This is pretty mind blowing to me, as I did not expect this would be achievable for many years:
It is mostly exciting because it uses a language model to understand and communicate with others.
If we take a step back, that’s the kind of technology that might be used to have an interactive experience in Blender where the artist describes what they want and some sort of agent might create some geometry node setup iteratively or ask clarifying questions.
Meanwhile, it looks like a future thread for this forum will be titled AI vs. Employees (as the insane advancement in this type of thing collides with robotics). On one hand we may finally get our own Rosy, on the other…
Was just watching a YT video by part_time_Larry that said that as Microsoft own GitHub and they are a supporter of OpenAI they are using all the content on GitHub to train AI to program in any language. Probably where your example comes from
Everyone can read the public repositories on GitHub. That’s how anyone can scrape the data for machine learning. There is no need to own GitHub for this. Everyone can legally get the data, with very little effort.
Game tools developer ZibraAI has unveiled Zibra VDB Compression, a new AI-based technology for compressing volumetric data in OpenVDB format “up to 20 times”.
ZibraAI claims that the tech, which will be available next year via Unity and Unreal Engine plugins, makes it possible to display smoke and fire simulations of a complexity previously only possible in VFX work in-game.
The firm also plans to release the technology as a plugin for Houdini.