AGI Index dimension · weight 9%
Learning
Acquiring new capability after training ends.
Whether a deployed system can acquire durable new skills from experience, rather than being re-trained. Includes in-context adaptation, but weights persistent improvement far more heavily.
The argument
Evidence and counter-evidence
Both sides are published at equal weight. A framework that only records what raises a score is not measuring anything.
Raises the score
- In-context adaptation is strong enough to substitute for fine-tuning across a widening band of tasks.
- Retrieval and memory systems let deployed models incorporate new information without weight updates.
Holds it down
- Almost nothing learned in deployment persists into the model itself. The weights are static; the scaffolding around them is not learning in any deep sense.
- Continual learning without catastrophic forgetting remains substantially unsolved.
- This is the dimension where progress most resembles engineering around a limitation rather than removing it.
What the score reads from
In-context adaptation
Durable skill acquisition
Related coverage
Everything bearing on learning
Context
Where this sits against the rest
- Multimodality82
- Coding81
- Tool use76
- Reasoning74
- Memory67
- Generalization66
- Planning61
- Autonomy53
- Social intelligence52
- Scientific discovery48
- Embodiment41
How these scores are producedMethodology v1.2