Summary of The Representation Landscape Of Few-shot Learning and Fine-tuning in Large Language Models, by Diego Doimo et al.
The representation landscape of few-shot learning and fine-tuning in large language models
by Diego Doimo, Alessandro Serra, Alessio Ansuini, Alberto Cazzaniga
First submitted to arxiv on: 5 Sep 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary Medium Difficulty summary: This paper investigates the internal workings of large language models (LLMs) when learning specific tasks using in-context learning (ICL) and supervised fine-tuning (SFT). The authors compare how LLMs solve question-answering tasks, revealing that ICL and SFT create distinct internal structures. During the first half of the network, ICL shapes interpretable representations hierarchically organized by semantic content. In contrast, SFT leads to fuzzier, semantically mixed representations. As the model progresses, fine-tuned representations develop modes better encoding answer identities, while ICL representations exhibit less defined peaks. This research sheds light on how LLMs compute optimal methods for extracting information and designing strategies to extract knowledge from language models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Low Difficulty summary: Imagine you have a super smart computer that can answer questions about anything. It’s called a large language model. Scientists want to know how this computer learns new things. They tried two ways: one where the computer just figures it out, and another where humans help guide it. The scientists looked inside the computer to see what was happening when it learned these new skills. They found that the two methods created different “maps” in the computer’s brain, showing how it organized information differently. This helps us understand how computers learn and can design better ways to use them. |
Keywords
» Artificial intelligence » Fine tuning » Large language model » Question answering » Supervised
