AI Coding Live News

Standard agentic reinforcement learning assumes single-session trajectories with continuously accumulated context, clashing with production agent techniques like context compaction and subagent forking. Researchers introduce Mosaic (arXiv:2610.22174), a critic-free hierarchical optimization framework for Multi-Session Agentic RL. By incorporating session-usage signals at the trajectory level and shaping per-session advantages by marginal contribution, Mosaic outperforms trajectory-level broadcasting on BrowseComp across both compaction and subagent scenarios while using fewer sessions.

Zoom

To improve credit assignment in long-horizon agentic tasks, combining reinforcement learning with verifiable rewards (RLVR) and on-policy self-distillation (OPSD) has emerged as a leading paradigm. However, existing pipelines neglect the dynamic evolution of agentic skills and the reliability of privileged teacher signals. Researchers from CUHK and Tencent introduced RSIPO, enabling skills and policies to co-evolve recursively. RSIPO extracts skills directly from trajectory rollouts, optimizes skill generation via replay, and employs Empirical Bayes shrinkage to calibrate student trust in teacher logits. Operating on Qwen2.5-7B, RSIPO achieves 98.6% success on ALFWorld, outperforming standard GRPO by 17.4 percentage points.

Zoom

Reinforcement learning for language model agents is constrained by sparse episodic rewards and inefficient surface token alignment. Researchers have introduced PR-OPD (Privileged Representation On-policy Self-Distillation), discovering that injecting hindsight skills alters less than 25% of token probabilities while reconfiguring over 80% of internal hidden states. By aligning multi-layer projected representations against a stop-gradient self-teacher, PR-OPD boosts WebShop accuracy by 14.0 points and ALFWorld by 4.7 points with zero inference overhead.

Zoom

DeepSeek (@deepseek_ai) officially disclosed its proprietary DSec (DeepSeek Elastic Compute) large-scale sandbox infrastructure. Processing over 3 million isolated, lightweight sandbox sessions daily, DSec powers reinforcement learning (RL) training for DeepSeek-V4.1-Flash and next-gen coding agents with sub-second cold starts and strict isolation.

Zoom

Prime Intellect (@PrimeIntellect) GA’d Prime Sandboxes—hardware-virtualized MicroVMs for agentic RL. Full Linux VM fidelity (Docker Compose, background jobs, kernel workloads), elastic to tens of thousands concurrent, with verifiers/prime-rl/Prime Tunnels integrations and 365k+ prebuilt environments. Blog: ~30M sandboxes created internally; default 1,024 concurrent; tier-free pricing claimed ~3× cheaper than peers through Dec 22 (e.g. vCPU $0.02/hr).

Zoom