Summary of I’ve Got the “answer”! Interpretation Of Llms Hidden States in Question Answering, by Valeriya Goloviznina and Evgeny Kotelnikov

I’ve got the “Answer”! Interpretation of LLMs Hidden States in Question Answering

by Valeriya Goloviznina, Evgeny Kotelnikov

First submitted to arxiv on: 4 Jun 2024

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 A novel investigation into the interpretability and explainability of large language models (LLMs) is presented, focusing on their behavior in knowledge-based question answering. The study hypothesizes that correct and incorrect model behavior can be distinguished at the level of hidden states using quantized models LLaMA-2-7B-Chat, Mistral-7B, Vicuna-7B, and the MuSeRC question-answering dataset. The results support this hypothesis, identifying specific layers with negative effects on model behavior. As a potential application, training these “weak” layers to improve task solution quality is proposed.
Low	GrooveSquid.com (original content)	Low Difficulty Summary Large language models are getting really good at answering questions! This research helps us understand how they make decisions by looking at what’s going on inside the model. It finds that correct and incorrect answers can be told apart by studying the hidden states of the model. The study uses four special models and a big dataset to test this idea, and it works! The results also show which parts of the model are causing problems and how we could fix them.