Provably Learning from Modern Language Models via Low Logit Rank

• URL: http://arxiv.org/abs/2512.09892v1
• Date: Wed, 10 Dec 2025 18:18:11 GMT
• Title: Provably Learning from Modern Language Models via Low Logit Rank
• Authors: Noah Golowich, Allen Liu, Abhishek Shetty,
• Abstract summary: Low logit rank models can encode hard-to-learn distributions such as noisy parities.<n>We show how this structure can be exploited algorithmically for obtaining provable learning guarantees.<n>Our result gives what we believe is the first end-to-end learning guarantee for a generative model that plausibly captures modern language models.
• Score: 22.148282143726835
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: While modern language models and their inner workings are incredibly complex, recent work (Golowich, Liu & Shetty; 2025) has proposed a simple and potentially tractable abstraction for them through the observation that empirically, these language models all seem to have approximately low logit rank. Roughly, this means that a matrix formed by the model's log probabilities of various tokens conditioned on certain sequences of tokens is well approximated by a low rank matrix. In this paper, our focus is on understanding how this structure can be exploited algorithmically for obtaining provable learning guarantees. Since low logit rank models can encode hard-to-learn distributions such as noisy parities, we study a query learning model with logit queries that reflects the access model for common APIs. Our main result is an efficient algorithm for learning any approximately low logit rank model from queries. We emphasize that our structural assumption closely reflects the behavior that is empirically observed in modern language models. Thus, our result gives what we believe is the first end-to-end learning guarantee for a generative model that plausibly captures modern language models.

Related papers

• Sequences of Logits Reveal the Low Rank Structure of Language Models [22.148282143726835]
  We introduce an approach to study the low-dimensional structure of language models at a model-agnostic level.<n>We first empirically demonstrate that a wide range of modern language models exhibit low-rank structure.<n>We then show that this low-rank structure can be leveraged for generation.
  arXiv  Detail & Related papers  (2025-10-28T20:55:58Z)
• Pretraining Language Models to Ponder in Continuous Space [50.52734567589996]
  We introduce this pondering process into language models by repeatedly invoking the forward process within a single token generation step.<n>We show that the model can learn to ponder in this way through self-supervised learning, without any human annotations.
  arXiv  Detail & Related papers  (2025-05-27T03:47:33Z)
• Model Stealing for Any Low-Rank Language Model [25.16701867917684]
  We build a theoretical understanding of stealing language models by studying a simple and mathematically tractable setting. Our main result is an efficient algorithm in the conditional query model, for learning any low-rank distribution. This is an interesting example where, at least theoretically, allowing a machine learning model to solve more complex problems at inference time can lead to drastic improvements in its performance.
  arXiv  Detail & Related papers  (2024-11-12T04:25:31Z)
• Promises and Pitfalls of Generative Masked Language Modeling: Theoretical Framework and Practical Guidelines [74.42485647685272]
  We focus on Generative Masked Language Models (GMLMs) We train a model to fit conditional probabilities of the data distribution via masking, which are subsequently used as inputs to a Markov Chain to draw samples from the model. We adapt the T5 model for iteratively-refined parallel decoding, achieving 2-3x speedup in machine translation with minimal sacrifice in quality.
  arXiv  Detail & Related papers  (2024-07-22T18:00:00Z)
• MiLe Loss: a New Loss for Mitigating the Bias of Learning Difficulties in Generative Language Models [40.992566245706996]
  We propose a MiLe Loss function for mitigating the bias of learning difficulties with tokens. We train generative language models at different scales of 468M, 1.2B, and 6.7B parameters. Experiments reveal that models incorporating the proposed MiLe Loss can gain consistent performance improvement on downstream benchmarks.
  arXiv  Detail & Related papers  (2023-10-30T13:33:21Z)
• Language models are weak learners [71.33837923104808]
  We show that prompt-based large language models can operate effectively as weak learners. We incorporate these models into a boosting approach, which can leverage the knowledge within the model to outperform traditional tree-based boosting. Results illustrate the potential for prompt-based LLMs to function not just as few-shot learners themselves, but as components of larger machine learning pipelines.
  arXiv  Detail & Related papers  (2023-06-25T02:39:19Z)
• Interpretability at Scale: Identifying Causal Mechanisms in Alpaca [62.65877150123775]
  We use Boundless DAS to efficiently search for interpretable causal structure in large language models while they follow instructions. Our findings mark a first step toward faithfully understanding the inner-workings of our ever-growing and most widely deployed language models.
  arXiv  Detail & Related papers  (2023-05-15T17:15:40Z)
• CodeGen2: Lessons for Training LLMs on Programming and Natural Languages [116.74407069443895]
  We unify encoder and decoder-based models into a single prefix-LM. For learning methods, we explore the claim of a "free lunch" hypothesis. For data distributions, the effect of a mixture distribution and multi-epoch training of programming and natural languages on model performance is explored.
  arXiv  Detail & Related papers  (2023-05-03T17:55:25Z)
• A Natural Bias for Language Generation Models [31.44752136404971]
  We show that we can endow standard neural language generation models with a separate module that reflects unigram frequency statistics as prior knowledge. We use neural machine translation as a test bed for this simple technique and observe that it: (i) improves learning efficiency; (ii) achieves better overall performance; and perhaps most importantly: appears to disentangle strong frequency effects.
  arXiv  Detail & Related papers  (2022-12-19T18:14:36Z)
• Language Model Pre-Training with Sparse Latent Typing [66.75786739499604]
  We propose a new pre-training objective, Sparse Latent Typing, which enables the model to sparsely extract sentence-level keywords with diverse latent types. Experimental results show that our model is able to learn interpretable latent type categories in a self-supervised manner without using any external knowledge.
  arXiv  Detail & Related papers  (2022-10-23T00:37:08Z)
• Quark: Controllable Text Generation with Reinforced Unlearning [68.07749519374089]
  Large-scale language models often learn behaviors that are misaligned with user expectations. We introduce Quantized Reward Konditioning (Quark), an algorithm for optimizing a reward function that quantifies an (un)wanted property. For unlearning toxicity, negative sentiment, and repetition, our experiments show that Quark outperforms both strong baselines and state-of-the-art reinforcement learning methods.
  arXiv  Detail & Related papers  (2022-05-26T21:11:51Z)
• Low-Rank Softmax Can Have Unargmaxable Classes in Theory but Rarely in Practice [18.296971636710985]
  We develop algorithms to detect emphunargmaxable tokens public language models. We find that 13 out of 150 models do indeed have such tokens; however, they are very infrequent and unlikely to impact model quality.
  arXiv  Detail & Related papers  (2022-03-12T15:34:54Z)

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.