Almeida's new company, TypeSafe, completed a $40 million seed funding round this month. He stated that the Jev model fundamentally avoids AI hallucinations by designing text through selection rather than generation, and by setting a hardcoded limit of 255 options. Almeida previously contributed to OpenAI's InstructGPT paper, which applied Reinforcement Learning from Human Feedback (RLHF) to language models, but in his promotion of Jev, he admitted that RLHF is "not very useful" and called ChatGPT "unreliable without human supervision."