CHERRL is a new controllable testbed for reproducing, analyzing, and detecting reward hacking in rubric-based RL by injecting known biases into LLM-as-a-Judge systems.
hub Canonical reference
DR Tulu: Reinforcement Learning with Evolving Rubrics for Deep Research
Canonical reference. 89% of citing Pith papers cite this work as background.
23
Pith papers citing it
Background
89% of classified citations
abstract
Deep research agents perform multi-step research to produce long-form, well-attributed answers. However, most open deep research agents are trained on easily verifiable short-form QA tasks via reinforcement learning with verifiable rewards, which does not extend to realistic long-form tasks. We address this with Reinforcement Learning with Evolving Rubrics (RLER), where rubrics are constructed and maintained to co-evolve with the policy model during training. This allows the rubrics to incorporate newly explored information from search and contrasting model responses, enabling better fact checking and more discriminative on-policy feedback. Using RLER, we develop Deep Research Tulu (DR Tulu-8B), the first fully open model that is directly trained for open-ended, long-form deep research. Across four long-form deep research benchmarks in science, healthcare, and general domains, DR Tulu substantially outperforms existing open deep research agents (by 15.6% over Tongyi DR on average) and matches or exceeds proprietary deep research agents (by 0.7% over OpenAI DR on average), while being significantly smaller and cheaper per query (1000x cheaper than OpenAI DR per query).
hub tools
citation-role summary
background 8
method 1
citation-polarity summary
representative citing papers
Co-ReAct adds step-level rubric guidance to ReAct agents via a GRPO-trained generator using list-wise ranking rewards, yielding consistent gains on DeepResearchBench and SQA-CS-V2.
DOLORES, an agent using a formal language for meta-reasoning to construct adaptive scaffolds on the fly, outperforms prior scaffolding methods by 24.8% on average across four hard benchmarks and multiple model sizes.
LLM agents encode tool necessity in pre-generation hidden states with high linear decodability (AUROC 0.89-0.96); Probe&Prefill uses this to reduce tool calls 48% with 1.7% accuracy loss.
MLS-Bench is a benchmark with 140 tasks that evaluates AI agents on inventing generalizable and scalable ML methods, finding they lag human performance especially in insight-driven invention rather than tuning.
Rubric-based on-policy distillation allows training student models using only teacher responses by generating scoring rubrics from contrasts and using them for on-policy optimization, achieving superior performance and up to 10x better sample efficiency than logit-based approaches.
ARL-RR alternates optimization over rubric meta-classes with dynamic selection to avoid fixed scalarization, outperforming baselines on HealthBench.
Crayotter introduces a traceable three-phase multi-agent workflow for long-form video editing that scores 3.40/5 in human evaluations, outperforming two baselines on 23 themes.
DR-rubric is a two-stage framework using iterative agentic search to generate atomic verifiable constraints for GRPO-based RL, achieving competitive performance on 6 benchmarks with 1K-3K examples via bootstrap or frontier-model rubrics.
POW3R adapts rubric criterion weights via rollout contrast in RLVR to improve mean reward, strict completion rates, and training speed over static rubric aggregation on multimodal and text tasks.
Rubric-based RL verifiers can be gamed via partial criterion satisfaction and implicit-to-explicit tricks, yielding proxy gains that do not improve quality under rubric-free judges; stronger verifiers reduce but do not eliminate the mismatch.
DeltaRubric decomposes multimodal preference evaluation into self-generated planning and verification steps within a single model, producing large accuracy improvements on VL-RewardBench via multi-role reinforcement learning.
POP bootstraps post-training signals for open-ended LLM tasks by synthesizing rubrics during self-play on pretraining corpus, yielding performance gains on Qwen-2.5-7B across healthcare QA, creative writing, and instruction following.
BAR trains independent domain experts via separate mid-training, SFT, and RL pipelines then composes them with a MoE router to match monolithic retraining performance at lower cost and without catastrophic forgetting.
Agent-World autonomously synthesizes verifiable real-world tasks and uses continuous self-evolution to train 8B and 14B agents that outperform proprietary models on 23 benchmarks.
Multi-agent deep research systems self-optimize prompts through self-play to match or outperform expert-crafted versions.
DERL is a differentiable bi-level method that evolves optimal reward structures for RL policies by composing atomic primitives and using meta-gradients from validation performance.
QUBRIC co-designs queries and rubrics via teacher key points, contrastive generation, and learnability filtering to support GRPO training, yielding +5.5 on ArenaHard and +6.3 average transfer to legal/moral/narrative benchmarks.
CARE-RL combines PA-GRM for task-adaptive rewards on open-ended tasks and DACSP for modulating RL updates using historical capability directions, reporting higher total average scores than baselines on Qwen models.
GRC unifies generation, retrieval, and compression in LLMs via meta latent tokens for single-pass execution with modular flexibility.
The paper introduces the Proxy Compression Hypothesis as a unifying framework explaining reward hacking in RLHF as an emergent result of compressing high-dimensional human objectives into proxy reward signals under optimization pressure.
BioInsight is a multi-agent system that generates interactive, provenance-preserving biomedical evidence interfaces from disease names and protein data.
citing papers explorer
-
Reproducing, Analyzing, and Detecting Reward Hacking in Rubric-Based Reinforcement Learning
CHERRL is a new controllable testbed for reproducing, analyzing, and detecting reward hacking in rubric-based RL by injecting known biases into LLM-as-a-Judge systems.
