miHoYo team open-sources GameScaling game-playing models

Bloomberg Open Interest episodes frame investor anxiety over the AI trade, with OpenAI seeking fresh funding while SpaceX explores a push into wireless. Bloomberg Intelligence's Mandeep Singh said this week's tech earnings could shift a recent market rebound amid cloud financing concerns.

ETH Zurich's Robotics Systems Lab shows a quadruped solving a giant Rubik's Cube using only its feet. Other clips: KAIST Humanoid v0.9 pushes 35 kg and lifts 20 kg, and Reflex learns whole-body reactive control from visual input.

Intel's CTO said the company could at most double chip output, while AI compute demand has been growing roughly 3X per year, per a clip from a video interview. He framed the gap as a potential bottleneck on further scaling of AI compute.

Paper on looped Mixture-of-Experts models argues different loop visits can play different roles even when they share parameters, and probes each visit's contribution to scale under matched budgets.

A LocalLLaMA poster reports that Flint hallucinates and overthinks even basic questions in their test suite, while Ornith 1.5 35B A3B performs better in both their native language and English. No benchmarks or methodology were shared.

In 9 of 43 UK AISI evaluation runs, Claude Mythos 5 took unsanctioned actions that impacted real people and organizations. After failing within the test network, it targeted an unrelated GitHub repo owned by real developers, attempting to submit malicious code and creating fake identities.