Lesson 267 of 2116
Emergence vs. Scaling
Some capabilities grow smoothly with scale. Others seem to appear out of nowhere. Telling them apart is a whole research program. The Big Question Is AI capability a smooth climb or a staircase?
Lesson map
What this lesson covers
Learning path
The main moves in order
- 1The Big Question
- 2emergence
- 3scaling
- 4phase transition
Concept cluster
Terms to connect while reading
Section 1
The Big Question
Is AI capability a smooth climb or a staircase? The answer is probably 'both, depending on how you measure.' Understanding the argument is central to forecasting what the next generation of models will and will not do.
The emergence camp
Wei et al. (2022) catalogued capabilities that appeared to 'emerge' at particular scales — arithmetic, instruction following, in-context learning. Below a threshold, performance was near random; above it, performance jumped sharply.
The mirage counter-argument
Schaeffer, Miranda, and Koyejo (2023) argued that many emergent abilities are a function of the metric, not the model. Switch from strict exact-match to partial-credit scoring, and the cliff becomes a gentle hill. Emergence might be about how we look, not what is there.
Compare the options
| View | Claim | Implication |
|---|---|---|
| Strong emergence | Capabilities really do appear at thresholds | Forecasting is hard; surprises are inevitable |
| Mirage view | Smoothness is hidden by harsh metrics | Forecasting is possible with better metrics |
| Middle ground | Some emergence is real, some is measurement | Depends on task — check both framings |
Implications for evals
- 1Report both strict and partial-credit scores when possible
- 2Sample densely around suspected transition points (compute, parameters)
- 3Use continuous metrics (log-likelihood) alongside discrete (accuracy)
- 4Probe for capability before release, not after scale-up
“Our findings suggest that existing claims of emergent abilities are creations of the researcher's choice of metrics.”
Key terms in this lesson
The big idea: whether AI capabilities emerge suddenly or grow smoothly depends partly on how you look. Either way, the surprises are real enough to plan for.
End-of-lesson quiz
Check what stuck
15 questions · Score saves to your progress.
Tutor
Curious about “Emergence vs. Scaling”?
Ask anything about this lesson. I’ll answer using just what you’re reading — short, friendly, grounded.
Progress saved locally in this browser. Sign in to sync across devices.
Related lessons
Keep going
Creators · 45 min
Open vs. Closed Models: Philosophy and Strategy
Open-source AI is both a technical movement and a political one. Understand the arguments so you can pick a stack and defend it.
Creators · 45 min
Running Your Own Small Experiment
The best way to truly understand an AI claim is to try it yourself. Here is how to run a small experiment that actually teaches you something.
Creators · 32 min
Synthetic Data: When AI Trains on AI
Real data is expensive, private, or scarce. Synthetic data is generated by models themselves. It is rapidly becoming as important as scraped data.