[AINews] Open Models, Model Labs vs Agent Labs, and the Untrainable★ 72
Latent Space6 hours agoCommentaryThis AINews issue uses Sarah Guo’s essay as a lens for current AI industry debates: where open models matter, how agent labs differ from model labs, and what cannot be trained away. It also recaps discourse around Anthropic Fable/Mythos, Fable 5’s capabilities, Google’s DiffusionGemma, and maturing agent infrastructure. The central takeaway is that durable value may lie in integration, customer translation, maintenance, and intent rather than model scores alone.
DiffusionGemma: Google Launches High-Speed Open-Weight Gemma Diffusion Model★ 76
Simon Willison's Weblog14 hours agoReleaseSimon Willison highlights Google’s new DiffusionGemma, an Apache 2 licensed open-weight Gemma model. He connects it to last year’s brief Gemini Diffusion preview, which he measured at 857 tokens per second. NVIDIA is currently hosting the model for free on its NIM cloud API, where Willison generated 2,409 tokens in 4.4 seconds, implying at least 500 tokens per second.