179 questions
No questions match those filters.
← All questions
Training & Fine-tuningHard· In-Context Learning· Reasoning· Evaluation
The task changes from four-way classification to multi-...
This is one of the questions in the full AI/ML interview bank. Pro unlocks all 1789 questions; Premium includes the same bank plus the highest daily Practice limit.
See plansThe shift to numerical reasoning breaks the local averaging theory of in-context learning, as a weighted average of labels cannot produce a derivation not present in the examples. Use a small number of demonstrations to establish the ‘show your steps’ format, but do not rely on them to cover all instances. Conduct a k-sweep: if accuracy remains flat from k=2 to k=16, it indicates the reasoning concept is not in the pre-training mixture, confirming that fine-tuning is the correct path.