ICL
The ability of pretrained language models to perform new tasks solely from examples provided in the prompt, without updating model parameters.
Timeline (last 90 days)
The cost of ICL is that every prediction must carry examples, and the longer the context, the higher the inference cost; the selection and ordering of examples affect performance; scenarios with extremely large data volumes or very high feature dimensions still require sampling and retrieval strategies.
Recent research provides the complete circuit for arbitrary labels in few-shot sentence classification, revealing the mechanism of ICL.
ICL in this setting is a gated associative lookup, built from a few identifiable components, with causality traced from demonstrations to predictions.
All Facts (3)
The cost of ICL is that every prediction must carry examples, and the longer the context, the higher the inference cost; the selection and ordering of examples affect performance; scenarios with extremely large data volumes or very high feature dimensions still require sampling and retrieval strategies.
50%UnverifiedRecent research provides the complete circuit for arbitrary labels in few-shot sentence classification, revealing the mechanism of ICL.
50%UnverifiedICL in this setting is a gated associative lookup, built from a few identifiable components, with causality traced from demonstrations to predictions.
50%