ChatGPT broke Diogo Almeida’s heart.
Almeida was an OpenAI researcher who helped build the chatbot and then invent reinforcement learning from human feedback (RLHF), the model-training technique perhaps most responsible for our current age of AI. But despite its capabilities, he was disappointed.
“We have lightning in a bottle, and yet it is not useful,” Almeida told TechCrunch. “I’ve been battling that problem since then. It took me a while to come to the conclusion: 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.”
Two years ago, Almeida left OpenAI to start TypeSafe AI, a startup trying to fix that problem. This week, the company released a new transformer-based model, Jev, that is not a large language model (LLM). It doesn’t output text, but instead produces probabilities, or what the company calls “calibrated decisions.”
Eschewing language does a few things: It makes the model incredibly cheap and fast, and because users define the outputs in advance, it cannot hallucinate. Its output tokens are free, and input tokens are metered by the billion, not the million.
Developers are taking a great interest in the product; the company briefly lost the ability to serve users from its API because demand was so high. Jev appears most useful for software automation. Thus far, software developers see it as a cheaper and more robust way to incorporate intelligence into their code.
For example, Pranit Sharma, a software engineer at Vercel, a company making agentic infrastructure, said his company had used OpenAI’s ChatGPT Luna 5.6 to run a classifier to review commands for safety. When Vercel replaced OpenAI’s Luna with Jev, it got results 5 to 18 times more quickly and with greater accuracy.
Another developer, Bryo AI CTO Nikhil Mudholkar, tested Jev against Gemini for classifying business emails. In his test, Gemini was slightly more accurate, but 10 to 20 times more expensive. More interesting to Mudholkar were Jev’s confidence scores — “it is the only one that hands back a real probability which makes it ideal for automating workflows!!”
Besides replacing LLMs in certain use cases, the new model can also augment them, acting as a smart check on misbehavior. Using agents to monitor agents can quickly become expensive, but using Jev to do so, Almeida argues, makes sense. He sees users deploying Jev to track LLM agent traces and prevent jailbreaks.
“At the end of the day, it delegates the hallucination problem a little bit to the user,” explained Armin Ronacher, the CTO of Earendil, which builds the open-source model harness Pi. “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.”
Another potential use for Jev is model routing, Ronacher said. Predicting whether a given workload requires a specific model would be useful, but using an LLM for the job would be expensive. Jev’s low cost and speed make that kind of real-time sorting possible.
And that’s Almeida’s hope. The model is named after William Stanley Jevons, the 19th-century economist whose eponymous paradox describes how the falling cost of a commodity can lead to it being used more and more. In this case, the falling cost of intelligence should lead to its widespread 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 the mega apps that people are trying to build right now,” Almeida said.
Almeida is tight-lipped about the model’s architecture, which outside observers suspect is built on top of an open-weight LLM. The company refers to Jev as a “System One model,” focused on intuition rather than reasoning, and specifically focused on the right task. Almeida says Jev is trained exclusively on synthetic data using a 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,” he told TechCrunch. “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 now, Jev stands alone as this kind of model, but Ronacher expects that competitors will spring up now that its utility is apparent.
“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 said.
TypeSafe itself will be building more versions of the model, in new modalities. Asked if TypeSafe is a frontier lab, Almeida said, “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.”
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.
Key Takeaways:
- Beyond Language for Automation:Diogo Almeida, an OpenAI veteran, founded TypeSafe AI to address the limitations of Large Language Models (LLMs) for automation, arguing they optimize for human language, not machine efficiency.
- Introducing Jev, a Non-LLM AI:TypeSafe AI’s new transformer-based model, Jev, outputs “calibrated decisions” (probabilities) instead of text. This approach makes it incredibly fast, cheap, and immune to hallucination, ideal for precise software automation.
- Practical Impact & Vision:Jev is already being adopted by developers like Vercel and Bryo AI for tasks requiring speed, accuracy, and cost-efficiency, often outperforming LLMs. TypeSafe envisions a future of ubiquitous, distributed “smart software” enabled by the falling cost of intelligence.
From LLM Disillusionment to a New AI Paradigm: TypeSafe AI Unveils Jev, the Model That Speaks Machines’ Language
Diogo Almeida, a name synonymous with the foundational innovations at OpenAI, including the very development of ChatGPT and the groundbreaking Reinforcement Learning from Human Feedback (RLHF), found himself facing a profound disappointment. Despite having helped usher in the current era of AI, he felt that the “lightning in a bottle” that was ChatGPT wasn’t translating into truly useful automation. His core realization? Large Language Models, for all their prowess in human communication, were fundamentally optimized for the wrong language when it came to computers.
“We have been super good at human language for four years, but it’s not useful for automation because computers speak a different language,” Almeidatold TechCrunch, articulating the chasm he perceived. This conviction led him to depart OpenAI two years ago and establish TypeSafe AI, a startup dedicated to bridging this gap. This week, TypeSafe AI pulled back the curtain on its answer: Jev, a novel transformer-based model engineered to serve the precise needs of machine automation.
