What happened
Researchers built a large dataset of 424 hours of primate video from research sites and the web, then trained a video-understanding AI model on it to recognize what different primates are actually doing in the wild. This approach works better than using human-centered AI models, and it needs far fewer labeled examples to learn new behaviors — making it practical for conservation and behavioral research where getting labels is expensive and time-consuming.
Why it matters
For the first time, an AI model trained on primate-specific video generalizes well to new datasets without extensive retraining, which means researchers can now deploy computer vision to monitor wild primate populations at scale without needing to manually label thousands of hours of footage for each new study site.