
IBM's Granite PatchTST-FM-r2 tops commercially licensed zero-shot forecasters on GIFT-Eval
IBM released Granite Time Series PatchTST-FM-r2, a ~385M-parameter forecasting foundation model dual-licensed under Apache 2.0 and OpenMDW that ranks second among replicable zero-shot models on GIFT-Eval and first among those with commercial-friendly licenses. The edge is licensing, not raw accuracy — it trails only models with more restrictive terms, and ships open weights plus a documented pretraining corpus that enterprise governance reviews require.
huggingface.co →- 02
Anthropic builds predictive monitoring to track activists who oppose AIAnthropic contracts Samdesk for real-time protest detection and runs a 'person-of-interest' process that flags escalation patterns to police, and a $180K-$230K intelligence role lists activism itself as a tracked threat category. It reported a Claude user to San Francisco police over an in-platform AR-15 remark, then refused to hand over the messages — a shift from reactive security to pre-crime monitoring at a lab that brands itself the responsible AI alternative.
prospect.org → - 03
Google commits €13B to Finnish AI data centers, its largest single investment in EuropeGoogle is investing at least €13 billion in Finnish AI infrastructure across 2027-2028 — its largest single investment in Europe — anchored by a 22-year power purchase agreement with Fortum that keeps the Loviisa nuclear plant, source of 10% of Finland's electricity, running past 2030. The structural shift is that hyperscaler demand is now the reason a national nuclear asset stays online, a template for how AI buildouts underwrite grid-scale clean power alongside 629 MW of new onshore wind and a 94 MW battery near Kajaani.
googlecloudpresscorner.com → - 04
OpenAI's 10,000-agent run builds a forced Navier–Stokes blowup, not the Clay solutionOpenAI ran roughly 10,000 agents in parallel for about 88 hours to construct a finite-time blowup for the Navier–Stokes equations, with the 165-page proof machine-checked in Lean in a further 17 hours. The construction adds a smooth external forcing term, so it does not meet the Clay formulation — Clay still lists the problem unsolved and OpenAI declines the $1M prize. The signal is the method, not the result: a large agent fleet producing a research-scale proof that verifies formally, a marker of how far AI has pushed into frontier mathematics.
openai.com → - 05
Qualcomm will design custom AI inference chips for AWS, its third data-center win since JuneQualcomm will design custom AI inference silicon for AWS across multiple chip generations, plus optical interconnects up to 1.6 terabits, aimed at cutting energy cost per token — and uses Amazon Bedrock to speed its own chip design in return. It is Qualcomm's third data-center win since June, after Meta's Dragonfly C1000 and Microsoft's HBC memory in Azure, giving weight to its $15B data-center revenue target for 2029. The chips sit alongside AWS's own Trainium, Graviton, and Nitro families, turning inference — where cost per token drives margins — into a multi-vendor race inside AWS itself.
the-decoder.com → - 06
Cognition raises $2B at a $48B valuation, doubling its price in four monthsCognition, maker of the Devin coding agent, raised $2 billion at a $48 billion valuation led by Andreessen Horowitz — double the $26 billion mark it set four months ago, on run-rate revenue that climbed from $492 million to $900 million. That a16z, which profited from Cursor's sale to SpaceX, is back leading a round in a Cursor rival signals investor conviction that AI coding stays a multi-player market, not winner-take-all.
techcrunch.com → - 07
NSA, CISA and FBI accuse Chinese AI firms of industrial-scale distillation of U.S. modelsA joint NSA, CISA and FBI advisory names DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI for industrial-scale distillation of U.S. frontier models since late 2024 — DeepSeek drawing on four Claude, two Gemini, five ChatGPT versions and Grok 4 to build R1 and R3, and Moonshot AI on 18 U.S. models to train Kimi K2 and K3. Calling distillation the core of Chinese AI industrial policy, not a copyright dispute, moves the fight from private lawsuits to national security, with the agencies pressing for a coordinated response across government, industry and allied nations.
cyberscoop.com →