Tech Desk: Diogo Almeida, a co-creator of ChatGPT and inventor of RLHF, has launched a revolutionary new AI model called Jev through his startup, TypeSafe AI. Unlike traditional Large Language Models (LLMs) that are optimized for human language and prone to ‘hallucinations,’ Jev is designed specifically for automation, producing calibrated decisions and probabilities instead of text.

This novel transformer-based model offers unprecedented advantages: it is incredibly fast, cheap, and inherently incapable of hallucinating because its outputs are pre-defined. Input tokens are metered by the billion, not the million, making its operational cost significantly lower than LLMs. Developers are already thrilled, with demand briefly overwhelming TypeSafe AI’s API.

Jev is proving to be a game-changer in practical applications. Vercel, for instance, saw 5 to 18 times faster and more accurate results replacing OpenAI’s Luna 5.6 for command classification. Similarly, Bryo AI found Jev 10 to 20 times cheaper than Gemini for email classification, praising its ‘real probability’ confidence scores as ideal for workflow automation.

Beyond replacement, Jev can augment existing LLMs, acting as a low-cost, smart check to prevent agent misbehavior or jailbreaks. Its speed and affordability also enable efficient model routing. Almeida envisions Jev leading to ubiquitous, distributed ‘smart software,’ echoing Jevons’ paradox where falling costs lead to widespread adoption. Trained on unique synthetic data, Jev stands alone for now, but its utility suggests competitors will soon emerge, transforming how intelligence is integrated into software.