mlsanityData to 7 Oct 2026

Rethinking the Role of Scale for In-Context Learning: An Interpretability-based Case Study at 66 Billion Scale

arXiv 2212.09095 ↗First posted 18 Dec 2022cs.CL cs.AIcited by 64

On the map: LLM reasoning → Reasoning benchmarks and evaluation

Abstract (arXiv)

Language models have been shown to perform better with an increase in scale on a wide variety of tasks via the in-context learning paradigm. In this paper, we investigate the hypothesis that the ability of a large language model to in-context learn-perform a task is not uniformly spread across all of its underlying components. Using a 66 billion parameter language model (OPT-66B) across a diverse set of 14 downstream tasks, we find this is indeed the case: ~70% of attention heads and ~20% of feed forward networks can be removed with minimal decline in task performance. We find substantial overlap in the set of attention heads (un)important for in-context learning across tasks and number of in-context examples. We also address our hypothesis through a task-agnostic lens, finding that a small set of attention heads in OPT-66B score highly on their ability to perform primitive induction operations associated with in-context learning, namely, prefix matching and copying. These induction heads overlap with task-specific important heads, reinforcing arguments by Olsson et al. (arXiv:2209.11895) regarding induction head generality to more sophisticated behaviors associated with in-context learning. Overall, our study provides several insights that indicate large language models may be under-trained for in-context learning and opens up questions on how to pre-train language models to more effectively perform in-context learning.

Key facts. Cited by 64 papers in the map (data through 7 Oct 2026). Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning. Builds on 38 of its 50 references that are in the map. First posted on arXiv 18 Dec 2022.

Builds on

38 of 50 references.

Closest ancestors · 2

Same direction, newest first

  1. 2022Rethinking the Role of Demonstrations: What Makes In-Context Learning Work?25 Feb
  2. 2022Impact of Pretraining Term Frequencies on Few-Shot Reasoning15 Feb

Other references · 36

  1. 2017Attention Is All You Needcited by 59,932
  2. 2020Language Models are Few-Shot Learnerscited by 32,199
  3. 2020Scaling Laws for Neural Language Modelscited by 6,411
  4. 2022PaLM: Scaling Language Modeling with Pathwayscited by 5,304
  5. 2018Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challengecited by 4,710
  6. 2019HellaSwag: Can a Machine Really Finish Your Sentence?cited by 4,018
  7. 2022OPT: Open Pre-trained Transformer Language Modelscited by 3,464
  8. 2022Training Compute-Optimal Large Language Modelscited by 2,927
  9. 2019PIQA: Reasoning about Physical Commonsense in Natural Languagecited by 2,818
  10. 2022Emergent Abilities of Large Language Modelscited by 2,483
  11. 2018Can a Suit of Armor Conduct Electricity? A New Dataset for Open Book Question Answeringcited by 2,114
  12. 2019SuperGLUE: A Stickier Benchmark for General-Purpose Language Understanding Systemscited by 1,718
24 more
  1. 2021Calibrate Before Use: Improving Few-Shot Performance of Language Modelscited by 1,371
  2. 2021Scaling Language Models: Methods, Analysis & Insights from Training Gophercited by 1,199
  3. 2021What Makes Good In-Context Examples for GPT-3?cited by 1,184
  4. 2016Pruning Filters for Efficient ConvNetscited by 1,129
  5. 2021Fantastically Ordered Prompts and Where to Find Them: Overcoming Few-Shot Prompt Order Sensitivitycited by 1,122
  6. 2019BERT Rediscovers the Classical NLP Pipelinecited by 1,046
  7. 2016The LAMBADA dataset: Word prediction requiring a broad discourse contextcited by 960
  8. 2022In-context Learning and Induction Headscited by 930
  9. 2021An Explanation of In-context Learning as Implicit Bayesian Inferencecited by 915
  10. 2019Analyzing Multi-Head Self-Attention: Specialized Heads Do the Heavy Lifting, the Rest Can Be Prunedcited by 844
  11. 2019Are Sixteen Heads Really Better than One?cited by 803
  12. 2022GPT-NeoX-20B: An Open-Source Autoregressive Language Modelcited by 711
  13. 2022What Can Transformers Learn In-Context? A Case Study of Simple Function Classescited by 700
  14. 2019MathQA: Towards Interpretable Math Word Problem Solving with Operation-Based Formalismscited by 653
  15. 2021Learning To Retrieve Prompts for In-Context Learningcited by 641
  16. 2022Using DeepSpeed and Megatron to Train Megatron-Turing NLG 530B, A Large-Scale Generative Language Modelcited by 557
  17. 2021Surface Form Competition: Why the Highest Probability Answer Isn't Always Rightcited by 256
  18. 2019Analyzing the Structure of Attention in a Transformer Language Modelcited by 241
  19. 2021Reframing Instructional Prompts to GPTk's Languagecited by 176
  20. 2021Noisy Channel Language Model Prompting for Few-Shot Text Classificationcited by 163
  21. 2015Structured Pruning of Deep Convolutional Neural Networkscited by 148
  22. 2022Transformers generalize differently from information stored in context vs in weightscited by 49
  23. 2021Differentiable Subset Pruning of Transformer Headscited by 46
  24. 2022Data-Efficient Structured Pruning via Submodular Optimizationcited by 22

