AI — Page 12

Jakub Pachocki wants frontier labs to set shared triggers for slowing development, three days after GPT-6 Astra and following an OpenAI agent containment failure.

The GA endpoint accepts images, video and audio; Fal's API allows 12 files, though the company ties its real-time performance claim to four references.

The controlled demonstration shows the model building an editable 3D scene and moving it into Unreal Engine 5, while artist Stefan Vaskevich warns about the pace of improvement.

Sylvain Kalache's warning accompanies Rootly's partnership with Uptime Labs, which uses simulated outages and AI role-play to train incident responders.

Ahead of a September 10th hearing, Reddit says Anthropic wants tighter controls on its lawyers, experts, technical records and electronic evidence.

Ivy prepares requests, Tulip batches them and ROSE runs models; Perplexity says the custom stack makes online and batch embeddings faster and cheaper, without disclosing production benchmarks.

Ling-3.0-flash-VL reads images and short videos, writes websites from screenshots, and operates interfaces through Ant Ling's hosted API.

The llama.cpp and vLLM updates target RTX 5090, RTX PRO 6000 and dual DGX Spark systems, with support through LM Studio and Ollama.

CEO Amjad Masad confirmed the rollout as Replit turns rapid access to frontier models into a core product advantage.

Meta's Max reasoning setting, which powered its strongest Muse Spark 1.3 benchmark results, is now available in Muse Code and the Meta Model API.

Sara Hooker and Sudip Roy's new tool turns a behavior description into synthetic training examples, then feeds them into AutoScientist.

The two AI companies will host Rio Shen, Max Ryabinin and Sarung Tripathi at a September 16th event on the production economics of open models.

New modules will arrive each month, while a September 19th virtual event will offer K-12 teachers live training and digital badges.

Mustafa Suleyman's lab paired Arena's No. 2 text-to-image model with a half-price Flash version for high-volume workloads.

The morning agent reads across Gmail, Calendar and Gemini chats, provided users enable Memory and connect their Google Workspace data.

The Copilot preview routes tasks through single-model, cascade, or critique workflows, trading extra inference for lower estimated costs.

Erik Arakelyan's team released a 9B-parameter base model, a 4.37M-document corpus and verified STEM data for Armenian developers.

Galit Desheh, a trauma and gender scholar, founded Ozz Health to give patients context clinicians often lack time to provide.

Li Wenhao's PlugClaw pairs a USB-C Android computer with cloud inference inside a hardware-protected environment that TrustKernel says keeps data hidden from both itself and the cloud provider.

Pengcheng Xu's preliminary study found that tool choice followed the task, while location-only LSP results increased follow-up reads and token use.

Figure targets initial deployment in Barstow, Texas, in the second half of 2027; Nscale will take an undisclosed stake and serve as Figure's preferred compute provider.

Matthew Prince's latest security bet uses GPT-5.6 Cyber to rank findings, propose patches and prepare WAF rules, with humans approving every change.

The five-year GPU contract would turn a former quant trader's energy startup into Jane Street's second multibillion-dollar AI cloud supplier.

The open-source beta routes parallel agent requests across Windows, Linux and macOS devices without pooling their GPU memory.