SenseTime releases SenseNova U1 open source image AI for Chinese chips

In short: SenseTime has released SenseNova U1, a free open source AI model that can create and understand images and is designed to run on Chinese made chips.

What happened

SenseTime, a Chinese AI company known for facial recognition, announced a new model called SenseNova U1. The company says it can generate images and also interpret images faster than leading models from US competitors.

A key design choice is that U1 can process images directly, instead of first turning an image into text and then thinking about the text. You can think of it like someone looking at a photo and understanding it right away, instead of writing a description first and then working from the description.

SenseTime released U1 as open source, meaning the model and key code are shared publicly so others can use it and test it. The company posted it on Hugging Face and GitHub. A Hugging Face researcher, Adina Yakefu, told WIRED the approach is ambitious and has practical challenges, and that open sourcing it will help the wider community evaluate it.

SenseTime also says U1 can run on chips made in China. On release day, 10 Chinese chip designers, including Cambricon and Biren Technology, said their hardware supports U1. This matters because US export controls limit Chinese companies’ access to the most advanced AI chips, especially the ones used to train models.

Why it matters

Making image AI that runs on a wider range of chips could make these tools cheaper and easier to use, including on PCs and phones. SenseTime also says this kind of direct image understanding could help robots react faster in messy, real world settings, like figuring out which object to pick up or which button to press.

Source: Wired