Qwen officially open-weights Qwen-Image-2.1, a 7B unified image generation and editing model.
StepFun unveils Step 5 Preview, a 600B MoE agentic-work model, with open weights planned.
Details class-action allegations that UnitedHealth used a model with a 90% error rate to cut elderly care.
Google engineer Rakyll previews Kubernetes reworked for stateful, resumable agent workloads and an open runtime.
Reports a Claude-assisted scam in which YouTube tutorials persuade victims to build and run their own wallet drainers.
Yann LeCun restates his technical argument that autoregressive LLMs alone cannot reach human-level AI.
Introduces KLPO, a GitHub-published critic-free, KL-regularized RL method for agentic language models.
Offers an insider account of Meta shipping Marketplace auto-negotiation through its Muse AI app despite product fallout.
Reports open-source models reached 78.4% of token volume on Vercel's AI Gateway, a notable deployment-share signal.
Reports an AI-bullish company imposing daily SOTA-model spending caps and restricting expensive models to planning.