CITATION — REFERENCE ENTRY
Shut up and calculate: Jev's new AI primitives for coders — The Register
- Key
- jackson2026register
- Authors
- Jackson, Joab
- Issued
- 2026-9-23
- Type
- article-newspaper
- Container
- The Register
Raw CSL JSON
{
"URL": "https://www.theregister.com/a/5298431",
"type": "article-newspaper",
"title": "Shut up and calculate: Jev's new AI primitives for coders",
"author": [
{
"given": "Joab",
"family": "Jackson"
}
],
"issued": {
"date-parts": [
[
2026,
9,
23
]
]
},
"accessed": {
"date-parts": [
[
2026,
11,
5
]
]
},
"language": "en",
"container-title": "The Register"
}
Claims
-
Engineer Archer Hume probed Jev with 10,000 API calls and concluded that it has the broad knowledge of an LLM but, rather than predicting text, calculates decision probabilities directly from the model's internal representations.
-
Developers built game-playing demos with Jev, including Doom, Tetris and chess (where it lost to an open-weight model but cost far less). AI YouTuber Mo Bitar questioned its accuracy; Andrej Karpathy said it exposed demand for low-latency models of acceptable intelligence. The Register notes few public benchmarks exist, and that Jev needs more upfront design than prompting an LLM.
"I know it's fast. I know it's cheap, but is it good?"
Available in