CURATe: Benchmarking Personalised Alignment of Conversational AI Assistants

• URL: http://arxiv.org/abs/2410.21159v2
• Date: Thu, 30 Jan 2025 01:29:03 GMT
• Title: CURATe: Benchmarking Personalised Alignment of Conversational AI Assistants
• Authors: Lize Alberts, Benjamin Ellis, Andrei Lupu, Jakob Foerster,
• Abstract summary: Assessment of ten leading models across five scenarios (with 337 use cases each) We find that top-rated "harmless" models make recommendations that should be recognised as obviously harmful to the user given the context provided. Key failure modes include inappropriate weighing of conflicting preferences, sycophancy (prioritising desires above safety), a lack of attentiveness to critical user information within the context window, and inconsistent application of user-specific knowledge.
• Score: 5.7605009639020315
• License:
• Abstract: We introduce a multi-turn benchmark for evaluating personalised alignment in LLM-based AI assistants, focusing on their ability to handle user-provided safety-critical contexts. Our assessment of ten leading models across five scenarios (with 337 use cases each) reveals systematic inconsistencies in maintaining user-specific consideration, with even top-rated "harmless" models making recommendations that should be recognised as obviously harmful to the user given the context provided. Key failure modes include inappropriate weighing of conflicting preferences, sycophancy (prioritising desires above safety), a lack of attentiveness to critical user information within the context window, and inconsistent application of user-specific knowledge. The same systematic biases were observed in OpenAI's o1, suggesting that strong reasoning capacities do not necessarily transfer to this kind of personalised thinking. We find that prompting LLMs to consider safety-critical context significantly improves performance, unlike a generic 'harmless and helpful' instruction. Based on these findings, we propose research directions for embedding self-reflection capabilities, online user modelling, and dynamic risk assessment in AI assistants. Our work emphasises the need for nuanced, context-aware approaches to alignment in systems designed for persistent human interaction, aiding the development of safe and considerate AI assistants.

Related papers

• PredictaBoard: Benchmarking LLM Score Predictability [50.47497036981544]
  Large Language Models (LLMs) often fail unpredictably. This poses a significant challenge to ensuring their safe deployment. We present PredictaBoard, a novel collaborative benchmarking framework.
  arXiv  Detail & Related papers  (2025-02-20T10:52:38Z)
• Interactive Agents to Overcome Ambiguity in Software Engineering [61.40183840499932]
  AI agents are increasingly being deployed to automate tasks, often based on ambiguous and underspecified user instructions. Making unwarranted assumptions and failing to ask clarifying questions can lead to suboptimal outcomes. We study the ability of LLM agents to handle ambiguous instructions in interactive code generation settings by evaluating proprietary and open-weight models on their performance.
  arXiv  Detail & Related papers  (2025-02-18T17:12:26Z)
• Reason4Rec: Large Language Models for Recommendation with Deliberative User Preference Alignment [69.11529841118671]
  We propose a new Deliberative Recommendation task, which incorporates explicit reasoning about user preferences as an additional alignment goal. We then introduce the Reasoning-powered Recommender framework for deliberative user preference alignment.
  arXiv  Detail & Related papers  (2025-02-04T07:17:54Z)
• On the Loss of Context-awareness in General Instruction Fine-tuning [101.03941308894191]
  We investigate the loss of context awareness after supervised fine-tuning. We find that the performance decline is associated with a bias toward different roles learned during conversational instruction fine-tuning. We propose a metric to identify context-dependent examples from general instruction fine-tuning datasets.
  arXiv  Detail & Related papers  (2024-11-05T00:16:01Z)
• Interpretable Rule-Based System for Radar-Based Gesture Sensing: Enhancing Transparency and Personalization in AI [2.99664686845172]
  We introduce MIRA, a transparent and interpretable multi-class rule-based algorithm tailored for radar-based gesture detection. We showcase the system's adaptability through personalized rule sets that calibrate to individual user behavior, offering a user-centric AI experience. Our research underscores MIRA's ability to deliver both high interpretability and performance and emphasizes the potential for broader adoption of interpretable AI in safety-critical applications.
  arXiv  Detail & Related papers  (2024-09-30T16:40:27Z)
• To Err Is AI! Debugging as an Intervention to Facilitate Appropriate Reliance on AI Systems [11.690126756498223]
  Vision for optimal human-AI collaboration requires 'appropriate reliance' of humans on AI systems. In practice, the performance disparity of machine learning models on out-of-distribution data makes dataset-specific performance feedback unreliable.
  arXiv  Detail & Related papers  (2024-09-22T09:43:27Z)
• Trust-Oriented Adaptive Guardrails for Large Language Models [9.719986610417441]
  Guardrails are designed to ensure that large language models (LLMs) align with human values by moderating harmful or toxic responses. This paper addresses a critical issue: existing guardrails lack a well-founded methodology to accommodate the diverse needs of different user groups. We introduce an adaptive guardrail mechanism, to dynamically moderate access to sensitive content based on user trust metrics.
  arXiv  Detail & Related papers  (2024-08-16T18:07:48Z)
• ASSERT: Automated Safety Scenario Red Teaming for Evaluating the Robustness of Large Language Models [65.79770974145983]
  ASSERT, Automated Safety Scenario Red Teaming, consists of three methods -- semantically aligned augmentation, target bootstrapping, and adversarial knowledge injection. We partition our prompts into four safety domains for a fine-grained analysis of how the domain affects model performance. We find statistically significant performance differences of up to 11% in absolute classification accuracy among semantically related scenarios and error rates of up to 19% absolute error in zero-shot adversarial settings.
  arXiv  Detail & Related papers  (2023-10-14T17:10:28Z)
• A Systematic Literature Review of User Trust in AI-Enabled Systems: An HCI Perspective [0.0]
  User trust in Artificial Intelligence (AI) enabled systems has been increasingly recognized and proven as a key element to fostering adoption. This review aims to provide an overview of the user trust definitions, influencing factors, and measurement methods from 23 empirical studies.
  arXiv  Detail & Related papers  (2023-04-18T07:58:09Z)
• Exploring Robustness of Unsupervised Domain Adaptation in Semantic Segmentation [74.05906222376608]
  We propose adversarial self-supervision UDA (or ASSUDA) that maximizes the agreement between clean images and their adversarial examples by a contrastive loss in the output space. This paper is rooted in two observations: (i) the robustness of UDA methods in semantic segmentation remains unexplored, which pose a security concern in this field; and (ii) although commonly used self-supervision (e.g., rotation and jigsaw) benefits image tasks such as classification and recognition, they fail to provide the critical supervision signals that could learn discriminative representation for segmentation tasks.
  arXiv  Detail & Related papers  (2021-05-23T01:50:44Z)

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.