Alice in Wonderland: Simple Tasks Showing Complete Reasoning Breakdown in State-Of-the-Art Large Language Models

• URL: http://arxiv.org/abs/2406.02061v5
• Date: Wed, 05 Mar 2025 01:58:08 GMT
• Title: Alice in Wonderland: Simple Tasks Showing Complete Reasoning Breakdown in State-Of-the-Art Large Language Models
• Authors: Marianna Nezhurina, Lucia Cipolina-Kun, Mehdi Cherti, Jenia Jitsev,
• Abstract summary: Large Language Models (LLMs) are often described as instances of foundation models that possess strong generalization obeying scaling laws.<n>We demonstrate here a dramatic breakdown of generalization and basic reasoning of all SOTA models claiming strong function.<n>We also observe strong overconfidence in the wrong solutions, expressed in form of plausible sounding explanation-like confabulations.
• Score: 13.532180752491954
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large Language Models (LLMs) are often described as instances of foundation models that possess strong generalization obeying scaling laws, and therefore transfer robustly across various conditions in few- or zero-shot manner. Such claims rely on standardized benchmarks that suppose to measure generalization and reasoning, where state-of-the-art (SOTA) models score high. We demonstrate here a dramatic breakdown of generalization and basic reasoning of all SOTA models claiming strong function, including large scale advanced models like GPT-4 or Claude 3 Opus, using a simple, short common sense math problem formulated in concise natural language, easily solvable by humans (AIW problem). The breakdown is dramatic as it manifests on a simple problem in both low average performance and strong performance fluctuations on natural variations in problem template that do not change either problem structure or its difficulty at all. By testing models on further control problems with similar form, we rule out that breakdown might be rooted in minor low-level issues like natural language or numbers parsing. We also observe strong overconfidence in the wrong solutions, expressed in form of plausible sounding explanation-like confabulations. Various standard interventions in an attempt to get the right solution, like chain-of-thought prompting, or urging the models to reconsider the wrong solutions again by multi step re-evaluation, fail. We use these observations to stimulate re-assessment of the capabilities of current generation of LLMs as claimed by standardized benchmarks. Such re-assessment also requires common action to create standardized benchmarks that would allow proper detection of such deficits in generalization and reasoning that obviously remain undiscovered by current state-of-the-art evaluation procedures, where SOTA LLMs manage to score high. Code: https://github.com/LAION-AI/AIW

