How Jev's decision model fits on a trading desk
The post frames Jev as a trading desk tool: it evaluates predefined decisions in parallel, returns three answer types with confidence scores in 70-500ms, and is used to route orders, flag risky trades and confirm fills.
A co-creator of ChatGPT just launched an AI model that can't hold a conversation. 200x faster. 400x cheaper. Structurally incapable of hallucinating. A trading desk doesn't need a smarter chatbot. It needs exactly this. It's called Jev. Diogo Almeida helped build the RLHF behind ChatGPT at OpenAI. His new company spent two years building the opposite of it. What it actually does: 1. Doesn't generate text. Evaluates decisions you define upfront, all at once, in parallel. 2. Three answer types: pick one option, score on a scale, or answer yes/no as a probability. 3. Every answer comes with a confidence number. 91 percent, act on it. 52 percent, stop and escalate. 4. Can't return an answer outside what you defined. Can still confidently pick the wrong one inside it. 5. 70 to 500 milliseconds per call. $0.042 per million input tokens. Output free. 6. On a trading desk: it's what routes an order to the right execution path, flags a risky trade before it fires, and confirms a fill actually happened before the next one goes out. The name isn't random. Jev is short for Jevons - the economist who noticed in 1865 that more efficient steam engines didn't cut coal use. They increased it. Cheaper intelligence doesn't shrink how much gets used. It multiplies it. That's not just an AI story. It's a market structure one. Cheatsheet's below.
This post quotes another post: “Building an HFT trading bot with Jev: one buy/sell decision per block”