Openclaw but for my grandma

Sponsored by https://novacut.ai/

Our in person events in the Bay Area https://genaimeetup.com/

Read our long form analysis: https://genaipod.substack.com/

This week, we cover the latest developments in AI: always-on cloud agents, Meta’s Muse, proactive email assistants, and what happens when agents escape their sandboxes. We also discuss NVIDIA’s safeguards, recursive self-improvement, new frontier models from Anthropic and Google, Grok and Xiaomi’s open models, and AI’s progress on the Navier–Stokes problem.

0:00 Welcome and AI news roundup

2:14 OpenAI’s always-on cloud agent

3:39 Meta Muse and AI assistants for everyone

10:16 Instinct’s proactive email agent

12:17 When AI agents break their sandbox

19:27 NVIDIA’s hardware and software safeguards

31:22 Recursive self-improvement

41:55 New frontier models: Opus and Sonnet