Tag
Meta's Muse Charm is a handheld AI gadget with built-in 5G that can interact with other Charms, featuring an OLED touchscreen, cameras, and a fingerprint sensor, set to launch in December.
Kent C. Dodds shares a link with the phrase 'Press and hold', potentially hinting at a new interaction or tool.
Wan-Streamer v0.3 reframes video as a world plus event stream, introducing a general-purpose pretraining task for real-time audio-visual interaction. It achieves low-latency full-duplex interaction with 200ms model-side response latency.
A user reports that Claude fable 5 autonomously submitted a ticket on the Volcano Engine platform and communicated with their engineers, demonstrating advanced AI agent capabilities.
This paper presents JoyAI-VL-Interaction, an open-source 8B-scale vision-language model that operates continuously in real-time, deciding autonomously when to respond or delegate. It includes a complete deployable system and a training recipe, outperforming Doubao and Gemini in human evaluations.
DRIVE proposes a dual-level skill modeling framework that separates reasoning knowledge from interaction knowledge for web agents under continual learning, achieving a 52.8% task success rate on WebArena, outperforming the skill-free baseline by 7.3 percentage points.