I Predict Therefore I Am: Is Next Token Prediction Enough to Learn Human-Interpretable Concepts from Data?

• URL: http://arxiv.org/abs/2503.08980v3
• Date: Mon, 12 May 2025 10:45:23 GMT
• Title: I Predict Therefore I Am: Is Next Token Prediction Enough to Learn Human-Interpretable Concepts from Data?
• Authors: Yuhang Liu, Dong Gong, Yichao Cai, Erdun Gao, Zhen Zhang, Biwei Huang, Mingming Gong, Anton van den Hengel, Javen Qinfeng Shi,
• Abstract summary: Large language models (LLMs) have led many to conclude that they exhibit a form of intelligence.<n>We introduce a novel generative model that generates tokens on the basis of human-interpretable concepts represented as latent discrete variables.
• Score: 76.15163242945813
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: The remarkable achievements of large language models (LLMs) have led many to conclude that they exhibit a form of intelligence. This is as opposed to explanations of their capabilities based on their ability to perform relatively simple manipulations of vast volumes of data. To illuminate the distinction between these explanations, we introduce a novel generative model that generates tokens on the basis of human-interpretable concepts represented as latent discrete variables. Under mild conditions, even when the mapping from the latent space to the observed space is non-invertible, we establish an identifiability result, i.e., the representations learned by LLMs through next-token prediction can be approximately modeled as the logarithm of the posterior probabilities of these latent discrete concepts given input context, up to an invertible linear transformation. This theoretical finding not only provides evidence that LLMs capture underlying generative factors, but also provide a unified prospective for understanding of the linear representation hypothesis. Taking this a step further, our finding motivates a reliable evaluation of sparse autoencoders by treating the performance of supervised concept extractors as an upper bound. Pushing this idea even further, it inspires a structural variant that enforces dependence among latent concepts in addition to promoting sparsity. Empirically, we validate our theoretical results through evaluations on both simulation data and the Pythia, Llama, and DeepSeek model families, and demonstrate the effectiveness of our structured sparse autoencoder.

