Jev AI claims it cannot hallucinate, but critics question the hype
Diogo Almeida launches Jev AI model with anti-hallucination claims
Diogo Almeida, who describes himself as a ChatGPT "co-creator," has emerged from stealth with a new AI startup called TypeSafe and a model named Jev. The company raised $40 million in seed funding and is marketing Jev as an AI that cannot hallucinate.
However, the article's investigation raises questions about both Almeida's role in creating ChatGPT and the novelty of Jev's technology.
Key numbers and claims
- TypeSafe raised a $40 million seed round.
- Jev has a hard-coded cap of 255 output options.
- Input tokens are priced at $42 per billion, with output tokens temporarily free.
- Jev is in early access and described as "days old."
What the sources say about Almeida's role with ChatGPT
The claim that Almeida "co-created ChatGPT" is not strictly accurate. He is listed as the fourth author of 20 on the 2022 InstructGPT paper, which applied a technique called Reinforcement Learning from Human Feedback (RLHF) to language models. That paper was one of several research efforts ChatGPT engineers used during development.
A footnote in that paper identifies Almeida as one of nine "primary" authors, not a team lead. Additionally, RLHF technology predates that paper by nearly five years, and its foundational method came from six other researchers.
How Jev's anti-hallucination feature actually works
Almeida himself has admitted that RLHF "is not useful," calling ChatGPT "unreliable without human supervision" and blaming the technology for issues like mode dropping and overconfidence.
Jev's supposed protection against hallucinations, according to its own documentation, is simply its design as a decision selector rather than a sentence-generating chatbot. It presents users with a limited set of predefined options — such as determining a likelihood percentage — with a hard cap of 255 choices. Because it does not generate text or factual claims, it cannot technically hallucinate. Almeida describes hallucinations as impossible in Jev by design, not through any new algorithmic breakthrough.
What is still unclear
Almeida claims Jev has speed, cost, and quality advantages over competitors, but these claims have not been independently verified. The model is very new and remains in early access, so its real-world performance is not yet well understood.
Why this matters
The Jev launch highlights a broader tension in the AI industry around marketing claims. Almeida's viral campaign leans heavily on his association with ChatGPT's development, even though his actual role was as one author on a supporting research paper. The anti-hallucination claim, while technically true given Jev's limited design, may not represent the kind of AI breakthrough that the marketing campaign implies.
Sources
Newisty Editorial Team covers technology, cryptocurrency, digital products, online platforms, developer tools and the wider digital economy. Our content is researched from official sources, company announcements, public documentation, market data and other primary or reputable sources. Articles are reviewed and edited before publication for clarity, accuracy and useful context.