Turning an LLM into a probabilistic decision classifier with logprobs

michellechen · x · 2026-09-19

The author shows how to approximate a jev-style "System One" decision model with an off-the-shelf LLM: the model picks from a predefined answer set, token logprobs score each choice, and the scores are normalized into a probability distribution returned via a jev-compatible API. Thinking mode is disabled for speed and temperature is set to 0 for determinism, yielding typed probabilistic outputs that application code can threshold, rank, or use to defer decisions.

Related event: Developers Turn LLM Logprobs into Probabilistic Decision Outputs(2 posts)→

Original post →

More from coding & agent

coding & agent channel →