Agent Safety Gaps, Memory Learning, and Leaner Inference
Three new papers expose how production agent frameworks fail under attack, why RLVR training discards useful cross-episode signals, and how calibrated confidence cuts inference compute by 12x.
Articles Tagged "Reinforcement Learning"
Science Agents, Jailbreak Defense, and Open-World Failures
Three papers from today's arXiv: graph-native RL generates traceable scientific hypotheses, HARC defeats jailbreaks by coupling internal safety directions, and ICML 2026's OpenAgent shows how distributional shift breaks tool-use agents.
Tandem Training, World Models, and Efficient Agents
Three new arXiv papers on making RL reasoning legible across models, fixing broken world model latent states, and training small agents to beat their teachers.
SkyReels V4
SkyReels V4 is Skywork AI's unified multi-modal video model that jointly generates 1080p/32FPS video and synchronized audio from a single dual-stream diffusion transformer.
Sakana Fugu
Sakana AI's orchestrator model that dynamically coordinates Claude Opus 4.8, GPT-5.5, and Gemini 3.1 Pro to beat each of them individually on SWE-Bench Pro, GPQA-Diamond, and eight other benchmarks.
Emergent Alignment, Agent Memory, and Smarter Reasoning
Three arXiv papers: a conscience mechanism for ethical training, shared memory for agent populations, and selective verification that cuts test-time compute waste.
Qwen-RobotManip
Alibaba's generalist VLA model for robotic manipulation, built on Qwen3.5-4B with a DiT action decoder, trained on 38,100+ hours of open-source data, and ranked first on the RoboChallenge generalist track.
Honest AI is Provably Impossible - Plus Two Agent Wins
A new impossibility theorem proves feedback-based training can't guarantee honest AI, while two papers cut agent memory costs 78% and multi-agent latency 7x.
Reasoning Leaks, Hard Limits, and Self-Aware LLMs
Three new papers expose how reasoning traces can be extracted from supposedly hidden model internals, where chain-of-thought hits an architectural ceiling, and how RL teaches models to know when to quit.
Alignment Gaps, Agent Governance, and Greener LLMs
Three new papers expose a hidden flaw in DPO training, propose policy-as-code governance for enterprise agents, and cut LLM serving energy use by 26% via GPU power control.
Self-Correcting Models, Smarter Monitors, AI Designs Itself
Three new papers tackle critique dependency in LLMs, ensemble monitoring for AI control, and agents that autonomously discover better neural architectures.
SU-01
SU-01 is a 30B-A3B MoE reasoning model from Shanghai AI Lab that achieves gold-medal performance on IMO 2025, USAMO 2026, and IPhO 2024/2025 using a three-stage training recipe and test-time scaling.
