mlsanityData to 14 Oct 2026

Emergence of a High-Dimensional Abstraction Phase in Language Transformers

arXiv 2405.15471 ↗First posted 24 May 2024cs.CLcited by 56

Not on the map. Why?

Abstract (arXiv)

A language model (LM) is a mapping from a linguistic context to an output token. However, much remains to be known about this mapping, including how its geometric properties relate to its function. We take a high-level geometric approach to its analysis, observing, across five pre-trained transformer-based LMs and three input datasets, a distinct phase characterized by high intrinsic dimensionality. During this phase, representations (1) correspond to the first full linguistic abstraction of the input; (2) are the first to viably transfer to downstream tasks; (3) predict each other across different LMs. Moreover, we find that an earlier onset of the phase strongly predicts better language modelling performance. In short, our results suggest that a central high-dimensionality phase underlies core linguistic processing in many common LM architectures.

Key facts. Cited by 56 papers in the map (data through 14 Oct 2026). Builds on 43 of its 63 references that are in the map. First posted on arXiv 24 May 2024.

Builds on

43 of 63 references.

References · 43

  1. 2017Attention Is All You Needcited by 60,361
  2. 2019PyTorch: An Imperative Style, High-Performance Deep Learning Librarycited by 20,305
  3. 2022FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awarenesscited by 3,673
  4. 2016Gaussian Error Linear Units (GELUs)cited by 3,502
  5. 2016Pointer Sentinel Mixture Modelscited by 3,096
  6. 2022Locating and Editing Factual Associations in GPTcited by 2,691
  7. 2020The Pile: An 800GB Dataset of Diverse Text for Language Modelingcited by 2,411
  8. 2020GLU Variants Improve Transformercited by 1,804
  9. 2023Pythia: A Suite for Analyzing Large Language Models Across Training and Scalingcited by 1,786
  10. 2019Similarity of Neural Network Representations Revisitedcited by 1,755
  11. 2023GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpointscited by 1,346
  12. 2015Aligning Books and Movies: Towards Story-like Visual Explanations by Watching Movies and Reading Bookscited by 1,043
31 more
  1. 2023Towards Automated Circuit Discovery for Mechanistic Interpretabilitycited by 724
  2. 2020Intrinsic Dimensionality Explains the Effectiveness of Language Model Fine-Tuningcited by 671
  3. 2019Designing and Interpreting Probes with Control Taskscited by 640
  4. 2023Dissecting Recall of Factual Associations in Auto-Regressive Language Modelscited by 528
  5. 2017Opening the Black Box of Deep Neural Networks via Informationcited by 503
  6. 2024Dolma: an Open Corpus of Three Trillion Tokens for Language Model Pretraining Researchcited by 483
  7. 2018What you can cram into a single vector: Probing sentence embeddings for linguistic propertiescited by 457
  8. 2024ShortGPT: Layers in Large Language Models are More Redundant Than You Expectcited by 383
  9. 2018Measuring the Intrinsic Dimension of Objective Landscapescited by 370
  10. 2021The Intrinsic Dimension of Images and Its Impact on Learningcited by 315
  11. 2023In-Context Learning Creates Task Vectorscited by 312
  12. 2019Intrinsic dimension of data representations in deep neural networkscited by 282
  13. 2018Estimating the intrinsic dimension of datasets by a minimal neighborhood informationcited by 250
  14. 2024The Unreasonable Ineffectiveness of the Deeper Layerscited by 226
  15. 2023Language Modeling Is Compressioncited by 225
  16. 2023Getting aligned on representational alignmentcited by 143
  17. 2023The geometry of hidden representations of large transformer modelscited by 112
  18. 2023Intrinsic Dimension Estimation for Robust Detection of AI-Generated Textscited by 112
  19. 2024A Primer on the Inner Workings of Transformer-based Language Modelscited by 85
  20. 2023White-Box Transformers via Sparse Rate Reductioncited by 80
  21. 2017Residual Connections Encourage Iterative Inferencecited by 77
  22. 2021On Transferability of Prompt Tuning for Natural Language Processingcited by 75
  23. 2017Classification and Geometry of General Perceptual Manifoldscited by 62
  24. 2019Distributional Semantics and Linguistic Theorycited by 53
  25. 2024Characterizing Truthfulness in Large Language Model Generations with Local Intrinsic Dimensioncited by 45
  26. 2020Hierarchical nucleation in deep neural networkscited by 31
  27. 2023Bridging Information-Theoretic and Geometric Compression in Language Modelscited by 26
  28. 2021Ranking the information content of distance measurescited by 22
  29. 2023Fine-tuning Happens in Tiny Subspaces: Exploring Intrinsic Task-specific Subspaces of Pre-trained Language Modelscited by 19
  30. 2023Representation Learning via Manifold Flattening and Reconstructioncited by 11
  31. 2024Neural population geometry and optimal coding of tasks with shared latent structurecited by 9

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2024: 9; 2025: 19; 2026: 28.

56 papers.

