The model is forced to show intermediate steps instead of jumping to a finish. Accuracy rises when the work is visible.
Everyone wants
System 2 Super Intelligence.
Only one can own the name.
System 2
Super
Intelligence
The idea
System 2 is the slow mind.
Daniel Kahneman split thinking into two modes. System 1 is fast, automatic, cheap. It recognizes a face, finishes a sentence, slams the brakes. System 2 is slow, deliberate, costly. It multiplies 27 × 14, checks a proof, holds two plans in mind and picks one.
Most of life runs on System 1. The work that changes outcomes runs on System 2. That is why the name moved from psychology into the center of AI.
System 1
- Fast, associative, effortless
- Pattern match. First answer.
- Good for the familiar
- Fails on novelty, logic, long chains
System 2
- Slow, sequential, expensive
- Reason, verify, revise
- Good for the unfamiliar
- The mode you engage when the first answer is not enough
Why it matters now
Labs are training machines to pause.
Base models behave like System 1: next token, first instinct, fluent and often wrong on hard problems. Reasoning models spend test-time compute — they write a hidden chain of thought, check it, and only then answer. That is System 2 made computational.
More thinking tokens at inference, not just more training data. Slower on purpose. Better on purpose.
Deliberate reasoning is the path the field names on the way to super intelligence. Method, then destination.
In practice
How to tell which system is running.
Useful because the label is now everywhere — in papers, product decks, and model cards. The test is simple.
If the answer arrives immediately and feels obvious, that is System 1. If it costs attention, that is System 2.
System 1 commits. System 2 can take the first draft back. Look for a check, a counterexample, a second pass.
Talking while you do it? System 1 can share the channel. System 2 usually cannot. The same is true of models that “think” in a private scratchpad.
Use System 2 when the cost of a fluent error is high: code, law, medicine, strategy, math. Use System 1 when the pattern is stable.
Everyone wants System 2 Super Intelligence.
Only one can own the name.
Acquire the name