mlsanityData to 7 Oct 2026

Auxiliary task demands mask the capabilities of smaller language models

arXiv 2404.02418 ↗First posted 3 Apr 2024cs.CL cs.AIcited by 54

Not on the map. Why?

Abstract (arXiv)

Developmental psychologists have argued about when cognitive capacities such as language understanding or theory of mind emerge. These debates often hinge on the concept of "task demands" -- the auxiliary challenges associated with performing a particular evaluation -- that may mask the child's underlying ability. The same issues arise when measuring the capacities of language models (LMs): performance on a task is a function of the model's underlying knowledge, combined with the model's ability to interpret and perform the task given its available resources. Here, we show that for analogical reasoning, reflective reasoning, word prediction, and grammaticality judgments, evaluation methods with greater task demands yield lower performance than evaluations with reduced demands. This "demand gap" is most pronounced for models with fewer parameters and less training data. Our results illustrate that LM performance should not be interpreted as a direct indication of intelligence (or lack thereof), but as a reflection of capacities seen through the lens of researchers' design choices.

Key facts. Cited by 54 papers in the map (data through 7 Oct 2026). Builds on 31 of its 50 references that are in the map. First posted on arXiv 3 Apr 2024.

Builds on

31 of 50 references.

References · 31

  1. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,302
  2. 2022Large Language Models are Zero-Shot Reasonerscited by 4,323
  3. 2022Emergent Abilities of Large Language Modelscited by 2,483
  4. 2023Pythia: A Suite for Analyzing Large Language Models Across Training and Scalingcited by 1,734
  5. 2016The LAMBADA dataset: Word prediction requiring a broad discourse contextcited by 960
  6. 2019Adversarial NLI: A New Benchmark for Natural Language Understandingcited by 798
  7. 2022Grokking: Generalization Beyond Overfitting on Small Algorithmic Datasetscited by 599
  8. 2019BLiMP: The Benchmark of Linguistic Minimal Pairs for Englishcited by 553
  9. 2023Are Emergent Abilities of Large Language Models a Mirage?cited by 515
  10. 2021Dynabench: Rethinking Benchmarking in NLPcited by 447
  11. 2016Assessing the Ability of LSTMs to Learn Syntax-Sensitive Dependenciescited by 350
  12. 2021AI and the Everything in the Whole Wide World Benchmarkcited by 340
19 more
  1. 2018Targeted Syntactic Evaluation of Language Modelscited by 239
  2. 2022Emergent Analogical Reasoning in Large Language Modelscited by 239
  3. 2021What Will it Take to Fix Benchmarking in Natural Language Understanding?cited by 200
  4. 2022Language models show human-like content effects on reasoning taskscited by 171
  5. 2024Political Compass or Spinning Arrow? Towards More Meaningful Evaluations for Values and Opinions in Large Language Modelscited by 162
  6. 2022Thinking Fast and Slow in Large Language Modelscited by 136
  7. 2020How Can We Accelerate Progress Towards Human-like Linguistic Generalization?cited by 114
  8. 2023The Generative AI Paradox: "What It Can Create, It May Not Understand"cited by 97
  9. 2023Language Model Behavior: A Comprehensive Surveycited by 92
  10. 2023Evaluating Cognitive Maps and Planning in Large Language Models with CogEvalcited by 77
  11. 2022minicons: Enabling Flexible Behavioral and Representational Analyses of Transformer Language Modelscited by 58
  12. 2024Language models align with human judgments on key grammatical constructionscited by 49
  13. 2023Prompting is not a substitute for probability measurements in large language modelscited by 43
  14. 2022Can language models handle recursively nested grammatical structures? A case study on comparing models and humanscited by 41
  15. 2021The Dangers of Underclaiming: Reasons for Caution When Reporting How NLP Systems Failcited by 38
  16. 2023On the Stepwise Nature of Self-Supervised Learningcited by 36
  17. 2024Beyond Probabilities: Unveiling the Misalignment in Evaluating Large Language Modelscited by 19
  18. 2024Predictions from language models for multiple-choice tasks are not robust under variation of scoring methodscited by 13
  19. 2019Language Tasks and Language Games: On Methodology in Current Natural Language Processing Researchcited by 6

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 18; 2025: 25; 2026: 11.

54 papers.

Sort:

By area:

