Opinion: OpenAI could copy TypeSafe's Jev classifier and embed it in future models

The piece argues that OpenAI, which has long used LLMs as implicit micro‑classifiers for tool calls, could fast‑follow by either releasing a dedicated Jev‑style model or folding the same classification logic into its flagship models via a special <prediction> tag. If successful, the capability would let the model judge its own reasoning steps, improve safety checks and route work between model sizes without leaving the GPU. The author notes that the key uncertainty is whether TypeSafe’s training data and reinforcement‑learning pipeline constitute a durable moat; Almeida remains confident that “if model quality matters, we will be in a very good position for a long time.”