AI24 Sept 2026
Adobe buys Topaz Labs: the licence question
Adobe closed its $340M Topaz Labs purchase on 23 September. Topaz keeps its own brand, and nothing was said about what the standalone apps will cost.
AI
Generative tools where they touch a working pipeline — what they take off an artist’s plate, what they quietly add to it, and which claims survive contact with a deadline.
14 articles · latest 24 Sept 2026
AI21 Sept 2026
CESA survey: 85.8% use AI, admin the top task
Japan's industry body puts generative AI use at 85.8%, up from 51%. The task breakdown comes from a far smaller sample, and puts visuals near the bottom.
AI14 Sept 2026
Level-5's AI apology and the five-to-two plan
Akihiro Hino apologised for AI in Level-5's Vision 2026 II broadcast, then restated the plan: turn artwork into polygons, five-year projects into two.
AI8 Sept 2026
Space Marine 3: no generative AI, by licence
Saber's CCO says there is absolutely no genAI in Space Marine 3 or Jurassic Park. Saber's own small game used it. The licence holder is the dividing line.
AI6 Sept 2026
Humankind 2 trailer: 150 crew, still called AI
Amplitude shot its Gamescom trailer at Nu Boyana with 25 actors, 150 crew and a pig. The internet called it AI slop anyway, and the reasons are worth reading.
AI3 Sept 2026
Hugging Face's WebGPU kernels: matmul gained 14%
Hugging Face put 207 WGSL kernels on the Hub on 1 September. The geometric mean is 2.57x. Matmul, which dominates inference time, came in at 1.14x.
AI3 Sept 2026
Nvidia buys Hugging Face: $12.9B for the models
Nvidia confirmed the $12.93 billion deal on 3 September. The models your local tools fetch on install now come from a repository the chip vendor owns.
AI26 Aug 2026
OpenAI's own chip: up to 3.6x ahead on latency
Jalapeño's first silicon results landed on 25 August — three open-weight models, two Nvidia rack generations, and a benchmark platform that isn't MLPerf.
AI25 Aug 2026
HyperNova 60B: 61GB to 32GB with quantization healing
Multiverse took HyperNova 60B from 61GB to 32GB of VRAM with Quantization-Aware Healing, and the smaller model holds its ground against the original.
AI23 Aug 2026
LinkedIn's AI slop button can't tell a render from AI
A "Seems like AI slop" button has been clicked over a million times since July, and LinkedIn says it's already suppressing reach by 40%.
AI20 Aug 2026
OpenAI keeps Zero Data Retention on frontier models
OpenAI says its no-retention promise will survive the next generation of frontier models, via a new architecture it calls Private Safety Processing.
AI14 Aug 2026
Gemini 3.7 Flash halves the price of looking at everything
A 1M-token, image-and-video-reading model at half its old price, three weeks after the last one.
AI14 Aug 2026
GPT-5.6's 14x speed-up matters for agents, not chat
Ultrafast mode runs GPT-5.6 Sol at up to 750 tokens a second on Cerebras silicon, which matters most for a model that keeps pace while you work.
AI28 Jul 2026
Where AI actually fits in a 3D pipeline
Not concept-to-final-render. The useful places are smaller, less glamorous, and they compound.