Why is everyone talking about Jev? The AI tool ‘200 times’ faster than ChatGPT

Why is everyone talking about Jev? The AI tool ‘200 times’ faster than ChatGPT

The AI industry has been upset once again with the launch of Jev, an artificial intelligence tool that promises to dramatically change how the technology works.

Its creators say that it is almost 200 times faster and 450 times cheaper than competitors such as ChatGPT and Claude. It achieves that performance by working in a fundamentally different way from those chatbots, not relying on the same technology that underpins large language models.

Jev was created by TypeSafe, an until now largely unknown AI company founded by an engineer who previously worked on ChatGPT. Founder Diogo Almeida said that the project was inspired by the recognition that there was "something really big missing" from OpenAI's systems, that would hold them back from reaching the kinds of human-level performance that artificial intelligence companies are searching for.

The system is built to output structured decisions, rather than more text. Users can feed it information alongside a series of questions, as well as a limited structure for how it should give the answer.

The system will then output a decision, with a level of confidence, rather than more text. A person might ask it to read an email and ask whether the person writing it is likely to take a certain action, for instance, and Jev is built to provide a "yes" or "no" alongside an expression of certainty, rather than giving out a chat response.

It is that structure – as well as the radically different technology and training process required to build the system – that allows Jev to be far quicker and more efficient than chatbots powered by large language models, TypeSafe said. It also means that the system is unable to hallucinate because it is inherently more restricted, the company claimed.

TypeSafe acknowledges that traditional, large language model systems are still likely to be useful in some cases: chatbots and coding agents where a human can watch what they are doing, for instance, as well as situations where the correctness of an answer can be easily checked. But Jev is more useful for situations where users are making specific decisions, working through large amounts of data, and real-time applications where the speed of the model will be useful, it said.

In a manifesto on its website, TypeSafe said that large language models had been "optimised for human preferences" through the way their training works. "This has led to models that are superhuman at instruction following, and are what we now call 'chat'," it writes.

But that means that the systems have "inherent issues" including the fact they are overly confident about their answers and are unreliable. That means that humans must oversee their work, they write.