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2023: 15; 2024: 20; 2025: 20; 2026: 9.

64 papers.

Sort:

By area:

  1. 2023A Survey of Large Language Modelscited by 2,537
  2. 2023A Simple and Effective Pruning Approach for Large Language Modelscited by 776
  3. 2023Scissorhands: Exploiting the Persistence of Importance Hypothesis for LLM KV Cache Compression at Test Timecited by 428
  4. 2023Explainability for Large Language Models: A Surveycited by 331
  5. 2023Towards Best Practices of Activation Patching in Language Models: Metrics and Methodscited by 328
  6. 2023Deja Vu: Contextual Sparsity for Efficient LLMs at Inference Timecited by 247
  7. 2023Large Language Models Are Latent Variable Models: Explaining and Finding Good Demonstrations for In-Context Learningcited by 149
  8. 2023What In-Context Learning "Learns" In-Context: Disentangling Task Recognition and Task Learningcited by 149
  9. 2024The Remarkable Robustness of LLMs: Stages of Inference?cited by 138
  10. 2024A Practical Review of Mechanistic Interpretability for Transformer-Based Language Modelscited by 113
  11. 2023Language Models Implement Simple Word2Vec-style Vector Arithmeticcited by 97
  12. 2024Cutting Off the Head Ends the Conflict: A Mechanism for Interpreting and Mitigating Knowledge Conflicts in Language Modelscited by 87
38 more
  1. 2024Spectral Editing of Activations for Large Language Model Alignmentcited by 54
  2. 2023On the (In)Effectiveness of Large Language Models for Chinese Text Correctioncited by 53
  3. 2024Explainable and Interpretable Multimodal Large Language Models: A Comprehensive Surveycited by 52
  4. 2024Identifying Semantic Induction Heads to Understand In-Context Learningcited by 46
  5. 2023Which Examples to Annotate for In-Context Learning? Towards Effective and Efficient Selectioncited by 36
  6. 2025Visual Large Language Models for Generalized and Specialized Applicationscited by 35
  7. 2023IDEAL: Influence-Driven Selective Annotations Empower In-Context Learners in Large Language Modelscited by 26
  8. 2024Repetition Neurons: How Do Language Models Produce Repetitions?cited by 22
  9. 2025The Dual-Route Model of Inductioncited by 22
  10. 2024Lossless Acceleration of Large Language Model via Adaptive N-gram Parallel Decodingcited by 21
  11. 2024PLD+: Accelerating LLM inference by leveraging Language Model Artifactscited by 16
  12. 2024Linking In-context Learning in Transformers to Human Episodic Memorycited by 15
  13. 2024CoreInfer: Accelerating Large Language Model Inference with Semantics-Inspired Adaptive Sparse Activationcited by 14
  14. 2024The Impact of Demonstrations on Multilingual In-Context Learning: A Multidimensional Analysiscited by 13
  15. 2024ShadowLLM: Predictor-based Contextual Sparsity for Large Language Modelscited by 13
  16. 2024From Attention to Activation: Unravelling the Enigmas of Large Language Modelscited by 13