Related papers

• Counterfactual Explanations in Medical Imaging: Exploring SPN-Guided Latent Space Manipulation [2.9810923705287524]
  In medical image analysis, deep learning models have demonstrated remarkable performance.<n>Deep generative models such as variational autoencoders (VAEs) exhibit significant generative power.<n>Probability models like sum-product networks (SPNs) efficiently represent complex joint probability distributions.
  arXiv  Detail & Related papers  (2025-07-25T15:19:32Z)
• From Flat to Hierarchical: Extracting Sparse Representations with Matching Pursuit [16.996218963146788]
  We show that MP-SAE unrolls its encoder into a sequence of residual-guided steps, allowing it to capture hierarchical and nonlinearly accessible features.<n>We also show that the sequential encoder principle of MP-SAE affords an additional benefit of adaptive sparsity at inference time.
  arXiv  Detail & Related papers  (2025-06-03T17:24:55Z)
• Understanding LLM Behaviors via Compression: Data Generation, Knowledge Acquisition and Scaling Laws [5.685201910521295]
  We offer a detailed view of how Large Language Models acquire and store information across increasing model and data scales. Motivated by this theoretical perspective and natural assumptions inspired by Heap's and Zipf's laws, we introduce a simplified yet representative hierarchical data-generation framework. Under the Bayesian setting, we show that prediction and compression within this model naturally lead to diverse learning and scaling behaviors.
  arXiv  Detail & Related papers  (2025-04-13T14:31:52Z)
• Cross-Entropy Is All You Need To Invert the Data Generating Process [29.94396019742267]
  Empirical phenomena suggest that supervised models can learn interpretable factors of variation in a linear fashion.<n>Recent advances in self-supervised learning have shown that these methods can recover latent structures by inverting the data generating process.<n>We prove that even in standard classification tasks, models learn representations of ground-truth factors of variation up to a linear transformation.
  arXiv  Detail & Related papers  (2024-10-29T09:03:57Z)
• Learning Discrete Concepts in Latent Hierarchical Models [73.01229236386148]
  Learning concepts from natural high-dimensional data holds potential in building human-aligned and interpretable machine learning models.<n>We formalize concepts as discrete latent causal variables that are related via a hierarchical causal model.<n>We substantiate our theoretical claims with synthetic data experiments.
  arXiv  Detail & Related papers  (2024-06-01T18:01:03Z)
• On the Origins of Linear Representations in Large Language Models [51.88404605700344]
  We introduce a simple latent variable model to formalize the concept dynamics of the next token prediction. Experiments show that linear representations emerge when learning from data matching the latent variable model. We additionally confirm some predictions of the theory using the LLaMA-2 large language model.
  arXiv  Detail & Related papers  (2024-03-06T17:17:36Z)
• The Information of Large Language Model Geometry [3.4003124816653143]
  We conduct simulations to analyze the representation entropy and discover a power law relationship with model sizes. We propose a theory based on (conditional) entropy to elucidate the scaling law phenomenon.
  arXiv  Detail & Related papers  (2024-02-01T12:50:43Z)
• Interpreting Pretrained Language Models via Concept Bottlenecks [55.47515772358389]
  Pretrained language models (PLMs) have made significant strides in various natural language processing tasks. The lack of interpretability due to their black-box'' nature poses challenges for responsible implementation. We propose a novel approach to interpreting PLMs by employing high-level, meaningful concepts that are easily understandable for humans.
  arXiv  Detail & Related papers  (2023-11-08T20:41:18Z)
• Evaluating and Explaining Large Language Models for Code Using Syntactic Structures [74.93762031957883]
  This paper introduces ASTxplainer, an explainability method specific to Large Language Models for code. At its core, ASTxplainer provides an automated method for aligning token predictions with AST nodes. We perform an empirical evaluation on 12 popular LLMs for code using a curated dataset of the most popular GitHub projects.
  arXiv  Detail & Related papers  (2023-08-07T18:50:57Z)
• A Recursive Bateson-Inspired Model for the Generation of Semantic Formal Concepts from Spatial Sensory Data [77.34726150561087]
  This paper presents a new symbolic-only method for the generation of hierarchical concept structures from complex sensory data. The approach is based on Bateson's notion of difference as the key to the genesis of an idea or a concept. The model is able to produce fairly rich yet human-readable conceptual representations without training.
  arXiv  Detail & Related papers  (2023-07-16T15:59:13Z)
• Explaining Language Models' Predictions with High-Impact Concepts [11.47612457613113]
  We propose a complete framework for extending concept-based interpretability methods to NLP. We optimize for features whose existence causes the output predictions to change substantially. Our method achieves superior results on predictive impact, usability, and faithfulness compared to the baselines.
  arXiv  Detail & Related papers  (2023-05-03T14:48:27Z)
• Provable concept learning for interpretable predictions using variational inference [7.0349768355860895]
  In safety critical applications, practitioners are reluctant to trust neural networks when no interpretable explanations are available. We propose a probabilistic modeling framework to derive (C)oncept (L)earning and (P)rediction (CLAP) We prove that our method is able to identify them while attaining optimal classification accuracy.
  arXiv  Detail & Related papers  (2022-04-01T14:51:38Z)
• Generative Counterfactuals for Neural Networks via Attribute-Informed Perturbation [51.29486247405601]
  We design a framework to generate counterfactuals for raw data instances with the proposed Attribute-Informed Perturbation (AIP) By utilizing generative models conditioned with different attributes, counterfactuals with desired labels can be obtained effectively and efficiently. Experimental results on real-world texts and images demonstrate the effectiveness, sample quality as well as efficiency of our designed framework.
  arXiv  Detail & Related papers  (2021-01-18T08:37:13Z)
• Interpretable Representations in Explainable AI: From Theory to Practice [7.031336702345381]
  Interpretable representations are the backbone of many explainers that target black-box predictive systems. We study properties of interpretable representations that encode presence and absence of human-comprehensible concepts.
  arXiv  Detail & Related papers  (2020-08-16T21:44:03Z)
• Leap-Of-Thought: Teaching Pre-Trained Models to Systematically Reason Over Implicit Knowledge [96.92252296244233]
  Large pre-trained language models (LMs) acquire some reasoning capacity, but this ability is difficult to control. We show that LMs can be trained to reliably perform systematic reasoning combining both implicit, pre-trained knowledge and explicit natural language statements. Our work paves a path towards open-domain systems that constantly improve by interacting with users who can instantly correct a model by adding simple natural language statements.
  arXiv  Detail & Related papers  (2020-06-11T17:02:20Z)

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.