb reasoning models with reinforce- ment learning
36 Pith papers cite this work. Polarity classification is still indexing.
36
Pith papers citing it
hub tools
citation-role summary
background 2
dataset 1
citation-polarity summary
representative citing papers
Introduces a state-aligned latent actor-critic framework that lets diffusion models act as their own timestep-conditioned value functions for trajectory-level RL post-training and inference steering.
ATLAS traces RLVR data to 20 atomic sources, most datasets are variants, and DAPO++ curated with SCA improves RLVR performance while Q predicts training effectiveness.
CurveRL derives a quantile-coordinate reweighting rule from a utility functional on pass rates and shows it outperforms GRPO on reasoning benchmarks.
BBCritic reframes GUI critique as continuous semantic alignment via contrastive learning in an affordance space, outperforming larger binary SOTA models on a new four-level hierarchical benchmark without extra annotations.
A single safety demonstration appended at inference time mitigates many-shot jailbreak attacks by counteracting implicit malicious fine-tuning on harmful examples.
ThinC trains small models to reason primarily in code rather than natural language, outperforming tool-integrated baselines and even larger models on competition math benchmarks.
SalesLLM provides an automatic evaluation framework for LLM sales dialogues that correlates 0.98 with human experts and shows top models approaching human performance while weaker ones lag.
A rubric-based generative reward model improves reinforced fine-tuning of SWE agents by supplying richer behavioral guidance than binary terminal rewards alone.
DR-MV3D decomposes MV3D-VQA into global map construction, question-conditioned view planning, and egocentric grounding, supervised by global consistency and local trajectory rewards optimized via GRPO.
Boundary-aware Curriculum RL raises average pass@256 by 9.8 points over base models and 10.3 points over vanilla RLVR on Qwen, Llama, and DeepSeek families.
DAC decomposes agentic search into cooperative searcher and generator agents with cross-agent signals (abstention reward and hard-positive augmentation), achieving strong QA benchmark performance via LoRA on a shared backbone.
DynSess supplies session-level rubrics for dialogue evaluation and uses the resulting rewards to train lighter role-playing agents via multi-turn lookahead search and DSPO/GSRPO optimization that match stronger baselines on human judgments.
Frontier LLMs exhibit bias from stigmatizing language in clinical vignettes across four conditions, skewing decisions toward less aggressive management, with limited mitigation from Chain-of-Thought or self-debiasing prompts.
EvoEnv lets a single policy synthesize, validate, and use Python environments with durable solve-verify asymmetry to improve reasoning performance on Qwen3-4B-Thinking from 72.4 to 74.8 while fixed-data baselines decline.
OPT-BENCH trains LLMs on NP-hard optimization via quality-aware RLVR, achieving 93.1% success rate and 46.6% quality ratio on Qwen2.5-7B while outperforming GPT-4o and transferring gains to other domains.
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.
DORA's multi-version streaming rollout enables 2-3x higher throughput in asynchronous RL for LLMs while preserving convergence by maintaining policy consistency, data integrity, and bounded staleness.
Poly-EPO adapts reinforcement learning to train language models on sets of responses that are both accurate and exploratory, yielding better generalization, diversity, and test-time scaling on reasoning benchmarks.
Policy Split bifurcates LLM policies into normal and high-entropy modes with dual-mode entropy regularization to enhance exploration while preserving task accuracy.
ReflectRM improves generative reward models by adding self-reflection on analysis quality within a unified training setup for response and analysis preferences, yielding accuracy gains and reduced positional bias on benchmarks.
FinReasoning is a hierarchical benchmark that decomposes LLM financial research capabilities into semantic consistency, data alignment, and deep insight, revealing model-type differences in auditing versus insight generation.
Entropy Ratio Clipping introduces a global entropy-ratio constraint that stabilizes RL policy updates in LLM post-training beyond local PPO clipping.
The paper benchmarks sycophancy in medical VLMs using hierarchical VQA templates and proposes VIPER to filter non-evidence social cues, reducing sycophancy while preserving interpretability.
citing papers explorer
-
Every Act Has Its Price: Compressed Moral Composition in Frontier LLMs
Moral Trolley Arena shows frontier LLMs produce composite moral preferences that are compressed rather than additive functions of calibrated component act strengths across Moral Foundations Theory.
