In partnership with
My name is Jev.
(Yapping starts. You can safely skip this part if you just wanna know what Jev does)
If you have opened social media these days, you will see a word that keeps popping up - Jev.
Thinking it was another fad, I had no intention of learning how it works…
Until I saw a YouTube channel that I watch regularly cover it.
I said, “Screw it, it’s just 10 minutes. What can I lose?”
The video actually made me interested in trying Jev.
And I did. It’s easy to use, and I can see some use cases for it.
That’s why I want to summarize what I learned and give you a quick walkthrough of what Jev is without the hype.
Jev is not a LLM.
It's more like a classifier (the exact training method is unknown).
What the heck is a classifier?
Given a photo,
How likely is this a cat girl?
A classifier is an algorithm that makes that judgment.
It will give you a probability from 0 - 1 (0 being zero chance and 1 being 100%)
How's Jev different from other existing classifiers?
While other classifiers are designed to evaluate a specific thing, Jev lets you choose what you want to evaluate.
Meaning, you can define your
state (the picture)
question (how likely is this a cat girl)
Let's run through an example.
You are triggered by social media posts more than you want. So you decided to create a filter to not show posts that are ragebait.
To use Jev, you define two main components:
State
The input data you want Jev to evaluate (right now it can only be text)
In our case, it's the post's text:
Question
This is what you want Jev to evaluate.
In our case, it's the question:
"Is this ragebait?"
Notice how the answer is either Yes/No.
This is called a Noul question.
Turning everything into code:
const post = "Claude or Codex have never suggested Java as a backend.\nI wonder why.";
const response = await client.systemOne({
state: {
post,
},
questions: {
isRagebait: noul("Is this ragebait?"),
},
});
const score = response.answers.isRagebait.noul;
console.log(score);
// Example 0.58 (probably ragebait)
Noul questions return an output value from 0 - 1.
There are two more question formats:
Choice (A,B,C,D)
Score (0-N)
This is essentially how Jev works.
To sum it up, Jev is (likely) a generalized classifier that produces a faster and more reliable output than LLMs.
- Fee from Anime Coders
