In my view, models like JEV will bring about enormous

Built self-play into a Pac-Man agent using the Jev API, replacing hand-coded if-statements with a dynamic classifier that picks moves from game state.

The original post, translated

In my view, models like JEV will bring about enormous changes. Current LLM models simulate the operation of System 2. For me, the implementation of JEV is one of the newest concepts to appear in the past few months.

Although it doesn't seem particularly special (just a dynamic classifier), it's surprising that we've had to wait so long for such a model. Actually, a general classifier, an intelligent "if" structure... In my previous articles, I've already explained the operating principles of this kind of model. I have access to the APIs of @typesafeai and @CompleteSkeptic, so I implemented self-learning functionality in my own "pac-man" model.

This example is a bit "playful," because the query pattern is fixed, but in many systems, such as intelligent agent systems, both the pattern and the state change over time (in the "pac-man" model, only the state changes). In more complex systems, the structure of the state and the values of the state attributes both change over time, and the same is true of the classifier's contract conditions (the number of categories and their semantics). And here, we have a System 1 model—it makes decisions.

Of course, this functionality can also be implemented on an LLM, but that would be slower, more expensive, and would require processing requests sequentially (the corresponding answers can be seen in the context). One can use response-based diffusion techniques instead of filling response templates with random data (this would be faster)—just generate test samples. However, what really matters is the concept of System 1 itself. I think its range of applications is very broad: intelligent applications, intelligent agent systems, client programs for coding—thereby reducing costs, increasing operating speed, improving the quality of solutions, and so on.

The post above is a machine translation from pl; the untranslated text is in the fold-out.

Engagement when collected

Numbers are a snapshot taken from X when the case was added to the library (schema v1, collected 2026-09-19); they will not match today.