ML Newsnew | past | best | rss | submit about
Is In-Context Learning Sufficient for Instruction Following in LLMs? (arxiv.org)
2 points by max-andr on Jun 1, 2024 | hide | past | pdf | 1 comment on HN

In plain words: Instead of retraining a chatbot to follow instructions, this study tests teaching it by placing a few example conversations in its prompt. Three examples still trail retraining, but tuning how the model picks words and adding carefully chosen examples closes much of the gap.

Abstract

In-context learning (ICL) allows LLMs to learn from examples without changing their weights: this is a particularly promising capability for long-context LLMs that can potentially learn from many examples. Recently, Lin et al. (2024) proposed URIAL, a method using only three in-context examples to align base LLMs, achieving non-trivial instruction following performance. In this work, we show that, while effective, ICL alignment with URIAL still underperforms compared to instruction fine-tuning on the established benchmark MT-Bench, especially with more capable base LLMs. We then uncover the most relevant elements for successful in-context alignment, finding the crucial role of the decoding parameters. Based on these insights, we show that the approach of URIAL can indeed be improved by adding high-quality, potentially carefully selected via greedy search, demonstrations in context, getting closer to the performance of instruct models. Finally, we provide the first, to our knowledge, systematic comparison of ICL and instruction fine-tuning (IFT) for instruction following in the low data regime, where ICL can be a viable alternative to IFT. Overall, our work advances the understanding of ICL as an alignment technique and its relationship to IFT. We provide our code at https://github.com/tml-epfl/icl-alignment.

Hao Zhao, Maksym Andriushchenko, Francesco Croce, Nicolas Flammarion
arXiv:2405.19874 · cs.CL, cs.AI, cs.LG · submitted May 30, 2024 · updated Apr 18, 2025
abstract · pdf · html · Accepted at ICLR 2025. This camera-ready version v3 adds multi-turn alignment via ICL, revisiting main results on instruct models, and simple mechanistic study. Updates in the v2: experiment with decoding schemes, scaling in-context alignment, ICL vs IFT for instruction following. Code at https://github.com/tml-epfl/icl-alignment

add comment on HN

max-andr on Jun 1, 2024 [–]
In-context learning (ICL) allows LLMs to learn from examples without changing their weights, which is a particularly promising capability for long-context LLMs that can potentially learn from many examples. Recently, Lin et al. (2024) proposed URIAL, a method using only three in-context examples to align base LLMs, achieving non-trivial instruction following performance. In this work, we show that, while effective, ICL alignment with URIAL still underperforms compared to instruction fine-tuning on established benchmarks such as MT-Bench and AlpacaEval 2.0 (LC), especially with more capable base LMs. Unlike for tasks such as classification, translation, or summarization, adding more ICL demonstrations for long-context LLMs does not systematically improve instruction following performance. To address this limitation, we derive a greedy selection approach for ICL examples that noticeably improves performance, yet without bridging the gap to instruction fine-tuning. Finally, we provide a series of ablation studies to better understand the reasons behind the remaining gap, and we show how some aspects of ICL depart from the existing knowledge and are specific to the instruction tuning setting. Overall, our work advances the understanding of ICL as an alignment technique.
reply
About | RSS | RSS (all) | HN arXiv