CITATION — REFERENCE ENTRY
A new kind of AI model from a ChatGPT inventor is thrilling developers — TechCrunch
- Key
- fernholz2026newkind
- Authors
- Fernholz, Tim
- Issued
- 2026-9-18
- Type
- article-newspaper
- Container
- TechCrunch
Raw CSL JSON
{
"URL": "https://techcrunch.com/2026/09/18/a-new-kind-of-ai-model-from-a-chatgpt-inventor-is-thrilling-developers/",
"type": "article-newspaper",
"title": "A new kind of AI model from a ChatGPT inventor is thrilling developers",
"author": [
{
"given": "Tim",
"family": "Fernholz"
}
],
"issued": {
"date-parts": [
[
2026,
9,
18
]
]
},
"accessed": {
"date-parts": [
[
2026,
11,
5
]
]
},
"language": "en",
"container-title": "TechCrunch"
}
Claims
-
Diogo Almeida was an OpenAI researcher who helped build ChatGPT and helped invent reinforcement learning from human feedback (RLHF); about two years before Jev's launch he left to found TypeSafe AI. He argued that language models optimise for human language, while automation needs output made for machines.
"We have lightning in a bottle, and yet it is not useful."
-
Shortly after launch, demand was high enough that TypeSafe briefly could not serve API users.
-
Jev is a transformer-based model that is not an LLM and outputs probabilities rather than text. It is trained only on synthetic data. Almeida declined to discuss the architecture in detail, and outside observers suspect it is built on an open-weight LLM.
-
Vercel engineer Pranit Sharma said Vercel switched a safety classifier for commands from an OpenAI model to Jev, and it ran five to 18 times faster and was more accurate.
Available in