An ex-OpenAI researcher just deleted language from the LLM...
Ex-OpenAI researcher's startup Typesafe AI launched Jev, a type-safe classifier model with guaranteed structured output.
“It deleted language from the large language model. And the result is a new type of classifier that's 200 times faster, 400 times cheaper with free output tokens and zero hallucinations.”
Fireship's satirical code report covers Jev, a model from ex-OpenAI researcher Diego Almeida's Typesafe AI ($40M raised), which strips language generation to act as a fast, cheap type-safe classifier returning strictly-typed shapes (choice, score, boolean). The claims (200x faster, 400x cheaper, zero hallucinations) are presented with heavy skepticism as 'trust me bro benchmarks,' signaling interest in specialized structured-output models over general-purpose LLMs. Notable as an industry direction but framed comedically with unverified metrics.