Desk
Generative models for text, image, audio, and video.
Anthropic is offering open-source maintainers free scans from its strongest models, including Mythos, with no human review. Faster warnings help, but the burden of sorting truth from noise lands on volunteers.
Google's AI Edge Foresight takes meeting notes without sending your meetings anywhere. On-device is the privacy story note-takers should have led with all along.
ChatGPT's new Intelligent UI swaps walls of text for interactive charts and tools, a genuine usability upgrade that also hands more of how information looks to the model itself.
NVIDIA wants telescope and detector images to stay on the GPU from first read to final human review, and its own fine print on the speedups deserves as much attention as the headline numbers.
Google just handed everyone a free AI-media checker, which is a real public good, as long as nobody treats a blank result as proof something is human-made.
Reka's Rho-1 folds text, images, video and robot control into one 19-billion-parameter model, an ambitious bet against agent pipelines that's still only a research preview.
ChatGPT's ads are moving from text into pictures, placed next to the images people ask it to make. The business logic is clear. The trust test is just starting.
ChatGPT’s ads are graduating from a logo and a link to sponsored images beside your generated pictures — a modest test that tells us where free AI is headed.
Yandex collapsed a music recommender’s whole cascade — 15-plus generators, a pre-ranker and a ranker — into one transformer, and live users listened more. The engineering is impressive; the engagement lever is the story.
Decision models return choices and calibrated probabilities, not paragraphs — TypeSafe’s Jev set the category, and Fastino plus open-source forks arrived within weeks.
Ai2 open-sources AstaBrief 8B — a one-pass scientific report model that cuts Asta Fast mode to ~51 seconds — betting data filters beat RL theater for grounded citations.
Astra Ultrafast on Blackwell promises up to 8x token speed for agent loops — OpenAI using its own models to tune inference is the quiet industrial story.