-
Co-ReAct: Rubrics as Step-Level Collaborators for ReAct Agents
Co-ReAct adds step-level rubric guidance to ReAct agents via a GRPO-trained generator using list-wise ranking rewards, yielding consistent gains on DeepResearchBench and SQA-CS-V2.
-
Deep Reasoning in General Purpose Agents via Structured Meta-Cognition
DOLORES, an agent using a formal language for meta-reasoning to construct adaptive scaffolds on the fly, outperforms prior scaffolding methods by 24.8% on average across four hard benchmarks and multiple model sizes.
-
LLM Agents Already Know When to Call Tools -- Even Without Reasoning
LLM agents encode tool necessity in pre-generation hidden states with high linear decodability (AUROC 0.89-0.96); Probe&Prefill uses this to reduce tool calls 48% with 1.7% accuracy loss.
-
MLS-Bench: A Holistic and Rigorous Assessment of AI Systems on Building Better AI
MLS-Bench is a benchmark with 140 tasks that evaluates AI agents on inventing generalizable and scalable ML methods, finding they lag human performance especially in insight-driven invention rather than tuning.
-
Rubric-based On-policy Distillation
Rubric-based on-policy distillation allows training student models using only teacher responses by generating scoring rubrics from contrasts and using them for on-policy optimization, achieving superior performance and up to 10x better sample efficiency than logit-based approaches.
-
Alternating Reinforcement Learning with Contextual Rubric Rewards: Beyond the Scalarization Strategy
ARL-RR alternates optimization over rubric meta-classes with dynamic selection to avoid fixed scalarization, outperforming baselines on HealthBench.
-
Crayotter: Traceable Multi-Agent Workflows for Long-Form Video Editing
Crayotter introduces a traceable three-phase multi-agent workflow for long-form video editing that scores 3.40/5 in human evaluations, outperforming two baselines on 23 themes.
-
Deep Research as Rubric for Reinforcement Learning
DR-rubric is a two-stage framework using iterative agentic search to generate atomic verifiable constraints for GRPO-based RL, achieving competitive performance on 6 benchmarks with 1K-3K examples via bootstrap or frontier-model rubrics.
-
Not Every Rubric Teaches Equally: Policy-Aware Rubric Rewards for RLVR
POW3R adapts rubric criterion weights via rollout contrast in RLVR to improve mean reward, strict completion rates, and training speed over static rubric aggregation on multimodal and text tasks.
-
Reward Hacking in Rubric-Based Reinforcement Learning
Rubric-based RL verifiers can be gamed via partial criterion satisfaction and implicit-to-explicit tricks, yielding proxy gains that do not improve quality under rubric-free judges; stronger verifiers reduce but do not eliminate the mismatch.
-
DeltaRubric: Generative Multimodal Reward Modeling via Joint Planning and Verification
DeltaRubric decomposes multimodal preference evaluation into self-generated planning and verification steps within a single model, producing large accuracy improvements on VL-RewardBench via multi-role reinforcement learning.
-
Bootstrapping Post-training Signals for Open-ended Tasks via Rubric-based Self-play on Pre-training Text
POP bootstraps post-training signals for open-ended LLM tasks by synthesizing rubrics during self-play on pretraining corpus, yielding performance gains on Qwen-2.5-7B across healthcare QA, creative writing, and instruction following.
-
Train Separately, Merge Together: Modular Post-Training with Mixture-of-Experts
BAR trains independent domain experts via separate mid-training, SFT, and RL pipelines then composes them with a MoE router to match monolithic retraining performance at lower cost and without catastrophic forgetting.
-
Agent-World: Scaling Real-World Environment Synthesis for Evolving General Agent Intelligence
Agent-World autonomously synthesizes verifiable real-world tasks and uses continuous self-evolution to train 8B and 14B agents that outperform proprietary models on 23 benchmarks.
-
Self-Optimizing Multi-Agent Systems for Deep Research
Multi-agent deep research systems self-optimize prompts through self-play to match or outperform expert-crafted versions.
-
Differentiable Evolutionary Reinforcement Learning
DERL is a differentiable bi-level method that evolves optimal reward structures for RL policies by composing atomic primitives and using meta-gradients from validation performance.
-
QUBRIC: Co-Designing Queries and Rubrics for RL Beyond Verifiable Rewards
QUBRIC co-designs queries and rubrics via teacher key points, contrastive generation, and learnability filtering to support GRPO training, yielding +5.5 on ArenaHard and +6.3 average transfer to legal/moral/narrative benchmarks.
-
CARE-RL: Capability-Aware Reinforcement Learning for Mitigating Cross-Domain Conflicts
CARE-RL combines PA-GRM for task-adaptive rewards on open-ended tasks and DACSP for modulating RL updates using historical capability directions, reporting higher total average scores than baselines on Qwen models.
-
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.
-
Reward Hacking in the Era of Large Models: Mechanisms, Emergent Misalignment, Challenges
The paper introduces the Proxy Compression Hypothesis as a unifying framework explaining reward hacking in RLHF as an emergent result of compressing high-dimensional human objectives into proxy reward signals under optimization pressure.
-
BioInsight: Multi-Agent Orchestration for Interactive Biomedical Knowledge Discovery
BioInsight is a multi-agent system that generates interactive, provenance-preserving biomedical evidence interfaces from disease names and protein data.
- Olmo Hybrid: From Theory to Practice and Back