World News, Politics and War, Reported from Boston
AI news from The Boston Report: OpenAI, Google, Anthropic, Meta and new AI tools, plus the safety, policy and business questions around artificial intelligence.
Nvidia posted four 16B-parameter Cosmos 3 Nano open-weight checkpoints on Nov. 5 for multi-view driving simulation and bimanual robot video prediction under OpenMDW 1.1.
Gemini Omni 1.2 Flash is generally available with native clips up to 15 seconds and a 30% lower 720p list price of $0.07 per second, Google said Nov. 5.
Google health AI evaluation combines benchmarks, automated graders and expert review. Its new framework is a testing method, not proof of clinical benefit.
MIT robot memory research checks doubtful object labels and improves test accuracy. A Cambridge team explains the gains, the method and its limits in use.
A Claude research usage study of 207,049 sessions finds analysis ahead of idea generation, with clear limits on what it proves about scientific progress.
GPT-6 Cyber arrives with restricted security access and a higher OpenAI-reported ExploitGym score. What the tests show, and where safeguards still matter.
Meta acquires Conductor, an AI coding multi-agent startup, on undisclosed terms; cloud shuts Nov. 20 while the local app keeps running into early 2027.
Moonshot AI sought Nvidia Blackwell access via European data centers amid U.S. export curbs, CNBC sources say, as remote cloud leases test chip rules.
OpenAI data center policy: no NDAs with officials, full power-cost pledges and community input after voters rejected most AI data-center ballot measures.
Boston’s Lazarus AI raised $43 million to expand enterprise AI that runs on customer hardware without per-token charges, backed by QBE, State Farm Ventures and others.