Key facts
- Jev is a new AI model from TypeSafe AI, founded by former OpenAI researcher Diogo Almeida.
- The model does not output text but instead produces probabilities, termed 'calibrated decisions'.
- Jev is described as faster, cheaper, and incapable of hallucination due to its output format.
- Developers are using Jev for software automation, finding it a more robust and cost-effective alternative to LLMs.
- Jev can also augment LLMs by acting as a 'System One' model for monitoring and preventing misbehavior.
- Almeida hopes Jev will lead to widespread, emergent, and distributed smart software.
Diogo Almeida, a former OpenAI researcher and co-inventor of reinforcement learning from human feedback (RLHF), has launched Jev, a new AI model from his startup TypeSafe AI. Almeida, who felt current AI models were not useful for automation despite their language capabilities, founded TypeSafe AI to address this gap.
Released this week, Jev is a transformer-based model that does not output text but instead generates probabilities, or 'calibrated decisions.' This approach makes the model significantly cheaper and faster than traditional large language models (LLMs) and prevents hallucination because users define the outputs in advance. Input tokens are metered by the billion, and output tokens are free.
Developers have shown strong interest in Jev, with high demand briefly overwhelming the company's API. Software engineers are finding it a more robust and cost-effective way to integrate intelligence into their code. For instance, Vercel used Jev to replace OpenAI's ChatGPT Luna 5.6 for command classification, achieving results 5 to 18 times faster and with greater accuracy. Bryo AI's CTO, Nikhil Mudholkar, found Jev to be 10 to 20 times cheaper than Gemini for classifying business emails, noting its valuable confidence scores for automating workflows.
Beyond replacing LLMs in certain tasks, Jev can also augment them by acting as a monitoring agent to check for misbehavior or 'jailbreaks,' according to Almeida. Armin Ronacher, CTO of Earendil, noted that Jev delegates the hallucination problem to the user, who must decide whether to act on a decision based on its probability score. Ronacher also sees potential for Jev in model routing, efficiently sorting workloads due to its low cost and speed.
Named after economist William Stanley Jevons, the model is designed to make intelligence more accessible and widely deployed. Almeida envisions a future with pervasive smart software, more akin to the early internet than current mega-app structures. TypeSafe AI is also developing Jev for new modalities, focusing on intelligence rather than hype.