Robot Watches One Demo, Hits 59% Success Rate
With no fine-tuning, Generalist AI's GEN-1.5 robot learns a new task from one video demo in its context, hitting 59% success versus 83% after training.
With no fine-tuning, Generalist AI's GEN-1.5 robot learns a new task from one video demo in its context, hitting 59% success versus 83% after training.
Pew Research sampled 490,000 English web pages across 49 Common Crawl snapshots and found AI-writing signals in 35% of pages posted after ChatGPT launched.
SGLang and Ant Ling Infra cut single-request TPOT for Ling-3.0-flash from 3.33ms to 0.78ms on 4 Blackwell GPUs by removing host-side stalls.
OpenAI is cutting GPT-5.6 Sol's API pricing by over 20% for three months, with output tokens down a third while subscription prices stay unchanged.
Ant Group and SGLang keep quantized weights resident in GPU memory, cutting Ling-2.6-1T restart time from 8.8 minutes to about half a minute.
Anthropic folds Claude Mythos 5 into partner security products, funds open-source security with $35M in credits, and widens its Cyber Verification Program.
Anthropic publishes a six-stage AI-native SDLC playbook that chains intent.md, spec.md and plan.md through Skills, Hooks, CLAUDE.md and evals.
OpenBMB open-sourced MathForm-8B, an 8B model for autoformalizing math into Lean 4, plus a 367,000-sample verified dataset and evaluation code, outperforming several 32B formalization models.
A Hume AI probe of 11 leading speech recognition models finds the top scorers on public benchmarks are also the most likely to echo wrong reference transcripts instead of what they actually hear.
A security researcher found ClarityCheck's unsecured S3 bucket with 9 million face photos and personal data downloadable by anyone with the link.