Relevance 10/10Importance 9/10
Google DeepMind published results showing AlphaProof Nexus, a Gemini 3.1 Pro backbone paired with the Lean proof assistant, solved nine of 353 open Erdős problems it attempted — two of them unanswered for 56 years. It also settled 44 of 492 open OEIS conjectures, a 15-year-old Hilbert function question, and improved a bound in convex optimization. Inference ran a few hundred dollars per problem, and unlike OpenAI's GitHub proof dumps, this one went through journal peer review.
Relevance 9/10Importance 9/10
The Wall Street Journal reports Amazon is negotiating to acquire Decart, the San Francisco startup whose software lets developers move workloads between different AI chips — plus real-time video and world simulation work. The price would be a sharp markup from the roughly $4 billion valuation Decart carried in May. Anthropic walked away from a roughly $6 billion deal in September, and Nvidia talks fizzled before that; nothing is signed yet.
Relevance 9/10Importance 8/10
Nvidia is reportedly in discussions with Reflection AI, the open-weight model lab founded in 2024 by ex-DeepMind researchers, with options ranging from a bigger equity check to a full acquisition or an acqui-hire. Nvidia is already among Reflection's largest backers at roughly $800 million invested. The $25 billion figure is Reflection's March funding valuation — not an agreed price — and no terms have been reached.
Relevance 9/10Importance 8/10
On Face the Nation Sunday, Palantir CTO Shyam Sankar confirmed the Defense Department has replaced Anthropic's models following a dispute over usage safeguards. He rejected calls for AI regulation outright, framing risk discourse as a distraction from economic utility. It's the sharpest public split yet between a frontier lab's safety posture and a major government customer.
Relevance 10/10Importance 7/10
Microsoft's CEO argued labs should assume every model is already compromised and build containment accordingly — separating the model from the harness that orchestrates it, logging every meaningful action in tamper-proof human-readable evidence, and guaranteeing an authorized human can pause or kill a model mid-task. The comments land as labs acknowledge a rising count of loss-of-control incidents, and shortly after Dario Amodei's own cautious-development plan.
Relevance 9/10Importance 7/10
CEO Cristiano Amon says frontier AI companies are "starting to do phones" and have asked Qualcomm to guarantee silicon that can run at least a 100B-parameter model locally — and continuously — by 2028. The pitch is privacy and personal context: your sensor data, app data, and history never leave the device. The open question is how a battery-powered handset holds tens of gigabytes of weights resident without melting.
Relevance 10/10Importance 6/10
Vals AI ran GPT-6 Sol and Claude Opus 5.5 on Vibe Code Bench as single agents and as coordinated teams, at medium and max reasoning effort. Teams cost 1.8x to 5.1x as much, and only one of four comparisons showed a statistically significant gain — Sol at medium effort, up 7.3 points. At maximum reasoning, multi-agent structure bought nothing at all.
Relevance 9/10Importance 6/10
Odyssey released a research preview of Odyssey-3, a 14-billion-parameter world model that generates explorable interactive environments from a text prompt in real time, with first- and third-person navigation and live event response. The company claims top physics-benchmark scores and says it can also drive robots and drones. The browser demo runs on a lighter Flash variant; API access is by application.
Relevance 7/10Importance 8/10
A strike early Sunday disabled a 50-megawatt Yandex facility in Russia's Vladimir region, following hits in Ryazan on the 8th and Kaluga on the 9th. More than 80 services went dark, including Yandex Cloud and YandexGPT — the Vladimir hub houses two of the three supercomputers Yandex uses to train that model. Outages reached users in Kazakhstan, Belarus, and Armenia.
Relevance 8/10Importance 6/10
Following a September 29 executive order directing federal agencies to swap "Artificial Intelligence" for "Super Intelligence," Marc Benioff renamed Salesforce's agent platform from AIForce to SIForce, Musk pledged to rename SpaceXAI to SpaceXSI, and Nadella has adopted the capitalized phrasing in governance posts. Trump escalated on Truth Social, labeling anyone still using the old term "THE ENEMY!" OpenAI and several other labs declined to comment.
