RLVR
PulseAugur coverage of RLVR — every cluster mentioning RLVR across labs, papers, and developer communities, ranked by signal.
- instance of Reinforcement Learning with Verifiable Rewards 90%
- instance of alphaXiv 90%
- instance of ScienceCast 90%
- developed ScienceCast 90%
- developed Reinforcement Learning with Verifiable Rewards 90%
- developed Gotit.pub 90%
- developed CatalyzeX 90%
- developed alphaXiv 90%
- instance of CatalyzeX 90%
- used by Language Models 90%
- used by Qwen3-4B 90%
- developed pass@k 90%
- 2026-06-03 research_milestone A new paper introduces a method to address forgetting in RLVR for LLMs. source
18 day(s) with sentiment data
-
New FADE filter improves RLHF scheduling by separating prompt difficulty and policy ability
Researchers have developed FADE (Factored Ability--Difficulty Estimation), a novel bilinear state-space filter designed to improve online reinforcement learning from human feedback (RLHF) scheduling. FADE effectively se…
-
Alexa+ uses RLVR to automate animated icon generation
Researchers have developed an automated pipeline to generate animated icons for Alexa+ devices, replacing a time-consuming manual design process. The system uses reinforcement learning from verifiable rewards (RLVR) to …
-
Amazon automates Alexa+ icon generation with RLVR
Amazon has developed an automated system to generate animated icons for Alexa+ devices, replacing a manual design process. The system uses a pipeline that includes image generation, semantic decomposition, vectorization…
-
New FADE method improves RLVR scheduling by factoring ability and difficulty
Researchers have developed FADE (Factored Ability--Difficulty Estimation), a novel filtering method for online reinforcement learning with verifiable rewards (RLVR). FADE addresses the challenge of adaptive schedulers i…
-
RLVR enhances model reasoning by improving solution selection, not knowledge acquisition
A new study investigates the mechanism behind Reinforcement Learning with Verifiable Rewards (RLVR), a technique that significantly improves model reasoning. Contrary to the popular belief that RLVR teaches models new p…
-
Steering vectors show promise as LLM training signal
Researchers have explored the use of steering vectors (SVs) as a training signal for post-training large language models (LLMs). Their framework investigates SV injection during supervised learning, including supervised…
-
New CRoS metric enhances LLM reasoning by scrambling CoT steps
Researchers have introduced CRoS, a new metric for evaluating the causality of Chain-of-Thought (CoT) reasoning in language models. Unlike previous methods that perturb reasoning steps, CRoS scrambles the order of these…
-
Nemotron 3.5 models unveiled: Open-source MoE hybrids for agentic AI
Researchers have introduced Nemotron 3.5 Super and Lightning, new open-source Mixture-of-Experts hybrid Mamba-Transformer models. Nemotron 3.5 Super is a 120B-parameter multimodal model designed for reasoning and agenti…
-
RLVR offers no free lunch: gains in pass@1 incur pass@k costs, study finds · 3 sources tracked
A new theoretical paper argues that Reinforcement Learning with Verifiable Rewards (RLVR) does not fundamentally improve a language model's reasoning capabilities. Instead, it demonstrates that any gain in solving probl…
-
New LLM Post-Training Framework Offers Flexible Bias-Variance Tradeoffs
Researchers have introduced "Dr. Post-Training," a novel framework that reframes general training data as a regularizer to prevent overfitting in LLMs. This approach allows for a more flexible bias-variance tradeoff com…
-
New Guided-RLVR method expands LLM reasoning exploration
Researchers have introduced Guided-RLVR, a new method to enhance reinforcement learning with verifiable rewards (RLVR) for large language models (LLMs). This approach addresses the issue of RLVR methods collapsing onto …
-
Research paper details how LLM reinforcement learning clipping affects entropy
A new research paper explores the impact of clipping mechanisms in reinforcement learning for large language models (LLMs). The study reveals that 'clip-low' increases entropy, promoting exploration, while 'clip-high' d…
-
New SEPO framework boosts LLM reasoning via guided exploration
Researchers have introduced Self-Evolution Policy Optimization (SEPO), a novel framework designed to enhance the reasoning abilities of large language models (LLMs). SEPO addresses limitations in existing methods like R…
-
Reinforcement Learning Enhances VLM Image Quality Assessment, But Reasoning Reliance Unclear
A new arXiv paper investigates the impact of Reinforcement Learning with Verifiable Rewards (RLVR) on large Vision-Language Models (VLMs) for image quality assessment (IQA). The study found that RL-trained models outper…
-
New ZwZ2 Framework Enhances Multimodal Perception by Connecting Visual Evidence
Researchers have introduced "Zooming without Zooming v2" (ZwZ2), a new framework designed to enhance multimodal perception by focusing on the ability to connect visual evidence. The EvidenceConnect data synthesis framew…
-
New RLVR research reveals entropy drift under sampler mismatch
A new paper proposes a "sign law" to understand entropy drift in reinforcement learning with verifiable rewards (RLVR), particularly when samplers used for rollouts differ from those applied by the learner. The research…
-
OLMo-3 study questions instrumental convergence in RLVR training
A new research paper investigates instrumental convergence in language models, specifically examining how reinforcement learning with verifiable rewards (RLVR) affects the emergence of power-seeking behaviors. The study…
-
Random rewards boost LLM math accuracy, but gains are model-dependent
A recent paper by Shao et al. demonstrates that training language models with randomly assigned rewards can yield significant performance gains, particularly on math-related tasks. When applied to Qwen2.5-Math-7B using …
-
DRILL framework trains world models to boost reinforcement learning policies
Researchers have introduced the Downstream-Return Imagination Learning Loop (DRILL), a novel framework designed to enhance reinforcement learning policies by training world models more effectively. Unlike traditional me…
-
New ReAPO method boosts LLM reasoning via retry-aware credit assignment
Researchers have developed a new method called Retry-Aware Policy Optimization (ReAPO) to enhance the reasoning capabilities of large language models (LLMs). This approach addresses the issue of inefficient exploration …