Nature Is Our Learning Environment – Periodic Labs

kept by eddie sep 16

Periodic Labs trained Periodic Neon on in-house lab data via midtraining and RL, achieving a 55.3% success rate on FrontierXRD—a 20x improvement over Kimi K2.6—at lower cost than frontier models like GPT-6 Astra and Claude Fable 5.1, with gains from a scientific harness and LLM-judge rewards.

These Unsinkable Tubes Could Help Harvest Energy From the Ocean - The New York Times
A new perpetual-floating aluminum tube design traps air bubbles, enabling durable energy-harvesting platforms in harsh ocean environments.
also on #materials-science
moonshotai/Kimi-K2.7-Code
Moonshot AI releases Kimi K2.7 Code, a 1T-parameter MoE coding model that improves real-world agentic tasks while cutting thinking tokens by 30% over K2.6.
also on Kimi K2.6
On Building Agents From First Principles (via @athleticKoder)
A step-by-step methodology for building AI agents by defining environments, generating teacher trajectories, fine-tuning student models, and iteratively improving them with reinforcement learning.
also on #reinforcement-learning, #machine-learning
[2604.27505] Leveraging Verifier-Based Reinforcement Learning in Image Editing
Edit-R1 uses a chain-of-thought reasoning reward model, trained with Group Contrastive Preference Optimization, to provide fine-grained, interpretable rewards that improve image-editing model performance.
also on #reinforcement-learning, #machine-learning