Cover Image

Old Habits Die Hard: How Conversational History Geometrically Traps LLMs

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Simhi, Adi, Barez, Fazl, Tutek, Martin, Belinkov, Yonatan, Cohen, Shay B.
Format:	Preprint
Published:	2026
Subjects:	Computation and Language Artificial Intelligence I.2.7
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Trust Me, I'm Wrong: LLMs Hallucinate with Certainty Despite Knowing the Answer
by: Simhi, Adi, et al.
Published: (2025)

ManagerBench: Evaluating the Safety-Pragmatism Trade-off in Autonomous LLMs
by: Simhi, Adi, et al.
Published: (2025)

Constructing Benchmarks and Interventions for Combating Hallucinations in LLMs
by: Simhi, Adi, et al.
Published: (2024)

Distinguishing Ignorance from Error in LLM Hallucinations
by: Simhi, Adi, et al.
Published: (2024)

HACK: Hallucinations Along Certainty and Knowledge Axes
by: Simhi, Adi, et al.
Published: (2025)

REVS: Unlearning Sensitive Information in Language Models via Rank Editing in the Vocabulary Space
by: Ashuach, Tomer, et al.
Published: (2024)

CRISP: Persistent Concept Unlearning via Sparse Autoencoders
by: Ashuach, Tomer, et al.
Published: (2025)

Reasoning Models Know What's Important, and Encode It in Their Activations
by: Nikankin, Yaniv, et al.
Published: (2026)

Are formal and functional linguistic mechanisms dissociated in language models?
by: Hanna, Michael, et al.
Published: (2025)

Have Faith in Faithfulness: Going Beyond Circuit Overlap When Finding Model Mechanisms
by: Hanna, Michael, et al.
Published: (2024)

A Dataset for Metaphor Detection in Early Medieval Hebrew Poetry
by: Toker, Michael, et al.
Published: (2024)

Masked by Consensus: Disentangling Privileged Knowledge in LLM Correctness
by: Ashuach, Tomer, et al.
Published: (2026)

Arithmetic Without Algorithms: Language Models Solve Math With a Bag of Heuristics
by: Nikankin, Yaniv, et al.
Published: (2024)

Same Task, Different Circuits: Disentangling Modality-Specific Mechanisms in VLMs
by: Nikankin, Yaniv, et al.
Published: (2025)

LLMs Know More Than They Show: On the Intrinsic Representation of LLM Hallucinations
by: Orgad, Hadas, et al.
Published: (2024)

ReFACT: Updating Text-to-Image Models by Editing the Text Encoder
by: Arad, Dana, et al.
Published: (2023)

Position-aware Automatic Circuit Discovery
by: Haklay, Tal, et al.
Published: (2025)

Diffusion Lens: Interpreting Text Encoders in Text-to-Image Pipelines
by: Toker, Michael, et al.
Published: (2024)

Large Language Models Generate Harmful Content Using a Distinct, Unified Mechanism
by: Orgad, Hadas, et al.
Published: (2026)

RomanLens: The Role Of Latent Romanization In Multilinguality In LLMs
by: Saji, Alan, et al.
Published: (2025)

The Personalization Trap: How User Memory Alters Emotional Reasoning in LLMs
by: Fang, Xi, et al.
Published: (2025)

Syntactic Blind Spots: How Misalignment Leads to LLMs Mathematical Errors
by: Williamson, Dane, et al.
Published: (2025)

Curveball Steering: The Right Direction To Steer Isn't Always Linear
by: Raval, Shivam, et al.
Published: (2026)

Hard to Be Heard: Phoneme-Level ASR Analysis of Phonologically Complex, Low-Resource Endangered Languages
by: Akavarapu, V. S. D. S. Mahesh, et al.
Published: (2026)

Blocks Architecture (BloArk): Efficient, Cost-Effective, and Incremental Dataset Architecture for Wikipedia Revision History
by: Li, Lingxi, et al.
Published: (2024)

WHoW: A Cross-domain Approach for Analysing Conversation Moderation
by: Chen, Ming-Bin, et al.
Published: (2024)

CauESC: A Causal Aware Model for Emotional Support Conversation
by: Chen, Wei, et al.
Published: (2024)

From Guessing to Asking: An Approach to Resolving the Persona Knowledge Gap in LLMs during Multi-Turn Conversations
by: Baskar, Sarvesh, et al.
Published: (2025)

Growing a Tail: Increasing Output Diversity in Large Language Models
by: Shur-Ofry, Michal, et al.
Published: (2024)

How Well Do LLMs Imitate Human Writing Style?
by: Jemama, Rebira, et al.
Published: (2025)

LLMs Are Not Scorers: Rethinking MT Evaluation with Generation-Based Methods
by: Cui, Hyang
Published: (2025)

RAG-Optimized Tibetan Tourism LLMs: Enhancing Accuracy and Personalization
by: Qi, Jinhu, et al.
Published: (2024)

Neither Valid nor Reliable? Investigating the Use of LLMs as Judges
by: Chehbouni, Khaoula, et al.
Published: (2025)

MAWARITH: A Dataset and Benchmark for Legal Inheritance Reasoning with LLMs
by: Bouchekif, Abdessalam, et al.
Published: (2026)

MIRIAD: Augmenting LLMs with millions of medical query-response pairs
by: Zheng, Qinyue, et al.
Published: (2025)

Credit C-GPT: A Domain-Specialized Large Language Model for Conversational Understanding in Vietnamese Debt Collection
by: Hong, Nhung Nguyen Thi, et al.
Published: (2026)

Fine-Tuning LLMs with Fine-Grained Human Feedback on Text Spans
by: CH-Wang, Sky, et al.
Published: (2025)

Overview of the Sensemaking Task at the ELOQUENT 2025 Lab: LLMs as Teachers, Students and Evaluators
by: Šindelář, Pavel, et al.
Published: (2025)

Just Pass Twice: Efficient Token Classification with LLMs for Zero-Shot NER
by: Ewais, Ahmed, et al.
Published: (2026)

MALT: Mechanistic Ablation of Lossy Translation in LLMs for a Low-Resource Language: Urdu
by: Bajwa, Taaha Saleem
Published: (2025)