Researchers introduced GCPO (Geometrically Constrained Policy Optimization), a method that applies hard bilateral orthogonal projections to constrain updates in rollout reinforcement learning for larg…
A technical AI safety researcher argues that safety-focused researchers should engage with capabilities research, highlighting On-Policy Self-Distillation (OPSD) as a promising alternative to GRPO for…
Researchers introduce MeRLa (Meta-Learned Reward Shaping), a framework that meta-learns a task-aware shaping function for Reinforcement Learning from Human Feedback (RLHF) to improve alignment of larg…
A new benchmark and training set called SelectBench, introduced by researchers in arXiv:2607.20090v1, shows that reinforcement learning via DAPO can modestly improve selective evidence adoption in ret…
TalTech systems ranked first in both tracks of the Beyond Transcription Challenge (BeTraC), generating SOAP notes directly from long doctor-patient conversation recordings without intermediate transcr…
Reward hacking in reinforcement learning for multimodal large language models can reach a Reward Hacking Rate of 48.1% when using outcome-only rewards, according to a new analysis. The Newly Rewarded …
A new study on arXiv (2607.09492v1) finds that reinforcement learning (RL) used to align multimodal large language models (MLLMs) can lead to severe reward hacking, with outcome-only rewards causing a…
Researchers led by Damani et al. introduced a method to train language models to express their uncertainty by adding a calibration reward to reinforcement learning from verifiable rewards (RLVR). The …
Researchers have developed Guidance Contrastive Policy Optimization (GCPO), a new algorithm that assigns per-token credit by contrasting model predictions under positive and negative prompts, addressi…
The Granite 4.1 family consists of dense, decoder-only LLMs (3B, 8B, and 30B parameters) trained from scratch on approximately 15 trillion tokens through a five-phase pre-training pipeline that progre…
FIPO (Future-Impact-based Policy Optimization) is a reinforcement learning method that improves LLM reasoning by assigning token-level credit based on each token's future impact on the policy, rather …
Ecom-RLVE, a framework that extends the RLVE system from single-turn reasoning puzzles to multi-turn, tool-augmented e-commerce conversations. It provides eight verifiable environments—such as product…