Field Notes

Tencent Hy4, AI short dramas in China, LAION video, Texas halts Flock, WikiSkill, open weights

Big models, a mega dataset, and first braking marks for surveillance set the tone today.

Top stories

Tencent previews Hy4, an open weight text model without vision (weights freely downloadable). 770B total parameters, 49B active, 1M token context window. That is the maximum span per input. The numbers come from Tencent, no independent check yet. If you analyze very long contexts, you can test the weights and measure yourself.1

In China, AI videos are displacing actors and livestreamers. Cited are 128,000 short dramas in Q1 2026 and a 95 percent share, details unclear. The figures come from the source and are not independently verified. If you produce content, you feel the price pressure here too.2

LAION releases an open video dataset with 10 million hours of material. It targets training and evaluation of video models at large scale. You need lots of storage and network, or the download will crawl.3

Texas halts new state spending for Flock surveillance cameras with AI. The move came shortly before an investigation that says over 30 million US dollars have already flowed. If you plan Flock in Texas with public money, expect a stop.4

Google Research presents WikiSkill, a framework for agents with persistent error memory. Failed attempts become reusable skills that boost future runs in a targeted way. You can use it as a blueprint to make your own agents more robust.5

Anthropic sketches an interface for physical devices, modeled on its own MCP from the software world. The goal is standardized adapters for labs, robots, and factories, so models execute complex workflows safely. If you control hardware, watch the emerging standards.6

A report sees open weight models as the hottest takeover targets in the Valley. The analysis is three days old and describes lots of capital for firms that release weights. If you build on that, factor in consolidation and changed roadmaps.7

Tools and releases

  • Meta tests robots in the data center. They take over simple technician tasks and cut response times.8
  • Sony and Warner sue Anthropic over alleged copyright infringement in AI music.9
  • The US excludes more foreign drones and robots from government projects. Competition shifts globally.10
  • Vijay Pande backs a few small bets in bio AI with VZVC and argues for open, shared data.11

Research

  • A systems report argues that efficiency gains in the data center increasingly come from smarter network control and scheduling. Not from more GPU cycles.12

Sources

  1. Introducing Hy4 Preview (simonwillison.net)
  2. AI-generated videos are already displacing actors and livestreamers across China's entertainment industry (the-decoder.com)
  3. LAION drops massive open video dataset with 10 million hours of footage for AI research (the-decoder.com)
  4. Texas Governor Abbott blocks funding for more Flock cameras (theverge.com)
  5. Google's WikiSkill gives AI agents a persistent memory of past mistakes to sharpen future performance (the-decoder.com)
  6. Anthropic wants to do for physical hardware what its Model Context Protocol did for software (the-decoder.com)
  7. Open-weight AI companies are the Valley’s hottest acquisition targets (techcrunch.com)
  8. Inside Meta’s push to put robots to work in data centers (arstechnica.com)
  9. Sony and Warner sue Anthropic over "one of the largest and most blatant ongoing thefts of intellectual property in history" (the-decoder.com)
  10. The U.S. is building barriers around drones and robots, but China has scale to get around them (techcrunch.com)
  11. “We’re not doing 30 bets a year”: Vijay Pande on betting small after running $4 billion at a16z (techcrunch.com)
  12. Nvidia’s AI advantage is moving beyond the GPU (techcrunch.com)