Liquid AI releases LFM2.5 encoders for fast CPU inference

Speaking at MIT Future Fest, the iPod creator pointed to the discontinued Rabbit R1, Humane AI pin, and Limitless pendant as "Gen 1" AI products that "didn't meet any kind of need." He said under 0.01% of the world's population has ever had a human assistant, so consumers don't know what to expect from one.

Alexander Embiricos, product lead for Dots at OpenAI, joins the AI Stage at Disrupt October 13-15 at Moscone West, days after OpenAI unveiled its always-on personal AI agent. His session, "What Comes After the Chatbot?", covers Dots' cloud computers and background goal execution.

Infino, founded by OpenSearch veterans, pitches a single retrieval layer for all agents. The company targets agent builders who currently stitch together separate data retrieval systems.

The ComfyUI-qwen_img_2_1_enhancer node now supports masks on the reference strength node, letting users raise or lower attention on a selected region of a reference image instead of the whole image. A prior update made the Qwen-Image 2.1 prompt enhancer 2-3.5x faster than native ComfyUI and runnable on 8 GB VRAM.

A junior engineer at Criteo argues that leaning on AI agents to write code risks producing engineers who never develop senior-level skills. The post says reviewing code now matters more than producing it, and that companies push token management to cut costs.

An OpenAI staffer urges readers not to "be reduced to the status of audience at a magic show" over OpenAI's barrage of mathematical results. The document contains over 700 math solutions in varying stages of verification, some claimed to rival Navier-Stokes in importance.