Loading slide

Next Token, Every Time

  1. 01Next Token, Every Time
  2. 02Memory card
  3. 03One word at a time
  4. 04Painting into a corner
  5. 05Temperature: how random is random?
  6. 06Turn the dial yourself
  7. 07Take a breath
  8. 08No inner monologue
  9. 09Thinking out loud actually helps
  10. 10Reinforce your understanding
  11. 11Question: No plan
  12. 12Question: Why reasoning out loud helps
  13. 13Quiz: answer
  14. 14The output is the thinking
  15. 15Want to go deeper?
8 / 14
BackNext

No inner monologue

People naturally imagine a model thinking before it speaks. Forming a complete thought somewhere private, then putting it into words. That is how it feels to be a person: you know what you mean, then you say it.

That is not what happens here. The model has no private place to hold a thought. It has the text so far, and it picks what comes next. There is no finished idea sitting behind the words, waiting to be expressed. The words are the thinking. The page is the only place any of it exists.

This sounds like a limitation, and in one way it is. But it has a strange and useful flip side. Because the model has no private scratchpad, the only way it can "work something out" is to write the working down where it can see it. Anything it puts on the page becomes part of the context, which means the model can attend back to it, in exactly the sense of attention we built last module, when choosing the next word.

So if you force a hard question straight to its answer, the model has nowhere to do the work and often fumbles. But give it room to lay the steps out first, and each step it writes becomes scaffolding the final answer can lean on. That single shift, from leaping to the answer to building toward it, turns out to matter enormously, and it is the next slide.

Citations
Citations