'Unsupervised ICL' refers to
APrompting with only problems, no solutions
BPretraining on unlabelled text data only
CSelf-supervised contrastive learning of prompts
DEmbedding queries via unsupervised encoders
Answer & Solution
Correct answer: A. Prompting with only problems, no solutions
Unsupervised ICL provides only problems (no exemplar solutions), inspired by the task-recognition view of ICL. Effective on MATH, GPQA, BBH where the model already knows the task.
Related questions
The implication for prompt design when shots are abundant isAn architectural risk of many-shot ICL exposed by the XSum result isWhy is many-shot ICL practically more attractive than fine-tuning?Long-context scaling laws (next-token loss)Many-shot ICL can overcomeMany-shot ICL can perform comparably toThe order of examples in many-shot promptsAbstractive summarization on XSum