Forget Chatbots: Meet Jev, the AI Making Smarter Decisions

AI is evolving faster than ever, and a new player has just entered the arena, promising to change how we think about intelligent systems. Meet Jev, the latest innovation from TypeSafe AI, and it's not another large language model (LLM) designed to generate text. Instead, Jev is built for one thing: making decisions.

This transformer-based model produces probabilities and “calibrated decisions,” a stark contrast to the conversational outputs of LLMs. What makes it even more compelling is the pedigree behind it: Diogo Almeida, a key researcher whose work underpinned OpenAI’s ChatGPT instruction-following capabilities, is a co-founder of TypeSafe AI.

Why Jev Changes the Game

Almeida's vision for Jev addresses a critical gap. As he put it,

The problem is we are optimising for human language. We have been super good at human language for four years, but it's not useful for automation because computers speak a different language.

Jev steps in to speak that computer language effectively.

Launched on September 15, Jev is incredibly fast and cheap. It can provide answers to complex decision-making questions in under half a second for a mere $0.042 per million input tokens, with output being completely free. Imagine the potential for AI agents struggling with costly LLM calls, Jev offers a game-changing alternative, positioning itself as the crucial “decision-making layer” beneath an expanding ecosystem of agents.

This shift could dramatically lower the operating costs for companies running multiple AI agents, significantly improving their margins. It's no surprise that major platforms like Vercel, Cloudflare, LangChain, and Langfuse have already integrated Jev into their stacks. TypeSafe AI recently made Jev available to everyone, with access starting at $5 in credits, equivalent to roughly 120 million tokens.

The Brains Behind the Decisions

After his time at OpenAI, Almeida, alongside co-founders Erik Gafni and Sashsa Sheng, launched TypeSafe AI to tackle the reinforcement learning (RL) problem head-on. Their first model, Jev, is aptly named after William Stanley Jevons, the economist behind the Jevons Paradox, which posits that increased efficiency in resource use leads to increased consumption.

The analogy here is that cheaper, more efficient AI intelligence will lead to its wider and more innovative application.

Jev is a “System One” model, focusing on intuition rather than reasoning.

It's trained exclusively on synthetically generated data using a method TypeSafe AI calls Reinforcement Learning for Calibrated Decisions.

The company emphasizes that Jev is incapable of hallucinating, as users define the context and expected outputs beforehand.

Instead of text, Jev returns structured answers for code, complete with probabilities indicating uncertainty.

Every question is evaluated in parallel, ensuring independent and reliable results.

What’s Next for Decision-Making AI?

With Jev, the industry is betting on a future where AI models serve as foundational building blocks for scalable products.

TypeSafe AI has indicated it will develop more versions of the model for various modalities.

This marks a pivotal moment, moving beyond text generation to a new era of highly efficient, reliable, and cost-effective AI decision-making.

We're just scratching the surface of what smart software can achieve.