  1. 2024Are Large Language Models Consistent over Value-laden Questions?cited by 58
  2. 2024One fish, two fish, but not the whole sea: Alignment reduces language models' conceptual diversitycited by 57
  3. 2024Elements of World Knowledge (EWoK): A Cognition-Inspired Framework for Evaluating Basic World Knowledge in Language Modelscited by 55
  4. 2024Social Bias Evaluation for Large Language Models Requires Prompt Variationscited by 39
  5. 2024LINGOLY: A Benchmark of Olympiad-Level Linguistic Reasoning Puzzles in Low-Resource and Extinct Languagescited by 28
  6. 2025Language Models Fail to Introspect About Their Knowledge of Languagecited by 28
  7. 2024Language Models as Models of Languagecited by 24
  8. 2025On Benchmarking Human-Like Intelligence in Machinescited by 24
  9. 2025Re-evaluating Theory of Mind evaluation in large language modelscited by 23
  10. 2024Robust Pronoun Fidelity with English LLMs: Are they Reasoning, Repeating, or Just Biased?cited by 17
  11. 2024LLMs as Models for Analogical Reasoningcited by 16
  12. 2025What Can String Probability Tell Us About Grammaticality?cited by 16
38 more
  1. 2025Aligned but Blind: Alignment Increases Implicit Bias by Reducing Awareness of Racecited by 15
  2. 2024Evidence from counterfactual tasks supports emergent analogical reasoning in large language modelscited by 14
  3. 2024DevBench: A multimodal developmental benchmark for language learningcited by 13
  4. 2025Language Model Behavioral Phases are Consistent Across Architecture, Training Data, and Scalecited by 13
  5. 2025A Survey on Collaborative Mechanisms Between Large and Small Language Modelscited by 12
  6. 2024Searching for Structure: Investigating Emergent Communication with Large Language Modelscited by 10
  7. 2024WinoPron: Revisiting English Winogender Schemas for Consistency, Coverage, and Grammatical Casecited by 9
  8. 2024Leaving the barn door open for Clever Hans: Simple features predict LLM benchmark answerscited by 9
  9. 2025Levels of Analysis for Large Language Modelscited by 9
  10. 2025When Do Large Language Models Exhibit Unsolicited Deception?cited by 9
  11. 2024Dissecting the Ullman Variations with a SCALPEL: Why do LLMs fail at Trivial Alterations to the False Belief Task?cited by 7
  12. 2025Random Scaling of Emergent Capabilitiescited by 7
  13. 2024Experimental Pragmatics with Machines: Testing LLM Predictions for the Inferences of Plain and Embedded Disjunctionscited by 5
  14. 2025Bigram Subnetworks: Mapping to Next Tokens in Transformer Language Modelscited by 5
  15. 2025Large Language Model probabilities cannot distinguish between possible and impossible languagecited by 5
  16. 2024Anthropocentric bias in language model evaluationcited by 4
  17. 2024Incremental Sentence Processing Mechanisms in Autoregressive Transformer Language Modelscited by 4
  18. 2025Understanding Common Ground Misalignment in Goal-Oriented Dialog: A Case-Study with Ubuntu Chat Logscited by 4
  19. 2025RankAlign: A Ranking View of the Generator-Validator Gap in Large Language Modelscited by 4
  20. 2026On the Limits of Layer Pruning for Generative Reasoning in Large Language Modelscited by 3
  21. 2025The potential -- and the pitfalls -- of using pre-trained language models as cognitive science theoriescited by 2
  22. 2025Signatures of human-like processing in Transformer forward passescited by 2
  23. 2025Not quite Sherlock Holmes: Language model predictions do not reliably differentiate impossible from improbable eventscited by 2
  24. 2025Metacognitive Sensitivity for Test-Time Dynamic Model Selectioncited by 2
  25. 2026How Open Must Language Models be to Enable Reliable Scientific Inference?cited by 2
  26. 2024Do Language Models Understand the Cognitive Tasks Given to Them? Investigations with the N-Back Paradigmcited by 1
  27. 2025MIAT: Maneuver-Intention-Aware Transformer for Spatio-Temporal Trajectory Predictioncited by 1
  28. 2025Multilingual LLMs Are Not Multilingual Thinkers: Evidence from Hindi Analogy Evaluationcited by 1
  29. 2025Uncertainty-Aware Collaborative System of Large and Small Models for Multimodal Sentiment Analysiscited by 1
  30. 2025Conflict Adaptation in Vision-Language Modelscited by 1
  31. 2025BeDiscovER: The Benchmark of Discourse Understanding in the Era of Reasoning Language Modelscited by 1
  32. 2026The Grammar of Transformers: A Systematic Review of Interpretability Research on Syntactic Knowledge in Language Modelscited by 1
  33. 2026Instructions Shape Production of Language, not Processingcited by 1
  34. 2025Investigating Syntactic Biases in Multilingual Transformers with RC Attachment Ambiguities in Italian and Englishcited by 0
  35. 2025Do Large Language Models know who did what to whom?cited by 0
  36. 2026Evaluating Pragmatic Reasoning in Large Language Models: Evidence from Scalar Diversitycited by 0
  37. 2026Investigating Concept Alignment Using Implausible Category Memberscited by 0
  38. 2026LEVANTE-bench: Multi-Scale Comparison of VLMs to Children Using Cognitive Tasks (or, "Is Your VLM Smarter Than a 5th Grader?")cited by 0

50 most cited of 54 All 54 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2022Define, Evaluate, and Improve Task-Oriented Cognitive Capabilities for Instruction Generation Models
  2. 2023Reasoning or Reciting? Exploring the Capabilities and Limitations of Language Models Through Counterfactual Tasks
  3. 2024From Babbling to Fluency: Evaluating the Evolution of Language Models in Terms of Human Language Acquisition
  4. 2023Comparing Machines and Children: Using Developmental Psychology Experiments to Assess the Strengths and Weaknesses of LaMDA Responses
  5. 2024Revisiting the Superficial Alignment Hypothesis
  6. 2025Evidence of conceptual mastery in the application of rules by Large Language Models
  7. 2024Core Knowledge Deficits in Multi-Modal Language Models
  8. 2023Psychometric Predictive Power of Large Language Models