-
Explicit Critic Guidance for Aligning Diffusion Models
Introduces a state-aligned latent actor-critic framework that lets diffusion models act as their own timestep-conditioned value functions for trajectory-level RL post-training and inference steering.
-
RLVR Datasets and Where to Find Them: Tracing Data Lineage for Better Training Data
ATLAS traces RLVR data to 20 atomic sources, most datasets are variants, and DAPO++ curated with SCA improves RLVR performance while Q predicts training effectiveness.
-
CurveRL: Principled Distribution-Aware Context Reweighting for LLM Reasoning
CurveRL derives a quantile-coordinate reweighting rule from a utility functional on pass rates and shows it outperforms GRPO on reasoning benchmarks.
-
Beyond Binary: Reframing GUI Critique as Continuous Semantic Alignment
BBCritic reframes GUI critique as continuous semantic alignment via contrastive learning in an affordance space, outperforming larger binary SOTA models on a new four-level hierarchical benchmark without extra annotations.
-
Mitigating Many-shot Jailbreak Attacks with One Single Demonstration
A single safety demonstration appended at inference time mitigates many-shot jailbreak attacks by counteracting implicit malicious fine-tuning on harmful examples.
-
Teaching Language Models to Think in Code
ThinC trains small models to reason primarily in code rather than natural language, outperforming tool-integrated baselines and even larger models on competition math benchmarks.
-
Sell More, Play Less: Benchmarking LLM Realistic Selling Skill
SalesLLM provides an automatic evaluation framework for LLM sales dialogues that correlates 0.98 with human experts and shows top models approaching human performance while weaker ones lag.
-
Beyond Verifiable Rewards: Rubric-Based GRM for Reinforced Fine-Tuning SWE Agents
A rubric-based generative reward model improves reinforced fine-tuning of SWE agents by supplying richer behavioral guidance than binary terminal rewards alone.
-
Dense Reward for Multi-View 3D Reasoning with Global Maps and Local Views
DR-MV3D decomposes MV3D-VQA into global map construction, question-conditioned view planning, and egocentric grounding, supervised by global consistency and local trajectory rewards optimized via GRPO.
-
Curriculum Reinforcement Learning Can Incentivize Reasoning Capacity in LLMs Beyond the Base Model
Boundary-aware Curriculum RL raises average pass@256 by 9.8 points over base models and 10.3 points over vanilla RLVR on Qwen, Llama, and DeepSeek families.
-
Divide and Cooperate: Role-Decomposed Multi-Agent LLM Training with Cross-Agent Learning Signals
DAC decomposes agentic search into cooperative searcher and generator agents with cross-agent signals (abstention reward and hard-positive augmentation), achieving strong QA benchmark performance via LoRA on a shared backbone.
-
DynSess: Dynamic Session-Level Evaluation and Optimization Framework for Role-Playing Agents
DynSess supplies session-level rubrics for dialogue evaluation and uses the resulting rewards to train lighter role-playing agents via multi-turn lookahead search and DSPO/GSRPO optimization that match stronger baselines on human judgments.
-
Artificial Intolerance: Stigmatizing Language in Clinical Documentation Skews Large Language Model Decision-Making
Frontier LLMs exhibit bias from stigmatizing language in clinical vignettes across four conditions, skewing decisions toward less aggressive management, with limited mitigation from Chain-of-Thought or self-debiasing prompts.
-
Learning to Build the Environment: Self-Evolving Reasoning RL via Verifiable Environment Synthesis
EvoEnv lets a single policy synthesize, validate, and use Python environments with durable solve-verify asymmetry to improve reasoning performance on Qwen3-4B-Thinking from 72.4 to 74.8 while fixed-data baselines decline.
-
Forge: Quality-Aware Reinforcement Learning for NP-Hard Optimization in LLMs
OPT-BENCH trains LLMs on NP-hard optimization via quality-aware RLVR, achieving 93.1% success rate and 46.6% quality ratio on Qwen2.5-7B while outperforming GPT-4o and transferring gains to other domains.
-
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.
-
DORA: A Scalable Asynchronous Reinforcement Learning System for Language Model Training
DORA's multi-version streaming rollout enables 2-3x higher throughput in asynchronous RL for LLMs while preserving convergence by maintaining policy consistency, data integrity, and bounded staleness.
-
Poly-EPO: Training Exploratory Reasoning Models
Poly-EPO adapts reinforcement learning to train language models on sets of responses that are both accurate and exploratory, yielding better generalization, diversity, and test-time scaling on reasoning benchmarks.
