Meta Platforms (META, Financials) has released a set of AI innovations created by its Fundamental AI Research branch. This shows that the company is still working to improve machine intelligence and encourage teamwork in the tech community.

• Warning! GuruFocus has detected 5 Warning Sign with META.

META shares are down 2% in Friday's trading. However, year to date, shares are up 78.5%.

A strong video watermarking system, a new way to handle virtual agents, and new ways for AI systems to expand their memory are some of the newest technologies.

At the release, Meta Motivo, a base model meant to control how virtual agents act, will be available. Meta Motivo solves problems like motion tracking and reward optimization without needing any extra code. It does this by imitating human-like moves using unsupervised learning methods, which builds on existing models. According to the company, the model is very adaptable to changes in its surroundings, such as changes in gravity and outside forces. This means that it could be used in realistic virtual reality and animation.

Meta also showed off Meta Video Seal, a method for adding small marks to videos that can be used as a stamp. These watermarks make it possible to find content that was made or changed by AI because they don't disappear when normal editing techniques like cutting and compression are used. This is an attempt to set a standard for content security in AI research, building on the company's previous work with audio watermarking. Researchers with an open license can use the model and the tools and standards that are provided to do more study.

In addition to these tools, Meta also showed off Flow Matching, a creative generative AI system that improves performance and scaling in content creation across a range of media, such as video, pictures, and 3D structures. Flow Matching has already been used in a lot of internal Meta projects and in other businesses that have switched to more modern diffusion models.

This article first appeared on GuruFocus.