A fresh approach to artificial intelligence that sidesteps traditional large language models (LLMs) is capturing the attention of software developers. Startup TypeSafe AI has launched Jev, a transformer-based model that avoids generating human text, focusing instead on calibrated probabilities and decision-making for automation tasks.
Co-founded by former OpenAI researcher Diogo Almeida, the company aims to solve the inefficiency of forcing computers to communicate via human language. By eschewing text generation, Jev operates at a fraction of the cost, runs significantly faster, and completely eliminates hallucinations since valid outputs are predetermined by the users.
Early adopters across the tech industry are already leveraging Jev for tasks like command safety classifiers and workflow automation, reporting dramatic speedups and cost reductions compared to standard LLMs. Moving forward, TypeSafe AI plans to introduce new modalities, hoping to usher in a wave of distributed, cost-effective smart software.
- Jev skips text generation to output calibrated probabilities and decisions.
- Built by former OpenAI researcher Diogo Almeida via TypeSafe AI.
- Eliminates hallucinations and achieves up to 18x faster processing speeds.
- Enables cheaper and more efficient software automation and agent routing.
Sources:
