jitllm moves running model layers between GPUs mid-reply

L-Acoustics released software that isolates a singer's voice to cut through concert noise, used by Céline Dion and Harry Styles. Bloomberg reports the tool targets live-sound vocal clarity.

Raye's Royal Albert Hall show drew 1.8M pre-sale sign-ups for 5,000 seats, with scalpers using AI bots to create hundreds of thousands of fake registrations. UMG named Òscar Celma SVP of applied AI and machine learning; Character•AI filed for bankruptcy with $50M-$100M in estimated liabilities.

Qwen-Image 2.1 ships under the Qwen Research License, which restricts it to research or evaluation purposes only, so a user emailed the licensing address to ask about commercial terms. Alibaba's reply, not publicly documented, sets a free tier of up to 300,000 images per month.

Digital Music News feature argues the record labels' litigation strategy against AI music companies may not restore their power, since copyright law needs rebuilding for the AI era. It weighs whether the legal battle produces a viable business for musicians and labels.

The Chan Zuckerberg Initiative-backed nonprofit launched an open platform with open knowledge graphs and AI evaluators to check whether generated lessons align with state standards. More than 70 partners, including OpenAI and Anthropic, have joined.

PathSeer, a predicate-agnostic filtered ANN algorithm from Tsinghua University's SIGMOD 2026 paper by Zhiyue Li and Guangyan Zhang, ships in Weaviate 1.40. It replaces ACORN as the default HNSW filter strategy, which lost recall on sparse filters and cost more above 40% selectivity.