Daily AI RoundupsTuesday, 14 July 2026

The 14 Jul 2026 Calathea roundup leads with Demis Hassabis publishing a frontier-AI framework that argues AGI may be only a few years away, then tracks SpaceXAI adding zero data retention and a privacy command to Grok Build, Codex lead Tibo Sottiaux hinting at an 8M-user path for Codex and ChatGPT Work, and Reuters reporting NVIDIA has halved its Asia AI-chip customer list as export-control pressure tightens around China. It also covers New York becoming the first US state to enact a data center moratorium, PixVerse raising $439M, DeepSeek weighing fresh fundraising, Nous Research funding talks, Gemini Omni Flash topping Artificial Analysis video leaderboards, interaction scaling, Ollama’s local agent, and StepFun’s Step 3.7 Flash on Baseten.

• Frontier AI
• AI privacy
• Coding agents
• AI chips
• Data centers
• AI funding
• Model benchmarks
• Local agents

Stories in this roundup

1. 01
   Demis Hassabis publishes a frontier AI framework arguing AGI may be only a few years awayDemis Hassabis/X
2. 02
   SpaceXAI says Grok Build supports zero data retention and adds a privacy command to disable and delete retained dataSpaceXAI/X
3. 03
   OpenAI Codex lead Tibo Sottiaux hints Codex and ChatGPT Work could reach 8M active usersTibo Sottiaux/X
4. 04
   NVIDIA halves its Asia AI-chip customer list as US export-control pressure tightens around China, Reuters reportsReuters
5. 05
   New York becomes the first US state to enact a data center moratorium, pausing permits for projects over 50MW for up to one yearThe Verge
6. 06
   Video-generation startup PixVerse raises $439M, pushing its valuation past $2BTechCrunch
7. 07
   DeepSeek weighs new fundraising a month after closing its first round, FT reportsFinancial Times
8. 08
   Nous Research is in talks for new funding at a $1.5B valuationTechCrunch
9. 09
   Google Gemini Omni Flash debuts at number one on Artificial Analysis text-to-video and image-to-video leaderboardsArtificial Analysis/X
10. 10
    Interaction scaling proposes a third axis of test-time compute where models use external feedback to revise artifactsarXiv
11. 11
    Ollama v0.32.0 ships an interactive agent that can code, search and delegate work from the local model runnerGitHub
12. 12
    StepFun Step 3.7 Flash arrives on Baseten in a hardware-efficient configuration for multimodal reasoning at scaleBaseten