mlsanityData to 7 Oct 2026

Language Models Understand Us, Poorly

arXiv 2210.10684 ↗First posted 19 Oct 2022cs.CL cs.AIcited by 3

Not on the map. Why?

Abstract (arXiv)

Some claim language models understand us. Others won't hear it. To clarify, I investigate three views of human language understanding: as-mapping, as-reliability and as-representation. I argue that while behavioral reliability is necessary for understanding, internal representations are sufficient; they climb the right hill. I review state-of-the-art language and multi-modal models: they are pragmatically challenged by under-specification of form. I question the Scaling Paradigm: limits on resources may prohibit scaled-up models from approaching understanding. Last, I describe how as-representation advances a science of understanding. We need work which probes model internals, adds more of human language, and measures what models can learn.

Key facts. Cited by 3 papers in the map (data through 7 Oct 2026). Builds on 35 of its 81 references that are in the map. First posted on arXiv 19 Oct 2022.

Builds on

35 of 81 references.

References · 35

  1. 2020Language Models are Few-Shot Learnerscited by 32,199
  2. 2022Hierarchical Text-Conditional Image Generation with CLIP Latentscited by 6,539
  3. 2020Scaling Laws for Neural Language Modelscited by 6,411
  4. 2022Photorealistic Text-to-Image Diffusion Models with Deep Language Understandingcited by 6,322
  5. 2022PaLM: Scaling Language Modeling with Pathwayscited by 5,304
  6. 2021Zero-Shot Text-to-Image Generationcited by 4,276
  7. 2022LaMDA: Language Models for Dialog Applicationscited by 1,218
  8. 2022In-context Learning and Induction Headscited by 930
  9. 2019Right for the Wrong Reasons: Diagnosing Syntactic Heuristics in Natural Language Inferencecited by 864
  10. 2022A Generalist Agentcited by 802
  11. 2022AudioLM: a Language Modeling Approach to Audio Generationcited by 611
  12. 2022Winoground: Probing Vision and Language Models for Visio-Linguistic Compositionalitycited by 523
23 more
  1. 2021Causal Abstractions of Neural Networkscited by 401
  2. 2020Experience Grounds Languagecited by 241
  3. 2022The Carbon Footprint of Machine Learning Training Will Plateau, Then Shrinkcited by 183
  4. 2021Implicit Representations of Meaning in Neural Language Modelscited by 179
  5. 2022SCROLLS: Standardized CompaRison Over Long Language Sequencescited by 170
  6. 2021How much do language models copy from their training data? Evaluating linguistic novelty in text generation using RAVENcited by 124
  7. 2020How Can We Accelerate Progress Towards Human-like Linguistic Generalization?cited by 114
  8. 2022A very preliminary analysis of DALL-E 2cited by 82
  9. 2018Image Chat: Engaging Grounded Conversationscited by 76
  10. 2019Approximate Causal Abstractioncited by 62
  11. 2022Testing Relational Understanding in Text-Guided Image Generationcited by 59
  12. 2021What Would Jiminy Cricket Do? Towards Agents That Behave Morallycited by 57
  13. 2021Provable Limitations of Acquiring Meaning from Ungrounded Form: What Will Future Language Models Understand?cited by 55
  14. 2020Word meaning in minds and machinescited by 51
  15. 2019Environmental drivers of systematicity and generalization in a situated agentcited by 50
  16. 2021Grounding 'Grounding' in NLPcited by 44
  17. 2020Effects of Parameter Norm Growth During Transformer Training: Inductive Bias from Gradient Descentcited by 40
  18. 2022What Do NLP Researchers Believe? Results of the NLP Community Metasurveycited by 28
  19. 2021Symbolic Behaviour in Artificial Intelligencecited by 24
  20. 2022Unit Testing for Concepts in Neural Networkscited by 23
  21. 2022Abstraction for Deep Reinforcement Learningcited by 19
  22. 2021Emergent Communication under Competitioncited by 17
  23. 2020PuzzLing Machines: A Challenge on Learning From Small Datacited by 15

Built on by

Cited by year

Citing papers per year: 2023: 1; 2024: 2.

3 papers.

  1. 2024Benchmarks as Microscopes: A Call for Model Metrologycited by 33
  2. 2023Mastering Symbolic Operations: Augmenting Language Models with Compiled Neural Networkscited by 12
  3. 2024Large Language Models, and LLM-Based Agents, Should Be Used to Enhance the Digital Public Spherecited by 1