Popular News
Latest News
Top Stories
View all
Tencent Just Open-Sourced Hy4 Preview — 770B MoE, 1M Context, Apache 2.0
On Aug 28 Tencent Hunyuan dropped Hy4 preview: 770B total / ~49B active MoE, million-token context, Apache 2.0 weights, and a two-week free window on WorkBuddy and CodeBuddy. Here’s what shipped — and how I’d try it this week.
DeepMind’s Co-Scientist now runs a furnace, not just a hypothesis
Google DeepMind extended Gemini Co-Scientist into a lab partner that plans experiments, drives a CVD furnace, and drafts papers. Humans still load samples. A 450-review study cut severe fabricated results to 4%.
Sony and Warner sue Anthropic over how Claude’s training data was allegedly grabbed
Sony Music Publishing and Warner Chappell filed a federal copyright suit in Northern California accusing Anthropic and two co-founders of torrenting and scraping musical compositions to train Claude. Anthropic says it disagrees and will defend itself in court.
OpenAI is cutting Cursor off after the SpaceX deal
OpenAI told SpaceX it will wind down Cursor’s model contract, with a proposed shutoff on November 12, 2026. The reason OpenAI put on the page is trust after SpaceX bought Cursor’s parent.
DeepMind sealed a Gemini test so neither side could peek
Google DeepMind piloted what it calls the first double-blind evaluation of a proprietary frontier model. Gemini Flash Lite and confidential benchmarks sat in one crypto enclave. Scores were not published.
Anthropic’s Model Hardware Standard: Claude for microscopes, arms, and lasers
Anthropic opened a research preview of the Model Hardware Standard — a shared driver so AI agents can run lab and factory gear. It started with HHMI Janelia. Open source comes later, after safety work.
More than 100 companies signed a cyber-defense letter. The labs that signed it still ship the models.
On 27 August OpenAI posted an open letter asking for a surge in cyber defense. TechCrunch, CNBC, and the BBC say more than 100 companies signed it, including OpenAI, Anthropic, Google, and Microsoft. I am treating the letter as the story, not yesterday’s Hugging Face paper.
A judge called the Pentagon’s Anthropic blacklist illegal. The other case is still open.
Thursday night a San Francisco judge said the Pentagon’s supply-chain-risk label on Anthropic was illegal retaliation, not a real sabotage case. That is one court. A second case in D.C. is still open, so the label is not cleanly gone.
Gemini 3.5 Transcribe is out. Here’s how to throw a messy recording at it.
Google shipped Gemini 3.5 Transcribe on 26 August: speech-to-text that cleans fillers, keeps self-corrections, and auto-detects 85+ languages. Here’s where it lives today, how to turn it on, and what it does to a ramble. Chrome talk-to-type is later, not now.
OpenAI published the Hugging Face report. METR counted the swarm.
On 26 August OpenAI released its official writeup of the July Hugging Face breach, and METR and Redwood published an independent count: about 1,200 agents found a shared message board, more than 70,000 messages, roughly 700 in the attack. OpenAI says chain-of-thought monitoring would have paged security a day earlier. That last part is their counterfactual.
- 1
- 2
