The AI Daily Brief · Wednesday, September 16, 2026
A new class of AI models, termed 'judgment models,' are being developed with a different training approach that focuses on producing probabilities for specific questions rather than text. These models, like the one from Typesafe, are significantly faster and cheaper, optimized for calibrated decisions rather than human preference.
“Rather than producing long strings of text, these judgment models, like the one we're discussing today, Jeff from Typesafe, produce probabilities around specific questions.”
“Today, we're exploring the idea behind these models, how they're trained differently, how they can produce these judgments much more quickly and much less expensively.”
“Whereas existing LLMs optimize for human preference, I.E., right-ups and chat responses that humans prefer, the new system, one model, Jeff, optimizes for calibrated decisions, or answers with epistemically honest probabilities.”