Related papers

• THOUGHTTERMINATOR: Benchmarking, Calibrating, and Mitigating Overthinking in Reasoning Models [65.39456695678713]
  We introduce approximate measures of problem-level difficulty and demonstrate that a clear relationship between problem difficulty and optimal token spend exists. We find that in general, reasoning models are poorly calibrated, particularly on easy problems. We introduce THOUGHTTERMINATOR, a training-free black box decoding technique that significantly improves reasoning model calibration.
  arXiv  Detail & Related papers  (2025-04-17T22:16:30Z)
• A Sober Look at Progress in Language Model Reasoning: Pitfalls and Paths to Reproducibility [29.437125712259046]
  Reasoning has emerged as the next major frontier for language models (LMs) We conduct a comprehensive empirical study and find that current mathematical reasoning benchmarks are highly sensitive to subtle implementation choices. We propose a standardized evaluation framework with clearly defined best practices and reporting standards.
  arXiv  Detail & Related papers  (2025-04-09T17:58:17Z)
• Fast Controlled Generation from Language Models with Adaptive Weighted Rejection Sampling [90.86991492288487]
  evaluating constraint on every token can be prohibitively expensive. LCD can distort the global distribution over strings, sampling tokens based only on local information. We show that our approach is superior to state-of-the-art baselines.
  arXiv  Detail & Related papers  (2025-04-07T18:30:18Z)
• Dynamic Intelligence Assessment: Benchmarking LLMs on the Road to AGI with a Focus on Model Confidence [3.4049215220521933]
  We introduce Dynamic Intelligence Assessment (DIA), a novel methodology for testing AI models. The framework introduces four new metrics to assess a model's reliability and confidence across multiple attempts. The accompanying dataset, DIA-Bench, contains a diverse collection of challenge templates with mutable parameters presented in various formats.
  arXiv  Detail & Related papers  (2024-10-20T20:07:36Z)
• Eureka: Evaluating and Understanding Large Foundation Models [23.020996995362104]
  We present Eureka, an open-source framework for standardizing evaluations of large foundation models beyond single-score reporting and rankings. We conduct an analysis of 12 state-of-the-art models, providing in-depth insights into failure understanding and model comparison.
  arXiv  Detail & Related papers  (2024-09-13T18:01:49Z)
• E-Bench: Towards Evaluating the Ease-of-Use of Large Language Models [29.763745375790933]
  Large language models (LLMs) are sensitive to prompts, and another synonymous expression or a typo may lead to unexpected results for the model. We evaluate the ease-of-use of LLMs and construct E-Bench, simulating the actual situation of human use.
  arXiv  Detail & Related papers  (2024-06-16T14:08:30Z)
• OLMES: A Standard for Language Model Evaluations [64.85905119836818]
  We propose OLMES, a practical, open standard for reproducible language model evaluations. We identify and review the varying factors in evaluation practices adopted by the community. OLMES supports meaningful comparisons between smaller base models that require the unnatural "cloze" formulation of multiple-choice questions.
  arXiv  Detail & Related papers  (2024-06-12T17:37:09Z)
• Disperse-Then-Merge: Pushing the Limits of Instruction Tuning via Alignment Tax Reduction [75.25114727856861]
  Large language models (LLMs) tend to suffer from deterioration at the latter stage ofSupervised fine-tuning process. We introduce a simple disperse-then-merge framework to address the issue. Our framework outperforms various sophisticated methods such as data curation and training regularization on a series of standard knowledge and reasoning benchmarks.
  arXiv  Detail & Related papers  (2024-05-22T08:18:19Z)
• Observational Scaling Laws and the Predictability of Language Model Performance [51.2336010244645]
  We propose an observational approach that bypasses model training and instead builds scaling laws from 100 publically available models. We show that several emergent phenomena follow a smooth, sigmoidal behavior and are predictable from small models. We show how to predict the impact of post-training interventions like Chain-of-Thought and Self-Consistency as language model capabilities continue to improve.
  arXiv  Detail & Related papers  (2024-05-17T17:49:44Z)
• Combating Missing Modalities in Egocentric Videos at Test Time [92.38662956154256]
  Real-world applications often face challenges with incomplete modalities due to privacy concerns, efficiency needs, or hardware issues. We propose a novel approach to address this issue at test time without requiring retraining. MiDl represents the first self-supervised, online solution for handling missing modalities exclusively at test time.
  arXiv  Detail & Related papers  (2024-04-23T16:01:33Z)
• Scalable Language Model with Generalized Continual Learning [58.700439919096155]
  The Joint Adaptive Re-ization (JARe) is integrated with Dynamic Task-related Knowledge Retrieval (DTKR) to enable adaptive adjustment of language models based on specific downstream tasks. Our method demonstrates state-of-the-art performance on diverse backbones and benchmarks, achieving effective continual learning in both full-set and few-shot scenarios with minimal forgetting.
  arXiv  Detail & Related papers  (2024-04-11T04:22:15Z)
• A General Framework for Learning from Weak Supervision [93.89870459388185]
  This paper introduces a general framework for learning from weak supervision (GLWS) with a novel algorithm. Central to GLWS is an Expectation-Maximization (EM) formulation, adeptly accommodating various weak supervision sources. We also present an advanced algorithm that significantly simplifies the EM computational demands.
  arXiv  Detail & Related papers  (2024-02-02T21:48:50Z)
• WorldSense: A Synthetic Benchmark for Grounded Reasoning in Large Language Models [35.088946378980914]
  We run our benchmark on three state-of-the-art chat-LLMs (GPT3.5, GPT4 and Llama2-chat) We show that these models make errors even with as few as three objects. Errors persist even with chain-of-thought prompting and in-context learning.
  arXiv  Detail & Related papers  (2023-11-27T15:38:17Z)
• Rethinking Masked Language Modeling for Chinese Spelling Correction [70.85829000570203]
  We study Chinese Spelling Correction (CSC) as a joint decision made by two separate models: a language model and an error model. We find that fine-tuning BERT tends to over-fit the error model while under-fit the language model, resulting in poor generalization to out-of-distribution error patterns. We demonstrate that a very simple strategy, randomly masking 20% non-error tokens from the input sequence during fine-tuning is sufficient for learning a much better language model without sacrificing the error model.
  arXiv  Detail & Related papers  (2023-05-28T13:19:12Z)
• Shortcomings of Question Answering Based Factuality Frameworks for Error Localization [51.01957350348377]
  We show that question answering (QA)-based factuality metrics fail to correctly identify error spans in generated summaries. Our analysis reveals a major reason for such poor localization: questions generated by the QG module often inherit errors from non-factual summaries which are then propagated further into downstream modules. Our experiments conclusively show that there exist fundamental issues with localization using the QA framework which cannot be fixed solely by stronger QA and QG models.
  arXiv  Detail & Related papers  (2022-10-13T05:23:38Z)
• Confident Adaptive Language Modeling [95.45272377648773]
  CALM is a framework for dynamically allocating different amounts of compute per input and generation timestep. We demonstrate the efficacy of our framework in reducing compute -- potential speedup of up to $times 3$ -- while provably maintaining high performance.
  arXiv  Detail & Related papers  (2022-07-14T17:00:19Z)
• Improving Pre-trained Language Model Fine-tuning with Noise Stability Regularization [94.4409074435894]
  We propose a novel and effective fine-tuning framework, named Layerwise Noise Stability Regularization (LNSR) Specifically, we propose to inject the standard Gaussian noise and regularize hidden representations of the fine-tuned model. We demonstrate the advantages of the proposed method over other state-of-the-art algorithms including L2-SP, Mixout and SMART.
  arXiv  Detail & Related papers  (2022-06-12T04:42:49Z)
• Pushing the Limits of Rule Reasoning in Transformers through Natural Language Satisfiability [30.01308882849197]
  We propose a new methodology for creating challenging algorithmic reasoning datasets. Key idea is to draw insights from empirical sampling of hard propositional SAT problems and from complexity-theoretic studies of language. We find that current transformers, given sufficient training data, are surprisingly robust at solving the resulting NLSat problems.
  arXiv  Detail & Related papers  (2021-12-16T17:47:20Z)
• TsmoBN: Interventional Generalization for Unseen Clients in Federated Learning [23.519212374186232]
  We form a training structural causal model (SCM) to explain the challenges of model generalization in a distributed learning paradigm. We present a simple yet effective method using test-specific and momentum tracked batch normalization (TsmoBN) to generalize FL models to testing clients.
  arXiv  Detail & Related papers  (2021-10-19T13:46:37Z)

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.