Interactive tool

A model isn't born finished. It goes through stages.

When people say "the model was trained," they usually mean one giant step. In reality it's four very different stages, each reshaping the model in a different way, at wildly different cost. Compare all four side by side below, then test your intuition on what changed the model's behavior, and finish with a quick quiz.

01 — The pipeline

Four stages, one model — side by side.

All four stages at a glance. Follow the strip below to watch a model transform from raw weights into a shipped product, then compare what data feeds each stage, how its cost and time stack up against the others, and what actually changes about the model's behavior.

Bars show each stage's cost relative to the most expensive stage (pretraining = full bar). T time · $ money · D data size
02 — Guess the stage

The main event: spot the stage from the symptom.

This is where the pipeline clicks into place. You'll see a behavior change a team noticed after working with a model — your job is to name the stage that most likely caused it. Build a streak, and watch your score climb.

● Interactive challenge
Observed behavior
Which stage caused it?
Score: 0 / 0 Streak: 0
03 — Quiz

Lock in the order and the purpose.

Five quick questions on what each stage does and where it sits in the sequence.

Total score: 0 / 5
04 — Why it matters

Key takeaways for the boardroom.

You don't need to run any of these stages yourself to make better decisions about AI — you just need to know which one you're actually touching.

06 — Go deeper

Knowing the stages changes what you ask vendors.

"Is this model fine-tuned or pretrained for our use case?" and "how was it aligned?" are very different questions with very different cost and risk implications. That's the kind of due diligence we help leadership teams do before they commit budget to an AI initiative.

Talk to us about your AI roadmap