AWS Certified AI PractitionerFoundation ModelsHard
A data science team is evaluating several pre-trained foundation models for a complex natural language understanding (NLU) task. They are particularly interested in models that exhibit 'emergent abilities' — capabilities that are not present in smaller models but appear at scale. Which of the following is a common example of an emergent ability observed in very large foundation models?
- AAbility to perform basic arithmetic operations (e.g., 2+2=4)
- BReduced training time compared to smaller models
- CStrong performance on classification tasks with labeled data
- DIn-context learning (performing new tasks from instructions and examples in the prompt)
Show answer & explanationAnswer & explanation
Correct answer: D. In-context learning (performing new tasks from instructions and examples in the prompt)
In-context learning, where a model can learn a new task by simply observing examples and instructions within the prompt without explicit weight updates, is a prominent emergent ability of large foundation models. This capability is not typically seen in smaller models and arises from the sheer scale of parameters and pre-training data.
Why the other options are wrong
- A. Basic arithmetic can be learned by smaller models and is not considered an emergent ability unique to large-scale FMs.
- B. Larger models typically require significantly MORE training time, not less.
- C. Strong classification performance is a general capability of many machine learning models, not an emergent ability specific to large FMs.
Emergent Abilities
Capabilities of foundation models that are not present in smaller models but appear at scale, often without explicit training for those specific tasks.
- Arise from scale and vast pre-training
- Not predictable from smaller models
- Examples: in-context learning, complex reasoning
Memory trick: Emergent abilities are like SURPRISE superpowers appearing at a certain size.