Yesterday, Nvidia Research published a blog post highlighting numerous developments in artificial intelligence research in 2024. Still, some are more interesting than simply generating content from other people’s copyrighted material, such as “StormCast” and “MaskedMimic”, which correspond to more advanced weather forecasting and motion restoration (reconstructing complete from partially visible motion locomotion) respectively for robotics.
There’s also “GluFormer,” which uses artificial intelligence to predict blood sugar levels for up to four years, although it does require past blood sugar monitoring data. GluFormer also makes it easier to determine how dietary changes will affect long-term blood sugar, with studies showing high accuracy for people with conditions like diabetes.
Some of the other touted improvements are slightly less impressive and more ethically questionable than other areas of generative AI. However, they still demonstrate the evolution of technology. “ConsiStory” allows for multiple AI image prompts with a consistent theme, increasing the usefulness of those trying to make certain content narratively consistent with these tools.
At the same time, “Edify3D” and “LATTE3D” are generative AI tools used to build simple 3D models. Existing 3D modelers are not enthusiastic about this. However, they note that AI retopology and UV mapping can be very useful for existing 3D art workloads without removing all the fun and/or paid labor of making the models yourself. There’s also “Fugatto,” a generative artificial intelligence model designed to create new sound profiles (including music) or modify existing ones (e.g. remove background music).
Finally, at the end of its highlight blog post, Nvidia Research summarizes the many improvements and benchmark wins. These wins include the autonomous driving framework “Hydra-MDP” that won the Autonomous Grand Challenge at CVPR 2024, and Nvidia Blackwell’s industry-leading performance in MLPerf benchmarkand “FoundationPose”, which achieved first place on the BOP leaderboard for model-based pose estimation of unseen objects.
In its original blog post, Nvidia linked to the relevant research paper A more detailed overview of these advances.