Last week my wife used Claude to sort through her emails.
Days later, a new AI model shipped that could've done the same, better, and for cents on the dollar.
I haven't had the heart to mention it to her. Yet.
That model is Jev by @TypeSafe, from one of ChatGPT's co-inventors.
It's been making waves because it doesn't work like the AI we've gotten used to:
You don't ask it anything. You don't chat with it. It doesn't execute a task for you.
With Jev, you hand it a bunch of data (your context), a related question, and the format of a single answer.
Want this running in your team?
evoilabs starts with one task, the one your team repeats every week, and builds the system that handles it. Then your team builds the next one.
See how it works →For instance:
- "Here is an email. Which of my usual labels should I apply? Give me a score for each label."
- "I just received a refund request from a user. Is this fraud, or should I refund? Give me a likelihood of fraud."
No paragraphs. No reasoning. You tell it what the output must look like. Not should. Must.
To the tech crowd, this model is a "classifiers," a type of work that's been around for a long time. I launched machine learning models like these at Amazon in the 2010s. What's new is that it can work on unsorted input (LLM-like context) and the price (hundreds of times cheaper than frontier models).
I used it on this post. It scored five opening lines we came up with, and my favourite lost.
Jev didn't write a word. It just scored possible hooks. I picked.
So, how much of your system or process today is just repeated classifying and categorizing? Where would you first use Jev?
------
My name's Mathieu, and I demystify AI for founders and execs, and help transform their teams from chatting with AI to building systems that compound in value.