There’s a new neural net product making waves in tech media right now, and it’s called Jev. Pioneered by a company called Typesafe, this model doesn’t scribble out pages of narrative when a human user asks it a question: it just spits out some probabilities.

In fact, they’re called Jev a sort of “A2A” tool, not a user-facing one. That in itself is interesting, in the context of a world where we inherently expect AI to be lingual, to be serving us in a chatbot-like way.

To date, most models have been this kind of verbal interface. That has led to knock-down, drag-out discussions about whether the AI is “conscious,” because in imitating human cognition, it’s also imitating human activity, through language.

A product like Jev is, in some ways, a return to something that looks more like yesterday’s deterministic programming. It’s not deterministic, of course, not in the traditional way, but perhaps it’s more bounded: if you give a system like Jev a prompt, you’d expect its response to be more bounded, and less capricious and/or wide-ranging.

That brings us to another consideration: cost.

Efficiency in Token Handling

The Jev model will generate fewer tokens than more chatty models, on a given prompt: much fewer. After all, all of that text takes token-crunching, translation of ideas into, well, words.

“Most enterprises are using general-purpose LLMs as a universal intelligence layer, even when the application only needs a fast, bounded decision. That is like using a full enterprise service bus to answer a yes/no routing question,” said David Linthicum, an independent tech consultant, as quoted in coverage by Anirban Ghoshal at InfoWorld . “Combine that with the dynamics of an agentic workflow, where a single task can trigger many model calls, the cost and latency of using a general-purpose LLM for every decision can quickly add up. A Jev-like model that can handle some of the bounded decisions of the workflow through probabilistic function calls gives enterprises a way to reserve the general-purpose LLMs for the parts that actually require them.”

In a way, all of that text generation is just wasted compute, so it makes sense to change the format of a model this way, to provide an alternative. Marvin Minsky, an eminent voice at MIT, used to say that the brain is a collection of many small computers. If so, Jev can represent a digital analog of a part of the brain distinct from language centers, that essentially does something different.

Github user Vinnylarouge describes the architecture of Jev this way:

“Each option becomes a query vector, which is a short list of numbers representing its text,” the programmer writes. “The query assigns attention weights to the context tokens. Those weights make one context vector for that option. A shared dot product turns each option and context pair into one score. A softmax, which converts scores into probabilities that sum to one, runs across the options.”

Yes, Jev is a different animal. In a guide to this model type, NNNN illustrates the use of browser state evaluation through element identification, click prediction, dynamic flow branching, and multi-tab orchestration, noting that Jev is trained using Reinforcement Learning from Calibrated Decisions (RLCD).

The reasoning here is that a non-conversational model like Jev can get applied to business workflows in new ways. Business queries rarely need a lot of commentary: they just need numbers returned. Jev is good at doing one without the other.

As for the name, Thomas Claburn has this to say over at The Register:

“The name ‘Jev’ is a reference to 19th century English economist William Stanley Jevons, known for Jevons paradox, who argued that technology efficiency related to coal usage increased coal consumption rather than reducing it,” Claburn writes. “The name also reflects the AI industry’s bet that greater token efficiency will increase token consumption even as token prices decline. That wager assumes the market for tokens will be as broad as the market for energy – an assumption that isn’t yet settled given that many people who have access to AI tools just don’t have a use for them or avoid AI tools due to moral objections.”

This makes sense, in that Jevon’s paradox basically assumes robust demand or appetite for whatever is being proliferated. We’ll see.

So yes, call it Jev, and look for it in business workflows.