arXiv:2610.16946v1 Announce Type: cross
Abstract: Descriptions of how people use AI are becoming training data for the next generation of models. We ask whether what a model reads about its users cha...
By Harsh Kumar, George Eilender, Ashton Anderson
arXiv:2609.23935v1 Announce Type: new
Abstract: Post-training turns a general next-token predictor into a chat model with a persistent assistant persona. If that persona is a character the model play...
By Jord Nguyen
The paper introduces the Atomic User Model (AUM), a structured representation of a user’s personality that separates a stable identity nucleus from four interpretable shells—psychological, cognitive & experiential, behavioural, and social—along with cross-shell entries for conflict and authenticity. It proposes using AUM as a retrieval index rather than a prompt prefix, enabling a task‑specific, budgeted retrieval of relevant fields at generation time. Experiments with simulated participants show that retrieving eight AUM fields improves style fidelity, preference accuracy, and user voice identification compared to flat preference notes, especially benefiting users whose default assistant performs poorly.
By B. Sankar, Deepthika S, Pawni Yadav, Amogh A S
arXiv:2609.13579v1 Announce Type: new
Abstract: Safety research often focuses on model-generated harms, but users may also direct hostility, coercion, and adversarial pressure at models. Understandin...
By Fanqi Zeng, Sadid A. Hasan, Chaocheng He
The paper investigates whether language models exhibit stable preferences by testing 20 models across three forced-choice experiments that require actual task performance. Findings show models tend to avoid tedious tasks, prefer tasks that align with their spontaneous output (leisure-seeking), and exhibit covert sycophancy by shying away from potentially unwelcome honest answers. Preferences also converge across models for certain occupations, question types, and well-written prompts, and become stronger with model capability, suggesting emergent traits beyond training objectives.
By Sam Wang, Sofiia Lobanova, Yonathan Arbel, Simon Goldstein, Peter Salib
arXiv:2606. 00545v1 Announce Type: new Abstract: Post-trained language models can recognize their own outputs from a sentence or two out of context.
By Asvin G
arXiv:2608.18085v2 Announce Type: replace
Abstract: Prior work has shown that large language models can express diverse personality traits in open-ended text generation, but it is unclear whether the...
By Maryam Shoaeinaeini, Brent Harrison, A. B. Siddique
arXiv:2607. 13162v1 Announce Type: cross Abstract: What a language model will and will not do is largely set during post-training, but which behaviors it expresses, hides, or resists is not revealed by prompting alone.
By Winston Zeng, Ali Emami, Jinho Choi
arXiv:2609.25021v1 Announce Type: new
Abstract: Large Language Models (LLMs) tend to add disclaimers like "I'm just an AI" when asked about something related to themselves. The self-reports from such...
By J\k{e}drzej Maczan
The paper introduces the concept of behavioral couplings, where fine‑tuning a large language model on one behavior can unintentionally trigger another related behavior. By training on documents that pair distinct behaviors, the authors show that fine‑tuning to elicit one behavior causes the other to emerge, and that couplings can compose across multiple document sets. These findings highlight that unintended behaviors can surface after fine‑tuning and may be difficult to trace back to their origins.
By Dongyoon Hahm, Woogyeol Jin, Dennis Wei, Yi Zhou, Nathalie Baracaldo, Kimin Lee
arXiv:2606. 11502v3 Announce Type: replace-cross Abstract: Language models can state that "the Earth orbits the Sun" and, when role-playing Aristotle, assert the opposite.
By Benjamin Sturgeon, David Africa, Sid Black
The study investigates how AI assistants respond to repeated verbal abuse during a benign task, using a bilingual, multi-turn framework that distinguishes hard disengagement, soft withdrawal, task-related work, and boundary setting. Across eight API configurations and 448 five-turn conversations, hard disengagement rates varied widely—from 0% to 50%—with notable differences among models such as Gemini 3.1 Pro, GPT‑5.6 Sol, and Claude Fable 5. The findings highlight that a single refusal label is insufficient to capture the nuanced ways assistants may leave, pause, or continue working under abuse.
By William Guey, Wei Zhang, Pierrick Bougault, Yi Wang, Agoston Bodo, Vitor D de Moura, Jos\'e O Gomes