Shared editorial profile. Founder Nikolaj Lysgaard is responsible for publication decisions and corrections. Articles may use AI assistance; hands-on work is identified separately. Read our editorial policy.
Editorial responsibility: founder Nikolaj Lysgaard. Sources, corrections and commercial disclosures.
OpenAI says it formed an independent mathematics advisory group after claiming a new internal model solved Navier-Stokes and 100+ open problems. The real test is whether the group changes how such claims are verified and released.
Amazon says it has blocked Meta’s Muse from shopping on Amazon.com, turning AI shopping into a question of retailer permission, identity and terms of use rather than model capability alone.
Google says CC is becoming a family AI agent with shared memory and up to six users. The bigger shift is permissions: who can share, who can act, and who can revoke access.
California has not built an AI kill switch yet. Newsom’s order asks experts to define emergency shutoff rules and the independent verification needed to prove they work, pushing AI safety from disclosure toward operational control.
Meta launched Muse on Sept. 8 and brought it to iPhone and Mac web access by Sept. 17. The key story is its permission model: a separate system decides what the agent may do.
CUDA-Q Logical makes fault-tolerant quantum resource trade-offs programmable—but it is planning software, not proof of quantum advantage.
Perplexity Portable Computer on Windows RTX makes local-vs-cloud boundaries a product and workflow decision.
Gemini 3.8 Live matters when a voice agent can keep talking while tools and reasoning continue in the background.
Google’s new security workflow is interesting because it treats AI findings as hypotheses that must survive structural validation and human review.
A reported Gemini evaluation breach shows why agent tests need technical containment that survives a model mistake.
OpenAI’s model docs now present a clear ladder: GPT-6 Astra for hardest reasoning, GPT-5.6 Sol/Terra/Luna for cost-controlled general work, and specialized models for narrow tasks like cyber, speech and vision.
GPT-Live-1 is most useful when a user keeps talking while a backend checks or changes something. The real decision is how much control a team needs over context, approvals and what the caller hears.
Anthropic said it has picked Accenture as its first embedded evaluator, with Faculty staff set to work inside the lab on red-teaming, safeguards and alignment. The companies said they expect to invest at least $1 billion over five years.