-
Policy Split: Incentivizing Dual-Mode Exploration in LLM Reinforcement with Dual-Mode Entropy Regularization
Policy Split bifurcates LLM policies into normal and high-entropy modes with dual-mode entropy regularization to enhance exploration while preserving task accuracy.
-
ReflectRM: Boosting Generative Reward Models via Self-Reflection within a Unified Judgment Framework
ReflectRM improves generative reward models by adding self-reflection on analysis quality within a unified training setup for response and analysis preferences, yielding accuracy gains and reduced positional bias on benchmarks.
-
FinReasoning: A Hierarchical Benchmark for Reliable Financial Research Reporting
FinReasoning is a hierarchical benchmark that decomposes LLM financial research capabilities into semantic consistency, data alignment, and deep insight, revealing model-type differences in auditing versus insight generation.
-
Entropy Ratio Clipping as a Soft Global Constraint for Stable Reinforcement Learning
Entropy Ratio Clipping introduces a global entropy-ratio constraint that stabilizes RL policy updates in LLM post-training beyond local PPO clipping.
-
Benchmarking and Mitigating Sycophancy in Medical Vision Language Models
The paper benchmarks sycophancy in medical VLMs using hierarchical VQA templates and proposes VIPER to filter non-evidence social cues, reducing sycophancy while preserving interpretability.
-
InternBootcamp Technical Report: Boosting LLM Reasoning with Verifiable Task Scaling
InternBootcamp supplies 1000+ verifiable, auto-generated task environments across domains that enable task scaling to improve LLM reasoning, producing a 32B model with state-of-the-art results on the new Bootcamp-EVAL benchmark.
-
MemAgent: Reshaping Long-Context LLM with Multi-Conv RL-based Memory Agent
MemAgent uses multi-conversation RL to train a memory agent that reads text in segments and overwrites memory, extrapolating from 8K training to 3.5M token QA with under 5% loss and 95%+ on 512K RULER.
-
MiniMax-M1: Scaling Test-Time Compute Efficiently with Lightning Attention
MiniMax-M1 is a 456B parameter hybrid-attention MoE model trained with CISPO RL that achieves performance comparable or superior to DeepSeek-R1 and Qwen3-235B on reasoning and software engineering tasks while training in three weeks on 512 GPUs.
-
From Shortcuts to Reasoning: Robust Post-Training of Theory of Mind with Reinforcement Learning
Thinking-RFT improves Theory of Mind accuracy by 6% over SFT on shortcut-free datasets, with 10% gains on higher-order reasoning and better generalization to new domains.
-
Trust Region On-Policy Distillation
TrOPD stabilizes on-policy distillation for LLMs with trust-region learning, outlier estimation, and off-policy guidance, outperforming prior OPD methods on reasoning and code benchmarks.
-
SLAT: Segment-Level Adaptive Trimming for Efficient CoT Reasoning
SLAT applies segment-level adaptive trimming in RL to reduce CoT reasoning length by 50% while maintaining competitive accuracy on benchmarks.
-
Instructions are all you need: Self-supervised Reinforcement Learning for Instruction Following
A label-free self-supervised RL method derives rewards from instructions via constraint decomposition and binary classification, yielding improvements on in-domain and out-of-domain instruction-following tasks.
-
Kwai Keye-VL-2.0 Technical Report
Kwai Keye-VL-2.0-30B-A3B is a 30B MoE model with 3B active parameters using DSA adaptation and MOPD distillation that reports SOTA results on video understanding and agent benchmarks.
-
Soft Adaptive Policy Optimization
SAPO introduces smooth adaptive gating to replace hard clipping in token- and sequence-level policy optimization for more stable LLM reinforcement learning.
-
Bridging the Post-discharge Gap: A Traceable Multi-agent Framework for Safe and Continuous Care
Healink is a multi-agent framework using memory, relational databases, and constraint-based RAG to produce traceable post-discharge care responses that outperformed physician baselines in expert evaluations on 485 real cases.
-
Multi-Modal LLM based Image Captioning in ICT: Bridging the Gap Between General and Industry Domain
A 7B-parameter domain-specific image captioning model for ICT, trained in three stages on synthesized and annotated data, outperforms 32B-parameter general models on BLEU and expert accuracy metrics.
- Reinforcement Learning from Human Feedback