So, yesterday I was doomscrolling YouTube and just saw a video byHarkirat Singh. This video was about the new System One Model Jev.
After watching that video, I thought of putting my thoughts on it.
So, basically, what I understood is that Jev is not an LLM; it is just a classification model trained using a method they call Reinforcement Learning for Calibrated Decisions (RLCD).
Now, the biggest difference b/w LLM and Jev is that a normal LLM could give answers to what we are querying for, maybe with some amount of hallucinations, but here in Jev, it only tells whether it's ok or not (yes or no, 0 or 1), similar to classic classification models but in a structured manner without hallucinations.
Jev is very fast and way cheaper than LLMs.
You can see more differences on this blog. Jev can be used for AI automations where, at certain points, classification is required.
Or any other application where a perfect classification or decision needs to be made and there should be no space for wrong answers, and the response should be faster and cheaper.
More use cases of Jev are listed here.