Jev: An AI That Prioritizes Precision Over Prose
Unlike the LLMs that dominate headlines, Jev is not designed to generate human-like text. Instead, it produces “calibrated decisions” – high-confidence probabilities that offer a direct, machine-readable output. This fundamental architectural choice delivers several game-changing advantages. By eschewing the complexities of natural language generation, Jev achieves unparalleled speed and cost-efficiency. Its output tokens are effectively free, and input tokens are metered by the billion, a dramatic reduction compared to the millions typically seen with LLMs.
Crucially, because users pre-define Jev’s potential outputs, the model cannot hallucinate. This inherent reliability makes it an ideal candidate for critical software automation tasks where accuracy and predictability are paramount. Jev represents a paradigm shift from models that interpret and generate human intent to one that directly provides the probabilistic decision-making machines truly understand.
Real-World Validation: Developers Embrace Jev’s Efficiency
The developer community has reacted with immediate enthusiasm, so much so that TypeSafe AI’s API briefly buckled under the initial surge in demand. Jev’s utility for software automation is quickly becoming apparent, with early adopters praising its ability to integrate intelligence into code more robustly and affordably than traditional LLMs.
Pranit Sharma, a software engineer at Vercel, a company specializing in agentic infrastructure, recounted how his team previously relied on OpenAI’s ChatGPT Luna 5.6 for a safety classifier. Switching to Jev yielded remarkable results: a 5 to 18 times increase in speed coupled with enhanced accuracy. This real-world performance underscores Jev’s capacity to streamline and accelerate critical operational workflows.
Similarly, Nikhil Mudholkar, CTO of Bryo AI, conducted a comparative test between Jev and Gemini for classifying business emails. While Gemini showed a slight edge in accuracy, Jev was an astonishing 10 to 20 times more cost-effective. Mudholkar was particularly impressed by Jev’s “real probability” confidence scores, highlighting them as “ideal for automating workflows!!” This feature allows developers to set clear thresholds for action, delegating decision-making confidently to the AI.
Beyond direct replacement, Jev also holds promise in augmenting existing LLM deployments. Almeida envisions Jev acting as an intelligent oversight layer, monitoring LLM agent traces to prevent misbehavior or “jailbreaks.” Integrating costly LLM agents to monitor other agents is often prohibitive, but Jev’s low operational cost makes such robust oversight a practical reality. Armin Ronacher, CTO of Earendil, also pointed to Jev’s potential for “model routing,” where its speed and affordability could enable real-time, intelligent allocation of workloads to the most appropriate AI model.
The “System One” Approach and the Power of Synthetic Data
Almeida maintains a degree of secrecy regarding Jev’s specific architecture, though external observers speculate it might leverage open-weight LLMs as a foundation. He describes Jev as a “System One model,” drawing a parallel to the intuitive, fast-thinking cognitive processes, as opposed to the more deliberate, reasoning-focused “System Two.” Jev is explicitly designed to focus on the “right task,” delivering intuition rather than complex reasoning.
A cornerstone of TypeSafe AI’s strategy is its unique approach to training data. Jev is trained exclusively on synthetic data, generated using a proprietary technique Almeida terms “reinforcement learning from calibrated decisions.” He considers the early bet on generating all their data to be “one of the best bets I’ve ever made in my life—better than our launch, in my opinion, better than RLHF.” This focus on statistically well-understood synthetic data allows TypeSafe AI to control the training environment with precision, further contributing to Jev’s reliability and targeted performance.
The Jevons Paradox of Intelligence: A Future of Distributed AI
The model’s name, Jev, is a deliberate nod to William Stanley Jevons, the 19th-century economist whose paradox describes how increasing efficiency in resource use can lead to increased overall consumption. Almeida hopes to invoke a similar phenomenon in the realm of artificial intelligence: as the cost of intelligence plummets, its deployment should become ubiquitous and widespread.
Almeida envisions a future where “smart software” is “all over the place in a way that’s emergent and distributed,” drawing a parallel to the early internet rather than the “mega apps” currently being pursued. This philosophy stands in stark contrast to the “frontier lab” ethos, which he suggests often prioritizes “fear or hype” over practical utility. TypeSafe AI, he asserts, is not building “God in a data center” but rather focusing on delivering tangible intelligence.
The Road Ahead: Innovation and Inevitable Competition
For the moment, Jev stands alone in its niche, but Armin Ronacher anticipates that competitors will inevitably emerge as the model’s profound utility becomes more widely recognized. The current “cheap and subsidized” nature of LLMs may have inadvertently stifled creativity in exploring alternative AI paradigms, but Jev’s demonstrated success is likely to change that. TypeSafe AI itself plans to expand Jev’s capabilities, developing new versions and exploring additional modalities, all while remaining steadfast in its mission to deliver practical, robust intelligence.
Bottom Line:
TypeSafe AI’s Jev represents a pragmatic and powerful shift in the AI landscape, directly addressing the limitations of LLMs for real-world automation. By focusing on calibrated decisions over human language, Jev delivers unparalleled speed, cost-efficiency, and reliability, poised to unlock a new era of ubiquitous, intelligent software that truly understands and acts in the language of machines. It’s a compelling vision of AI’s future, built not on hype, but on tangible utility.
Source:TechCrunch