Latest AI News

By complete coincidence, the day we released Neil Zeghidour (CEO of Gradium, the for profit spinoff of the vaunted Kyutai Moshi )’s talk on what remains to be built for realtime voice, Thinking Machines emerged for only the third time in a ~year (despite much drama) to drop Interaction Models: A Scalable Approach to Human-AI Collaboration ,…

The proximal cause of today’s op-ed is OpenAI’s deprecation of their finetuning APIs. For years, OpenAI stood out among the big labs for their finetuning support, and many many many talks and content pieces and AI engineers promoted how you can get some variant of “get o1 performance at 4o prices” and insisting that it was an important part of the toolkit.

Building Blocks for Foundation Model Training and Inference on AWS For a long time, "scaling" in foundation models mostly meant one thing: spend more compute on pre-training and capabilities rise. That intuition was supported by empirical work such as Kaplan et al.