Supporting @MITEECS and @nlp_mit’s Multimodal Machine Learning course (Spring 2026). 🎓 Students are leveraging the multimodal capabilities of Kimi K2.5 to power their final research projects. We look forward to seeing the innovative applications that will emerge this semester. 🔗 Happy coding! ✨

On Jev... When I studied NLP, discriminative AI performed miles better than generative AI. For instance, discriminative API could tell you the sentiment of some text far better than generative AI could write text given a sentiment. Generative AI's output was much like repeatedly hitting the next autocomplete suggestion on your phone's keyboard. Obviously generative AI has tremendously advanced and the output is serviceable in many domains much of the time, though even the latest models still overlook simple solutions. Several things spark my interest in Jev based on my experience: - There will be features where discriminative AI is a much better fit than generative AI, like customer support services instantly routing you to the right department (sometimes I want to talk to a person). - Tomorrow's agents can combine generative and discriminative AI. Imagine Claude Code and Codex's "auto permissions" being extremely accurate and instant. - What if there were an open version of Jev? Discriminative AI will run better on your own devices. Most personal AI tasks are about decision-making, like deciding how important a push notification is to you at this moment, or which apps you might want to use now.