IC-1088ICL predictions in LLaMA, LLaMA-2, and Falcon models depend on in-context label information and can learn truly novel label relationships

Jannik Kossen, Yarin Gal, Tom Rainforth

SourceIn-Context Learning Learns Label Relationships but Is Not Conventional Learning

The paper tests whether ICL predictions are independent of the conditional label distribution of in-context examples (NH1). By randomizing in-context labels, they show that log likelihoods and entropies degrade significantly across all 10 model variants and 9 tasks, rejecting NH1. They further demonstrate that ICL can learn a completely novel author-identification task (private Slack messages between two authors) that is guaranteed absent from pre-training data, with all models except Falcon-7B-Instruct achieving better-than-random accuracy. This confirms ICL genuinely uses and learns from in-context label information rather than merely recognizing task format.

Evidence
correlational
Key metric
Δ log likelihood (default minus random) at max context size for llama-2 70b: 0.51 (SST-2), 0.53 (Subj), 0.57 (FP), 0.34 (HS), 0.80 (AGN), 0.29 (MQP), 0.04 (MRPC), 0.22 (RTE), 0.18 (WNLI); for falcon 40b: 0.34, 0.35, 0.31, 0.18, 0.90, 0.06, 0.01, 0.01, 0.02
Caveat
On entailment tasks (MQP, MRPC, RTE, WNLI) smaller models struggle to reach better-than-random performance on default labels, limiting the scope of the randomization comparison for those tasks. The novel author-identification dataset is not publicly released.
Model
Llama 2 / Llama 2 base, LLaMA, Falcon
Datasets
SST-2 [eval], Subjective [eval], Financial Phrasebank [eval], AG-News [eval], MQP [eval], MRPC [eval], RTE [eval], WNLI [eval]
Related work
Min et al. 2022b (Rethinking the Role of Demonstrations) [compared-to], Xie et al. 2022 (An Explanation of In-Context Learning as Implicit Bayesian Inference) [compared-to]
Related findings
IC-1089, IC-1090
Extraction
automatic-extraction