Explainer: System One models like Jev and calibrated confidence
An explainer says Jev skips text generation and outputs structured decisions directly, trained for honest confidence rather than human preference, suited to high-frequency small decisions like spam checks and refund eligibility.
the guy who co-invented ChatGPT just announced a whole new class of AI models... and it could get us to AGI much faster they're called system one models, and the name comes straight from how your own brain works your brain runs two modes: - system 2: slow, deliberate thinking, like doing long division on paper - system 1: instant gut decisions, like catching a ball or reading a face every AI model you've ever used is system 2 only ask it a simple yes or no and it still thinks out loud, word by word, burning seconds and money so half of what a brain does... but now we have Jev instead of writing text, it takes messy information in and spits out a direct decision, in a strict format software can use instantly the training method is the game changer here normal models get rewarded for answers humans like, so they learn to sound confident even when they're wrong... that's literally where hallucinations come from so Jev gets rewarded for honest confidence instead when it says it's 90% sure, it's right about 90% of the time, and because it only answers in that strict format, it can't ramble or make things up and this is why it's so interesting: the bulk of AI work inside real products is tiny decisions... is this spam, does this refund qualify, which folder does this go in right now companies point big slow chat models at those questions a system 1 model answers in under half a second, at a fraction of the cost, and tells you exactly how much to trust it thanks to this new tech, AI stops being a chatbot you wait on and becomes a reflex built into EVERY piece of software and if AGI is supposed to work like a real brain, it needs both modes... until today we only had one super excited for this