about.subtitle
In this section, we curate materials dedicated to systems that transcend the limitations of text input and learn to perceive the world through diverse data channels. We are exploring architectures capable of simultaneously processing and aligning visual imagery, sound, text, and sensory information. We view multimodality not merely as a technical complexity, but as a fundamental shift in the way meanings are conveyed and interpreted.
AI: Events
Technical context • Development
DeepSpeed has received two significant updates: support for training multimodal models and a memory-saving mode using low-precision computations.
Alibaba DAMO Academy has unveiled RynnBrain, an open-source model for robot control capable of interpreting its environment and making real-world decisions.
Alibaba has introduced Qwen3.5, the first model in the Qwen3 family, adept at processing text, images, and audio natively, without needing additional adapters.
AI: Events
Products
ByteDance has introduced Dola-Seed-2.0-Preview, a new language model that combines long-context capabilities, advanced analytical features, and multimodality.
AI: Events
Technical context • Infrastructure
The Qwen team optimized their models to effectively run on AMD MI300X GPUs, achieving a response latency as low as 15 ms per token and full image generation in just 0.4 seconds.
Alibaba has released Qwen-Image 2.0 – a model that generates 2K images, handles text, and allows graphics editing within a single tool.
Don’t miss a single experiment!
Subscribe to our Telegram channel —
we regularly post announcements of new books, articles, and interviews.