Loss as a scoreboard
Is this prediction good? By how much?
The idea inside
Loss = surprise; lower is better; it is the game's score.
After this lesson
You can explain what training loss measures: the model's surprise at the truth.
Where it leads
Millions of knobs: how do we tune them?
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
Loss = surprise; lower is better; it is the game's score.
The question it opens with
Is this prediction good? By how much?
The walkthrough, in the lesson's own words
- The true word was milk. Which model bet more on it, the better predictor?
- Now put a number on your gut feel. Drag the confidence, lower surprise is better.
- That one number is the model's score. Lower means a better predictor.
- Right. A bet 80% on the true word, B only 5%. A's surprise column is short; B's towers. Shorter surprise = better predictor.
- Look at the columns: A bet 80% on the true word, so its surprise is short. B bet only 5%, so its surprise towers. Shorter surprise = better predictor.
- It is the model's score on one guess, how surprised it was by the true word. Lower means a better predictor.
- Next question: the model has millions of knobs. How does it tune them all to push this score down?
- A model's score goes up as it gets better, like a test grade.
- Loss is a surprise score, so lower is better. It measures how surprised the model was by the true next word, and training drives it down. It's not the higher-is-better accuracy numbers a lab headlines at launch.
- Two chatbots answer the same question. One sounds confident and right, the other hedges wildly. What single number would training use to tell them apart?
- Loss, the surprise score you just watched. Training only ever sees one thing: how surprised the model was by the right answer. The confident-and-right one is the low-surprise one, and that is exactly the number training pushes down.
- Low confidence in the truth = tall surprise = high loss. Lower is better.
Key takeaway
The model's loss is now your score.
What you can do after this lesson
You can explain what training loss measures: the model's surprise at the truth.
Check yourself: What does training loss measure?
- The model's surprise at the correct next word(correct)
- How fast it runs
- Its memory usage
- The number of parameters
Loss is how surprised the model was by the actual next word. Low loss means it put high probability on what really came next.
Where it leads: Millions of knobs: how do we tune them?
This is the written summary. The lesson itself is interactive: you predict, drag and operate the mechanism above, and the reveal answers you.