Training-data curation
What actually goes into the training set?
The idea inside
Drive a funnel: raw web → dedup → quality filter → decontaminate.
After this lesson
You can explain how training data is cleaned, and why quality beats quantity.
Where it leads
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
Drive a funnel: raw web → dedup → quality filter → decontaminate.
The question it opens with
What actually goes into the training set?
The walkthrough, in the lesson's own words
- The web is mostly junk. Drag the cleaning slider and watch how few survive.
- One worked run. Watch the corpus shrink, stage by stage.
- Now you drive the funnel. Drag through the four stages.
- A quarter the size, and far cleaner. Quality beats quantity.
- Only about a quarter survives. Three of four pages are junk.
- Keep dragging. Each filter drops more, until barely a quarter is left.
- Illustrative counts and quality scores, the real filtering ratios vary by pipeline.
- 100 → 25: a quarter the size, far cleaner. Quality beats quantity.
- A model learns from exactly what it's fed. Duplicates waste training on the same page twice; spam and gibberish teach it to write spam and gibberish; leaked test questions let it memorize answers instead of learning. So a smaller, cleaner pile teaches more than a bigger, dirtier one.
- A team fine-tunes a model on every support ticket they have, raw, and it gets worse, not better. What went wrong?
- A model learns from exactly what it's fed, so raw tickets full of duplicates, junk, and noise teach it duplicates, junk, and noise. Dedup, filter for quality, and remove anything that overlaps your test set first: a smaller clean pile beats a big dirty one.
Key takeaway
The corpus shrinks and 'quality' rises.
What you can do after this lesson
You can explain how training data is cleaned, and why quality beats quantity.
This is the written summary. The lesson itself is interactive: you predict, drag and operate the mechanism above, and the reveal answers you.