Do Natural Language Descriptions of Model Activations Convey Privileged Information?

• URL: http://arxiv.org/abs/2509.13316v2
• Date: Mon, 29 Sep 2025 16:57:58 GMT
• Title: Do Natural Language Descriptions of Model Activations Convey Privileged Information?
• Authors: Millicent Li, Alberto Mario Ceballos Arroyo, Giordano Rogers, Naomi Saphra, Byron C. Wallace,
• Abstract summary: We critically evaluate popular verbalization methods across datasets used in prior work.<n>We run controlled experiments which reveal that verbalizations often reflect the parametric knowledge of the verbalizer LLM.
• Score: 17.883814422888264
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Recent interpretability methods have proposed to translate LLM internal representations into natural language descriptions using a second verbalizer LLM. This is intended to illuminate how the target model represents and operates on inputs. But do such activation verbalization approaches actually provide privileged knowledge about the internal workings of the target model, or do they merely convey information about its inputs? We critically evaluate popular verbalization methods across datasets used in prior work and find that they can succeed at benchmarks without any access to target model internals, suggesting that these datasets may not be ideal for evaluating verbalization methods. We then run controlled experiments which reveal that verbalizations often reflect the parametric knowledge of the verbalizer LLM which generated them, rather than the knowledge of the target LLM whose activations are decoded. Taken together, our results indicate a need for targeted benchmarks and experimental controls to rigorously assess whether verbalization methods provide meaningful insights into the operations of LLMs.

Related papers

• Natural Language-based Assessment of L2 Oral Proficiency using LLMs [5.931245606235828]
  Natural language-based assessment (NLA) is an approach to second language assessment that uses instructions expressed in the form of can-do descriptors.<n>In this work, we explore the use of such descriptors with an open-source LLM, Qwen 2.5 72B, to assess responses from the publicly available S&I Corpus.<n>Our results show that this approach achieves competitive performance: while it does not outperform state-of-the-art speech LLMs fine-tuned for the task, it surpasses a BERT-based model trained specifically for this purpose.
  arXiv  Detail & Related papers  (2025-07-14T12:13:50Z)
• Language Bottleneck Models: A Framework for Interpretable Knowledge Tracing and Beyond [55.984684518346924]
  We recast Knowledge Tracing as an inverse problem: learning the minimum natural-language summary that makes past answers explainable and future answers predictable.<n>Our Language Bottleneck Model (LBM) consists of an encoder LLM that writes an interpretable knowledge summary and a frozen decoder LLM that must reconstruct and predict student responses using only that summary text.<n> Experiments on synthetic arithmetic benchmarks and the large-scale Eedi dataset show that LBMs rival the accuracy of state-of-the-art KT and direct LLM methods while requiring orders-of-magnitude fewer student trajectories.
  arXiv  Detail & Related papers  (2025-06-20T13:21:14Z)
• MoRE-LLM: Mixture of Rule Experts Guided by a Large Language Model [54.14155564592936]
  We propose a Mixture of Rule Experts guided by a Large Language Model (MoRE-LLM)<n>MoRE-LLM steers the discovery of local rule-based surrogates during training and their utilization for the classification task.<n>LLM is responsible for enhancing the domain knowledge alignment of the rules by correcting and contextualizing them.
  arXiv  Detail & Related papers  (2025-03-26T11:09:21Z)
• Interpreting and Steering LLMs with Mutual Information-based Explanations on Sparse Autoencoders [29.356200147371275]
  Large language models (LLMs) excel at handling human queries, but they can occasionally generate flawed or unexpected responses.<n>We propose using a fixed vocabulary set for feature interpretations and designing a mutual information-based objective.<n>We propose two runtime steering strategies that adjust the learned feature activations based on their corresponding explanations.
  arXiv  Detail & Related papers  (2025-02-21T16:36:42Z)
• LatentQA: Teaching LLMs to Decode Activations Into Natural Language [72.87064562349742]
  We introduce LatentQA, the task of answering open-ended questions about model activations in natural language.<n>We propose Latent Interpretation Tuning (LIT), which finetunes a decoder LLM on a dataset of activations and associated question-answer pairs.<n>Our decoder also specifies a differentiable loss that we use to control models, such as debiasing models on stereotyped sentences and controlling the sentiment of generations.
  arXiv  Detail & Related papers  (2024-12-11T18:59:33Z)
• Traffic Light or Light Traffic? Investigating Phrasal Semantics in Large Language Models [41.233879429714925]
  This study critically examines the capacity of API-based large language models to comprehend phrase semantics. We assess the performance of LLMs in executing phrase semantic reasoning tasks guided by natural language instructions. We conduct detailed error analyses to interpret the limitations faced by LLMs in comprehending phrase semantics.
  arXiv  Detail & Related papers  (2024-10-03T08:44:17Z)
• $\forall$uto$\exists$val: Autonomous Assessment of LLMs in Formal Synthesis and Interpretation Tasks [21.12437562185667]
  This paper presents a new approach for scaling LLM assessment in translating formal syntax to natural language. We use context-free grammars (CFGs) to generate out-of-distribution datasets on the fly. We also conduct an assessment of several SOTA closed and open-source LLMs to showcase the feasibility and scalability of this paradigm.
  arXiv  Detail & Related papers  (2024-03-27T08:08:00Z)
• Characterizing Truthfulness in Large Language Model Generations with Local Intrinsic Dimension [63.330262740414646]
  We study how to characterize and predict the truthfulness of texts generated from large language models (LLMs) We suggest investigating internal activations and quantifying LLM's truthfulness using the local intrinsic dimension (LID) of model activations.
  arXiv  Detail & Related papers  (2024-02-28T04:56:21Z)
• Generative Context-aware Fine-tuning of Self-supervised Speech Models [54.389711404209415]
  We study the use of generative large language models (LLM) generated context information. We propose an approach to distill the generated information during fine-tuning of self-supervised speech models. We evaluate the proposed approach using the SLUE and Libri-light benchmarks for several downstream tasks: automatic speech recognition, named entity recognition, and sentiment analysis.
  arXiv  Detail & Related papers  (2023-12-15T15:46:02Z)
• Let Models Speak Ciphers: Multiagent Debate through Embeddings [84.20336971784495]
  We introduce CIPHER (Communicative Inter-Model Protocol Through Embedding Representation) to address this issue. By deviating from natural language, CIPHER offers an advantage of encoding a broader spectrum of information without any modification to the model weights. This showcases the superiority and robustness of embeddings as an alternative "language" for communication among LLMs.
  arXiv  Detail & Related papers  (2023-10-10T03:06:38Z)
• On Learning to Summarize with Large Language Models as References [101.79795027550959]
  Large language models (LLMs) are favored by human annotators over the original reference summaries in commonly used summarization datasets. We study an LLM-as-reference learning setting for smaller text summarization models to investigate whether their performance can be substantially improved.
  arXiv  Detail & Related papers  (2023-05-23T16:56:04Z)
• The Internal State of an LLM Knows When It's Lying [18.886091925252174]
  Large Language Models (LLMs) have shown exceptional performance in various tasks. One of their most prominent drawbacks is generating inaccurate or false information with a confident tone. We provide evidence that the LLM's internal state can be used to reveal the truthfulness of statements.
  arXiv  Detail & Related papers  (2023-04-26T02:49:38Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.