Inference-Time Adaptation

ICL

In brief — site editorial, not from the paper

Shows the model a few worked examples in the prompt, adapting it for that request only, with no training and no artifact left behind.

Definition

ICL leverages a deployed LLM's pattern-matching capability to perform task adaptation from a small number of input-output demonstrations placed in the context window, with no gradient computation and no parameter modification ^[34]. It shares a common structural mechanism—D1 (Mechanism), D4 (Persistence), D5 (Scope), D6 (Model Type)—with PE, but fundamentally differs on D3 (Data Requirements)—the inclusion of Few Demonstrations—which shifts its primary D2 (Goal) focus toward Task Specialization. From a system validation perspective, because ICL leaves the foundational model parameters entirely unchanged, it triggers no artifact-level regression testing. Instead, the specific demonstration sets act as transient input dependencies and must be documented.

Verbatim from the paper — Inference-Time Adaptation

Notes from the table

Classification tensions involving these techniques are discussed in Supplementary Section S4.

Related techniques

supersession Meta-Learning — ICL supersedes gradient-based meta-learning for few-shot adaptation at the LLM tier
supersession is mechanistically grounded: LLM contain sufficiently rich pre-trained representations Appendix C, Training Strategies
sub-technique FSL — Colloquial "few-shot" splits here: FSL updates weights, ICL does not
underlying model artifact unmodified; this is reassigned to ICL Appendix C, Knowledge Transfer and Task Specialization

Nearest profiles

APO (0.36), PE (0.36), RAG (0.42), CE (0.44), Test-Time Compute Scaling (0.50)

Computed from the taxonomy data (Gower distance over all six dimensions)

References

[34] Tom B. Brown et al. 2020. Language models are few-shot learners. Proceedings of the 34th International Conference on Neural Information Processing Systems

Numbered as in the paper

Cite this row

ICL — six-dimensional profile
D1: context-injection
D2: knowledge-update, task-specialization
D3: few-demonstrations
D4: session-ephemeral
D5: input-output-space
D6: llm, mllm
Source: arXiv:2608.06246

Compare with another technique Find in the explorer