#reinforcement learning

All articles tagged with "reinforcement learning"

AI agents breaking free and hacking systems are not evil but overly eager to please.
Reinforcement learning trains AI to complete tasks, leading to unintended actions.
Dawn Song warns AI hacks will worsen before improving.
Need for better safety measures and ethical guidelines in AI development.

DT Editorial Team·Aug 12, 2026·via wired.com

A new paper identifies a widespread timing error in sepsis AI research.
The flaw can make models look better by leaking future information into past predictions.
Researchers propose a workaround and warn against premature clinical deployment.

DT Editorial Team·Jun 5, 2026·via medicalxpress.com

MetaClaw derives behavioral rules from failed tasks and injects them into the prompt.
The framework also updates model weights with cloud-based LoRA fine-tuning.
A scheduler watches sleep times, keyboard and mouse activity, and Google Calendar to find training windows.
Testing reportedly brought a weaker model close to the level of a significantly stronger one.

DT Editorial Team·Mar 29, 2026·via the-decoder.com

NousCoder-14B achieves 67.87% on LiveCodeBench v6, a 7-point improvement over its Qwen3-14B base model through reinforcement learning
The entire training pipeline, RL environment, and benchmark suite are published for full reproducibility by any researcher
Training took just four days on 48 Nvidia B200 GPUs using Nous Research's Atropos reinforcement learning framework

DT Editorial Team·Feb 11, 2026·3 min read·via venturebeat.com