The model thinks before it answers
Hard problems need working-out. But the model answers in a single pass, no scratch paper to think on.
The idea inside
Reasoning models spend tokens on a private chain of thought before the final answer.
After this lesson
You can explain why reasoning models think before answering, and that the 'thinking' is generated tokens.
Where it leads
If thinking is just tokens, you can buy more of it, a new way to scale.
Inside this lesson
That's the real lesson stage, paused. Claim your pass to operate it.
See how AI actually works, end to end.
This lesson is one stop on the full arc. Unlock all of it, and keep it for life.
What you get
- The 34-lesson main path, a finishable route from a word to agents
- Goal tracks for using AI at work and building AI features
- Boss labs that make you apply a whole act, not just recognize it
- Spaced recall that brings each idea back before you forget
- Course memory: every term defined, with links to where it first appears
- A shareable capability card when you finish the main path
- Lifetime access on every device, every future lesson included
Not videos to watch. You predict, operate the machine, then prove it. That is why it stays.
99 interactive lessons and challenges. No videos, no code.
Free launch pass: lifetime access, no card needed
New here? The first lessons are free to try. Start with lesson 0.1
What this lesson shows
Reasoning models spend tokens on a private chain of thought before the final answer.
The question it opens with
Hard problems need working-out. But the model answers in a single pass, no scratch paper to think on.
The walkthrough, in the lesson's own words
- Thinking on, the model worked it out. Flip it back and forth to compare.
- The model just blurted “12”. Flip the switch to make it think first.
- Answer in one shot: the model must commit to the very first token it predicts.
- Watch the tokens get spent first, then flip thinking off to compare.
- Thinking is just tokens spent before the answer. That's the whole trick.
- "Thinking" is just generated tokens spent before the answer.
- There's no scratch paper inside the model, only a stream of tokens.
- The model writes out its reasoning (privately), and each of those tokens gives it more steps to get the hard part right.
- Reasoning models are set up to do this by default, the same predictor, just tuned to work it out first (how that tuning works is an elective later).
- A reasoning model literally thinks, pondering the problem the way a person does.
- It just spends more inference-time compute writing intermediate steps before it answers. That's more computation, not consciousness, and more steps can still reach a wrong answer.
- On a hard multi-step problem, a reasoning model is slower and costs more than a quick chat model, but gets it right where the fast one blurts a wrong answer. What are you paying for?
- Reasoning means the model generates a private chain of thought, spending tokens to work the problem out before it commits to an answer. Those extra tokens are the slower, pricier part, so reach for a reasoning model on hard multi-step tasks and a faster one for simple lookups where working-out is wasted effort.
- Done thinking. The model worked the problem out, then answered.
- No working-out, the first token IS the commitment. The prompt is saturated with 12s, so the guess gets pulled toward 12 before any working-out can happen.
- The reasoning never leaves the model. You just see the answer it earned.
- Answered in one shot. The prompt is saturated with 12s, so the first token gets pulled toward 12 before any working-out can happen.
- Thinking off: same model, no tokens spent first, same wrong blurt as before.
Key takeaway
Thinking is tokens: the model reasons out loud (privately), then answers, and gets harder problems right.
What you can do after this lesson
You can explain why reasoning models think before answering, and that the 'thinking' is generated tokens.
Check yourself: What is a reasoning model's 'thinking'?
- Extra tokens it generates to work things out before answering(correct)
- A separate database lookup
- A larger vocabulary
- Slower hardware
Thinking is just more tokens generated before the answer, a scratchpad. Spending those tokens often helps, but it is text generation, not a hidden lookup.
Prefer reading first? This lesson has a companion explainer: Reasoning models
Where it leads: If thinking is just tokens, you can buy more of it, a new way to scale.
This is the written summary. The lesson itself is interactive: you predict, drag and operate the mechanism above, and the reveal answers you.