Hustler Words – A groundbreaking artificial intelligence model, Jev, developed by former OpenAI researcher Diogo Almeida, is rapidly captivating the developer community by offering a compelling alternative to traditional large language models (LLMs). Almeida, a key figure in the creation of ChatGPT and the inventor of reinforcement learning from human feedback (RLHF), found himself increasingly frustrated with the limitations of LLMs for practical automation, despite their impressive linguistic prowess.
"We have lightning in a bottle, and yet it is not useful," Almeida candidly shared with hustlerwords. He elaborated on his epiphany: "The problem is we are optimizing 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." This realization spurred him to depart OpenAI two years ago and establish TypeSafe AI, a startup dedicated to bridging this critical gap.
This week, TypeSafe AI unveiled Jev, a novel transformer-based model that distinctly diverges from the LLM paradigm. Unlike its text-generating counterparts, Jev produces probabilities, which the company terms "calibrated decisions." This fundamental shift away from linguistic outputs brings several transformative advantages: Jev is remarkably inexpensive, operates at high speeds, and, crucially, is immune to hallucinations because its outputs are predefined by users. Its output tokens incur no cost, and input tokens are metered by the billion, not the million, drastically reducing operational expenses.
The immediate interest from developers has been immense, with TypeSafe AI’s API briefly struggling to meet the overwhelming demand. Jev’s primary utility lies in software automation, where engineers are discovering it to be a more robust and economical method for embedding intelligence into their codebases.
Real-world applications are already demonstrating Jev’s superiority. Pranit Sharma, a software engineer at Vercel, a company specializing in agentic infrastructure, reported significant improvements after replacing OpenAI’s ChatGPT Luna 5.6 with Jev for a safety classifier. Vercel observed results that were five to eighteen times faster and notably more accurate.
Similarly, Nikhil Mudholkar, CTO of Bryo AI, benchmarked Jev against Google’s Gemini for classifying business emails. While Gemini showed a slight edge in accuracy, Jev proved to be ten to twenty times more cost-effective. Mudholkar was particularly impressed by Jev’s "confidence scores," noting to hustlerwords that "it is the only one that hands back a real probability which makes it ideal for automating workflows!!"
Beyond merely replacing LLMs in specific scenarios, Jev also offers powerful augmentation capabilities. It can serve as an intelligent check against misbehavior in LLM-driven agents, a task that can quickly become prohibitively expensive with other models. Almeida envisions Jev being deployed to monitor LLM agent traces and prevent "jailbreaks." Armin Ronacher, CTO of Earendil, which develops the open-source model harness Pi, explained, "At the end of the day, it delegates the hallucination problem a little bit to the user. The user has to say, okay, if this only comes back with 50% probability, maybe this is a coin toss, and I disregard it. But if it’s 95%, sure, then I can do something with it."
Ronacher also highlighted Jev’s potential for efficient model routing, where its speed and low cost could enable real-time predictions about which specific model is best suited for a given workload.
Almeida’s vision for Jev is deeply rooted in the Jevons Paradox, named after the 19th-century economist William Stanley Jevons, which posits that falling commodity costs lead to increased consumption. In this context, Almeida anticipates that the reduced cost of intelligence will pave the way for its ubiquitous deployment. "We think that there’s just going to be smart software all over the place in a way that’s emergent and distributed… much more like the early internet than you know like the mega apps that people are trying to build right now," he mused.
While Almeida remains discreet about Jev’s precise architecture, external observers speculate it may leverage an open-weight LLM foundation. TypeSafe AI characterizes Jev as a "System One model," emphasizing its focus on intuition and task-specific execution rather than complex reasoning. Almeida confirms that Jev is trained exclusively on synthetic data, employing a proprietary technique he calls "reinforcement learning from calibrated decisions."
"We made an early bet that we will be making all of our data, and that has been one of the best bets I’ve ever made in my life – better than our launch, in my opinion, better than RLHF," Almeida proudly stated to hustlerwords. He added, "Half of [our company] is a lab that basically owns this entire subfield of statistically well-understood synthetic data, and that is now my life joy."
For the moment, Jev stands as a unique offering in the AI landscape, though Ronacher expects competitors to emerge as its practical utility becomes undeniable. "We should have seen this earlier in many ways, but presumably because the LLMs are so cheap and subsidized, you often don’t have to be creative yet," he observed.
TypeSafe AI is already planning further iterations of the model, exploring new modalities. When asked if TypeSafe is a "frontier lab," Almeida clarified their mission: "the main product of frontier labs is fear or hype. I would like our main product to be intelligence… [but we are] not a lab in the sense of, you know, like bet on infinite wealth, or a religion, or building God in a data center, or whatever is the thing of today." TypeSafe AI’s focus remains firmly on delivering tangible, cost-effective intelligence for the automation age.
Leave a Comment