Module 1 · Decisions, not text · Lesson 1
What a System One model is
How a System One model differs from an LLM and a reasoning model, why it is trained for calibrated decisions instead of human preference, what it gives up, and how to tell which kind of model a task needs.
By the end of this lesson you can
- Distinguish an LLM, a reasoning model and a System One model by their outputs, training objective, latency and typical use
- Explain why software wants typed decisions with probabilities rather than prose, and what parsing prose costs
- Explain why optimising for human preference (RLHF) or verifiable correctness (RLVR) does not by itself yield calibrated probabilities
- Classify real tasks as code, System One, LLM, reasoning model or human work, and justify each choice
- Separate what is established, what is vendor-measured and what is vendor-asserted in claims about Jev