Interpreting the Repeated Token Phenomenon in Large Language Models

• URL: http://arxiv.org/abs/2503.08908v1
• Date: Tue, 11 Mar 2025 21:40:58 GMT
• Title: Interpreting the Repeated Token Phenomenon in Large Language Models
• Authors: Itay Yona, Ilia Shumailov, Jamie Hayes, Federico Barbero, Yossi Gandelsman,
• Abstract summary: Large Language Models (LLMs) often fail to accurately repeat a single word when prompted to, and instead output unrelated text.<n>We aim to explain the causes for this phenomenon and link it to the concept of attention sinks''<n>Our investigation identifies the neural circuit responsible for attention sinks and shows how long repetitions disrupt this circuit.
• Score: 31.1226642501095
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large Language Models (LLMs), despite their impressive capabilities, often fail to accurately repeat a single word when prompted to, and instead output unrelated text. This unexplained failure mode represents a vulnerability, allowing even end-users to diverge models away from their intended behavior. We aim to explain the causes for this phenomenon and link it to the concept of ``attention sinks'', an emergent LLM behavior crucial for fluency, in which the initial token receives disproportionately high attention scores. Our investigation identifies the neural circuit responsible for attention sinks and shows how long repetitions disrupt this circuit. We extend this finding to other non-repeating sequences that exhibit similar circuit disruptions. To address this, we propose a targeted patch that effectively resolves the issue without negatively impacting the model's overall performance. This study provides a mechanistic explanation for an LLM vulnerability, demonstrating how interpretability can diagnose and address issues, and offering insights that pave the way for more secure and reliable models.

Related papers

• Don't Take Things Out of Context: Attention Intervention for Enhancing Chain-of-Thought Reasoning in Large Language Models [32.71672086718058]
  Few-shot Chain-of-Thought (CoT) significantly enhances the reasoning capabilities of large language models (LLMs) We observe that isolated segments, words, or tokens within CoT demonstrations can unexpectedly disrupt the generation process of LLMs. We propose a Few-shot Attention Intervention method (FAI) that dynamically analyzes the attention patterns of demonstrations to accurately identify these tokens.
  arXiv  Detail & Related papers  (2025-03-14T07:46:33Z)
• Causality can systematically address the monsters under the bench(marks) [64.36592889550431]
  Benchmarks are plagued by various biases, artifacts, or leakage.<n>Models may behave unreliably due to poorly explored failure modes.<n> causality offers an ideal framework to systematically address these challenges.
  arXiv  Detail & Related papers  (2025-02-07T17:01:37Z)
• Reflective Instruction Tuning: Mitigating Hallucinations in Large Vision-Language Models [36.119299938503936]
  Large vision-language models (LVLMs) have shown promising performance on a variety of vision-language tasks. They remain susceptible to hallucinations, generating outputs misaligned with visual content or instructions. We propose reflective instruction tuning, which integrates rationale learning into visual instruction tuning.
  arXiv  Detail & Related papers  (2024-07-16T06:32:45Z)
• Rethinking harmless refusals when fine-tuning foundation models [0.8571111167616167]
  We investigate the degree to which fine-tuning in Large Language Models (LLMs) effectively mitigates versus merely conceals undesirable behavior. We identify a pervasive phenomenon we term emphreason-based deception, where models either stop producing reasoning traces or produce seemingly ethical reasoning traces that belie the unethical nature of their final outputs.
  arXiv  Detail & Related papers  (2024-06-27T22:08:22Z)
• Preemptive Answer "Attacks" on Chain-of-Thought Reasoning [7.233752893356647]
  Large language models (LLMs) showcase impressive reasoning capabilities when coupled with Chain-of-Thought prompting. In this paper, we introduce a novel scenario termed preemptive answers, where the LLM obtains an answer before engaging in reasoning. Experiments reveal that preemptive answers significantly impair the model's reasoning capability across various CoT methods and a broad spectrum of datasets.
  arXiv  Detail & Related papers  (2024-05-31T15:15:04Z)
• Navigating the OverKill in Large Language Models [84.62340510027042]
  We investigate the factors for overkill by exploring how models handle and determine the safety of queries. Our findings reveal the presence of shortcuts within models, leading to an over-attention of harmful words like 'kill' and prompts emphasizing safety will exacerbate overkill. We introduce Self-Contrastive Decoding (Self-CD), a training-free and model-agnostic strategy, to alleviate this phenomenon.
  arXiv  Detail & Related papers  (2024-01-31T07:26:47Z)
• Improving Input-label Mapping with Demonstration Replay for In-context Learning [67.57288926736923]
  In-context learning (ICL) is an emerging capability of large autoregressive language models. We propose a novel ICL method called Sliding Causal Attention (RdSca) We show that our method significantly improves the input-label mapping in ICL demonstrations.
  arXiv  Detail & Related papers  (2023-10-30T14:29:41Z)
• Gaining Wisdom from Setbacks: Aligning Large Language Models via Mistake Analysis [127.85293480405082]
  The rapid development of large language models (LLMs) has not only provided numerous opportunities but also presented significant challenges. Existing alignment methods usually direct LLMs toward the favorable outcomes by utilizing human-annotated, flawless instruction-response pairs. This study proposes a novel alignment technique based on mistake analysis, which deliberately exposes LLMs to erroneous content to learn the reasons for mistakes and how to avoid them.
  arXiv  Detail & Related papers  (2023-10-16T14:59:10Z)
• Exposing Attention Glitches with Flip-Flop Language Modeling [55.0688535574859]
  This work identifies and analyzes the phenomenon of attention glitches in large language models. We introduce flip-flop language modeling (FFLM), a family of synthetic benchmarks designed to probe the extrapolative behavior of neural language models. We find that Transformer FFLMs suffer from a long tail of sporadic reasoning errors, some of which we can eliminate using various regularization techniques.
  arXiv  Detail & Related papers  (2023-06-01T17:44:35Z)
• Remembering for the Right Reasons: Explanations Reduce Catastrophic Forgetting [100.75479161884935]
  We propose a novel training paradigm called Remembering for the Right Reasons (RRR) RRR stores visual model explanations for each example in the buffer and ensures the model has "the right reasons" for its predictions. We demonstrate how RRR can be easily added to any memory or regularization-based approach and results in reduced forgetting.
  arXiv  Detail & Related papers  (2020-10-04T10:05:27Z)

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.