Sort:

By area:

  1. 2025Layer by Layer: Uncovering Hidden Representations in Language Modelscited by 311
  2. 2025Latent Thinking Optimization: Your Latent Reasoning Language Model Secretly Encodes Reward Signals in Its Latent Thoughtscited by 19
  3. 2025The Intrinsic Dimension of Prompts in Internal Representations of Large Language Modelscited by 17
  4. 2024Geometric Signatures of Compositionality Across a Language Model's Lifetimecited by 16
  5. 2024The representation landscape of few-shot learning and fine-tuning in large language modelscited by 15
  6. 2024Evidence from fMRI Supports a Two-Phase Abstraction Process in Language Modelscited by 13
  7. 2024Persistent Topological Features in Large Language Modelscited by 12
  8. 2025Layers at Similar Depths Generate Similar Activations Across LLM Architecturescited by 12
  9. 2025Decoding Emotion in the Deep: A Systematic Study of How LLMs Represent, Retain, and Express Emotioncited by 12
  10. 2025Representation Potentials of Foundation Models for Multimodal Alignment: A Surveycited by 11
  11. 2025Geometry of Decision Making in Language Modelscited by 9
  12. 2025A quantitative analysis of semantic information in deep representations of text and imagescited by 7
38 more
  1. 2026Differential syntactic and semantic encoding in LLMscited by 7
  2. 2025Model Internal Sleuthing: Finding Lexical Identity and Inflectional Features in Modern Language Modelscited by 5
  3. 2025From Topology to Retrieval: Decoding Embedding Spaces with Unified Signaturescited by 5
  4. 2024Training the Untrainable: Introducing Inductive Bias via Representational Alignmentcited by 4
  5. 2024The Narrow Gate: Localized Image-Text Communication in Native Multimodal Modelscited by 4
  6. 2025Prediction hubs are context-informed frequent tokens in LLMscited by 3
  7. 2025LF2AR: Accounting for Layerwise Dynamics to Improve Multimodal Adaptation of Language Modelscited by 3
  8. 2025TRACE for Tracking the Emergence of Semantic Representations in Transformerscited by 3
  9. 2025Controlling the Risk of Corrupted Contexts for Language Models via Early-Exitingcited by 3
  10. 2026Tracing the complexity profiles of different linguistic phenomena through the intrinsic dimension of LLM representationscited by 3
  11. 2026Disentangling Geometry, Performance, and Training in Language Modelscited by 3
  12. 2026Uncovering the Latent Potential of Deep Intermediate Representationscited by 3
  13. 2026Automatic Layer Selection for Hallucination Detectioncited by 3
  14. 2024Unsupervised detection of semantic correlations in big datacited by 2
  15. 2025The Generalization Ridge: Information Flow in Natural Language Generationcited by 2
  16. 2025Towards Understanding the Shape of Representations in Protein Language Modelscited by 2
  17. 2026GenCtrl -- A Formal Controllability Toolkit for Generative Modelscited by 2
  18. 2026Abstraction Induces the Brain Alignment of Language and Speech Modelscited by 2
  19. 2026Sampling Data with Chains of Forward-Backward Diffusion Stepscited by 2
  20. 2026Where Do Models Find Happiness? Emotion Vectors in Open-Source LLMscited by 2
  21. 2024Understanding Variational Autoencoders with Intrinsic Dimension and Information Imbalancecited by 1
  22. 2025Emergent effects of scaling on the functional hierarchies within large language modelscited by 1
  23. 2026Pseudo-Unification: Entropy Probing Reveals Divergent Information Patterns in Unified Multimodal Modelscited by 1
  24. 2026On the Persistent Effects of Lexicality in Large Language Modelscited by 1
  25. 2024Absolute abstraction: a renormalisation group approachcited by 0
  26. 2025Probing Geometry of Next Token Prediction Using Cumulant Expansion of the Softmax Entropycited by 0
  27. 2025Navigating through the hidden embedding space: steering LLMs to improve mental health assessmentcited by 0
  28. 2026Emergent Causal-Geometric Dynamics Across Depth in Large Language Modelscited by 0
  29. 2026A High-Resolution Landscape Dataset for Concept-Based XAI With Application to Species Distribution Modelscited by 0
  30. 2026Rethinking Intrinsic Dimension Estimation in Neural Representationscited by 0
  31. 2026Causal Dimensionality of Transformer Representations: Measurement, Scaling, and Layer Structurecited by 0
  32. 2026Relational Linear Properties in Language Models: An Empirical Investigationcited by 0
  33. 2026Cordyceps: Covert Control Attacks on LLMs via Data Poisoningcited by 0
  34. 2026Visual Instruction Tuning Aligns Modalities through Abstractioncited by 0
  35. 2026Riemannian Geometry for Pre-trained Language Model Embeddingscited by 0
  36. 2026What Makes Linguistic Representations Good Models of High-Level Visual Perception in the Human Brain?cited by 0
  37. 2026Isotropy Cliffs: The Geometric Signature of Decision-Making in Large Language Modelscited by 0
  38. 2026Local and Global Regimes of Geometric Complexity in Language Model Representationscited by 0

50 most cited of 56 All 56 (JSON)