Jev & computer-use
Current work: jev-computer-use My TL for the last few days has been taken over by Jev , a system one model from typesafe.ai . So i applied for the waitlist and got in. To me a interesting use-case for this kind of model is on browser/computer-use, lot of decision to be taken which may not need long reasoning before doing so, the UX improves with the speed we get and automations feel right.…
Jev, a system one model from Typesafe.ai, has become the focus of the reporter's work over the past few days. Impressed by its potential, the reporter applied for the waitlist and was granted access. The reporter sees great value in using Jev for computer-use scenarios, where numerous decisions must be made quickly without extensive reasoning.
Jev can be understood as a classifier, accepting structured text inputs and providing three primitive decision-making options: Choice, Score, and Noul. In contrast to general-purpose language models, Jev offers a decision interface, enabling agents to interact using state machines and offering a typed result.
The reporter is currently experimenting with Jev on a computer-use flow. They note that although Jev's age of LLMs lends it an air of novelty, the true power lies in the data it was trained on. Jev has been trained on 100% synthetic data and claims to have almost no hallucinations. However, they caution that even though the model is deterministic, the same prompt can yield different probabilities each time, making the confidence score less reliable.
The reporter suggests that users may need to develop their own definition of confidence using the provided probability list.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
