TypeSafe AI Launches Jev, a Text-Free Decision Model Inspired by Daniel Kahneman
Former OpenAI developer Diogo Almeida has launched TypeSafe AI and introduced Jev, a text-free AI model inspired by Daniel Kahneman that delivers structured data decisions at high speed.
A new startup named TypeSafe AI, founded by former OpenAI developer Diogo Almeida, has introduced an unconventional AI model called Jev. Unlike traditional large language models that generate text, code, or images, Jev produces no text at all. Instead, it focuses entirely on structured data output, such as multiple-choice selections, numerical scores, and calibrated probabilities.
Inspired by Daniel Kahneman
Jev is built upon System One, a concept inspired by the principles of the late Israeli Nobel laureate and psychologist Daniel Kahneman. Rather than the slow, deliberate reasoning of System Two found in most contemporary models, Jev simulates fast, intuitive thinking. This approach allows developers to receive reliable answers without the "hallucinations" typically associated with generative text models, making it ideal for direct integration into software code.
We trapped lightning in a bottle, and yet it is not useful.
Speed, Cost Efficiency, and Architecture
The system utilizes a unique training method known as RLCD to refine statistical probabilities rather than attempting to satisfy human evaluators. Its parallel processing architecture scans data only once and computes answers to multiple queries simultaneously. Consequently, response times range between 70 milliseconds and half a second, which is up to 200 times faster than a traditional language model.
Pricing is similarly disruptive, charging only for input text measured in billions of tokens. The cost stands at 0.042 dollars per million input tokens with entirely free output and a context window of up to 64,000 tokens, undercutting standard large language models significantly.
