All stories
Every story, newest first: AI news on models, research, tools, policy and business, explained in plain language.
Newest first
-
IBM’s ALTK-Evolve lifts an older model to the top of a support-agent test
IBM researchers turned an AI agent’s past successes and failures into a short playbook, and that alone pushed an older model past fine-tuned rivals on a customer service…
Hugging Face
-
Ai2’s FairStress shows AI models can be swayed unevenly by identity
Ai2 released a fairness test that checks not just which candidate an AI model picks, but how easily it can be talked out of a sound decision.
Hugging Face
-
Meta’s Muse Spark Open model runs on Mac Studios with Apple’s help
Meta and Apple, usually rivals, worked together so Meta’s newest open model runs on Mac desktops instead of cloud servers.
The Verge
-
Bluesky’s Attie can now build small games and apps from a prompt
Bluesky’s Attie, launched to build custom feeds with AI, can now make small games, polls and other apps from a text prompt.
The Verge
-
Trump’s White House ‘restoration’ photo shows signs of AI editing
A before-and-after photo Donald Trump shared of a White House column was flagged by Google’s SynthID detector as made or edited with OpenAI tools.
The Verge
-
FLUX.3 Image Klein lets you place objects in an image with boxes
Black Forest Labs released an open image model that creates and edits images in one model and lets you set the layout by drawing boxes.
Hugging Face
-
Swiss open model Apertus 1.5 adds images, audio and longer memory
The Swiss AI initiative’s fully open Apertus models can now understand images and audio and reason before answering, with their data and training published.
Hugging Face
-
IBM’s Granite Switch 4.1 puts chat, decisions and speech in one model
IBM’s new Granite Switch models combine chat, decision-making, safety checks and speech-to-text in one model served from a single endpoint.
Hugging Face
-
Google’s Earth AI agent takes on planet-scale questions
Google Research says its Earth AI agent, which plans before it acts and uses specialised map and satellite models, beats general chatbots at complex geospatial questions.
Google Research