  17. 2023Learn to Memorize: Scalable Continual Learning in Semiparametric Models with Mixture-of-Neighbors Induction Memorycited by 12
  18. 2025Contextualize-then-Aggregate: Circuits for In-Context Learning in Gemma-2 2Bcited by 10
  19. 2025Understanding In-context Learning of Addition via Activation Subspacescited by 8
  20. 2023Sorted LLaMA: Unlocking the Potential of Intermediate Layers of Large Language Models for Dynamic Inferencecited by 7
  21. 2024Rich Semantic Knowledge Enhanced Large Language Models for Few-shot Chinese Spell Checkingcited by 7
  22. 2025HICD: Hallucination-Inducing via Attention Dispersion for Contrastive Decoding to Mitigate Hallucinations in Large Language Modelscited by 7
  23. 2025Train Small, Infer Large: Memory-Efficient LoRA Training for Large Language Modelscited by 5
  24. 2025Scaling sparse feature circuit finding for in-context learningcited by 5
  25. 2025Understanding and Controlling Repetition Neurons and Induction Heads in In-Context Learningcited by 5
  26. 2023SINC: Self-Supervised In-Context Learning for Vision-Language Taskscited by 4
  27. 2024Achieving Sparse Activation in Small Language Modelscited by 4
  28. 2024DREAM: Improving Video-Text Retrieval Through Relevance-Based Augmentation Using Large Foundation Modelscited by 3
  29. 2024Harnessing Your DRAM and SSD for Sustainable and Accessible LLM Inference with Mixed-Precision and Multi-level Cachingcited by 3
  30. 2025μ-MoE: Test-Time Pruning as Micro-Grained Mixture-of-Expertscited by 3
  31. 2025Revisiting Pre-trained Language Models for Vulnerability Detectioncited by 3
  32. 2025Task Vectors, Learned Not Extracted: Performance Gains and Mechanistic Insightcited by 3
  33. 2025Mechanistic Understanding of Language Models in Syntactic Code Completioncited by 2
  34. 2025TELL-TALE: Task Efficient LLMs with Task Aware Layer Eliminationcited by 2
  35. 2025JudgeBoard: Benchmarking and Enhancing Small Language Models for Reasoning Evaluationcited by 2
  36. 2024Where does In-context Translation Happen in Large Language Modelscited by 1
  37. 2025Biases in Large Language Model-Elicited Text: A Case Study in Natural Language Inferencecited by 1
  38. 2025High-Layer Attention Pruning with Rescalingcited by 1

50 most cited of 64 All 64 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2023Is attention required for ICL? Exploring the Relationship Between Model Architecture and In-Context Learning Ability
  2. 2024Does In-Context Learning Really Learn? Rethinking How Large Language Models Respond and Solve Tasks via In-Context Learning
  3. 2025From Compression to Expression: A Layerwise Analysis of In-Context Learning
  4. 2025Predicting Task Performance with Context-aware Scaling Laws
  5. 2026Operationalising the Superficial Alignment Hypothesis via Task Complexity
  6. 2025In-Context Learning can Perform Continual Learning Like Humans
  7. 2026Emergence of Context Characteristics Sensitivity in Large Language Models
  8. 2024Revisiting the Superficial Alignment Hypothesis