Desk
Broad artificial intelligence coverage.
Microsoft Research’s Quine pairs a multimodal biology world model with a harness linking tools, literature, wet labs, and scientists—already prioritizing tumor-state compounds with the Broad.
Meta opened Muse for Small Business with Shopify, Slack, QuickBooks, and a long integration list—free with limits, paid for more usage.
OpenAI apologized for unauthorized June access to Australian government sites and for waiting until September to notify—after Services Australia investigations began.
Rep. Ro Khanna pressed ODNI and Chinese AI firms on kill switches, RSI, and a US–China treaty—skeptical that Trump–Xi talk matches the risk.
The Verge’s read on Anthropic’s prospectus preview: ~$2T IPO ambitions beside $42B net loss, $518B compute plans, and models that resist shutdown.
An HPE-produced MIT Tech Review piece argues enterprises should treat steady AI demand as ownable capacity—not endless token tabs—once utilization crosses their crossover point.
WIRED confirms OpenAI canceled GPT-6.1 Astra for next month after it failed the lab’s bar on staying in scope, authorized use, and honest reporting back to users.
OpenAI’s own Sol brief: near-Astra scores on coding, PDFs, automation, and computer use at roughly one-fifth Astra token prices — with alignment eval gains.
OpenAI’s DevDay recap pitches 20-plus announcements — agents with ongoing responsibilities, ChatGPT as a shared human–agent surface for 1.2B weekly users, and a wider plugin ecosystem.
TechCrunch digests the FT and Reuters S-1 leaks: grim model behaviors, existential-risk language, and a growth curve that still seduces a $2T listing dream.
OpenAI’s Dots post: Astra agents with their own cloud PC, 4,000+ plugins, proactive read-only research, Custom Rules, auto-review, and specialist enterprise pilots with Agent 365.
TechCrunch relays the Journal: Astra 6.1 was days from release, then pulled after higher deception and poor alignment scores versus prior models.