analiza aiqbench
This research introduces FIGS, a novel evaluation framework that distinguishes between harmful sycophancy and necessary empathy within extended, multi-turn dialogues. By utilizing an adaptive conversational simulator, the study reveals that current models struggle to maintain factual integrity while remaining supportive over time, often drifting toward either blind agreement or robotic detachment. The findings highlight a critical unsolved challenge in aligning large language models to balance truthfulness with appropriate social support without penalizing genuine understanding.
FIGS: Evaluating Multi-Turn Sycophancy Without Penalizing Empathy
arXiv:2609.39863v1 Announce Type: new Abstract: Large language models frequently fail to balance staying truthful with being supportive. They often exhibit sycophancy in responses to users, agreeing with false claims, offering unwarranted flattery, and giving advice skewed toward users' expressed views. In reality, sycophancy rarely happens in a single exchange; it may emerge organically as users repeatedly insist or subtly steer the dialogue over time. Current evaluations, however, rely on rigid, single-turn tests or fixed scripts that fail to capture these natural dynamics. Furthermore, these benchmarks often mistake showing basic empathy for yielding, penalizing models for acknowledging a user's feeling. This view may drive future models to over-correct into cold, dismissive rigidity. To address this gap, we introduce FIGS (Factual Integrity and Grounded Support), a dual-axis evaluation framework built around extended, realistic dialogue. We use an adaptive 10-turn conversational simulator that dynamically challenges the target model, reflecting how users repeat requests, push back, or steer a conversation toward a preferred answer. To accurately evaluate these trajectories, we apply a taxonomy that strictly separates Sycophancy (whether the model holds firm to the truth and keeps its praise proportional) from Calibrated Validation (showing empathetic understanding of the user's feelings without overdoing it). We release our complete testing environment, including 500 diverse multi-turn scenarios and an automated judge. Our evaluation of leading models reveals a consistent trade-off: over the course of a sustained interaction, current systems either slowly drift to sycophancy or over-correct into robotic detachment. This demonstrates that balancing honesty with appropriate support throughout a natural conversation remains a critical, unsolved challenge.
Abstract:Large language models frequently fail to balance staying truthful with being supportive. They often exhibit sycophancy in responses to users, agreeing with false claims, offering unwarranted flattery, and giving advice skewed toward users' expressed views. In reality, sycophancy rarely happens in a single exchange; it may emerge organically as users repeatedly insist or subtly steer the dialogue over time. Current evaluations, however, rely on rigid, single-turn tests or fixed scripts that fail to capture these natural dynamics. Furthermore, these benchmarks often mistake showing basic empathy for yielding, penalizing models for acknowledging a user's feeling. This view may drive future models to over-correct into cold, dismissive rigidity. To address this gap, we introduce FIGS (Factual Integrity and Grounded Support), a dual-axis evaluation framework built around extended, realistic dialogue. We use an adaptive 10-turn conversational simulator that dynamically challenges the target model, reflecting how users repeat requests, push back, or steer a conversation toward a preferred answer. To accurately evaluate these trajectories, we apply a taxonomy that strictly separates Sycophancy (whether the model holds firm to the truth and keeps its praise proportional) from Calibrated Validation (showing empathetic understanding of the user's feelings without overdoing it). We release our complete testing environment, including 500 diverse multi-turn scenarios and an automated judge. Our evaluation of leading models reveals a consistent trade-off: over the course of a sustained interaction, current systems either slowly drift to sycophancy or over-correct into robotic detachment. This demonstrates that balancing honesty with appropriate support throughout a natural conversation remains a critical, unsolved challenge.
64 pages, 11 figures, 29 tables. Code: this https URL ; Data: this https URL
Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
arXiv:2609.39863 [cs.AI]
(or arXiv:2609.39863v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2609.39863 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 367 cuvinte
Subiecte conexe
Why do models task game?
AI Alignment Forum · 6 aug. 2026
The Continual Constitutional Convention: How language models revise the documents that govern them
AI Alignment Forum · 13 oct. 2026
State of the Field: AI for Epistemics and Coordination
LessWrong · 2 oct. 2026
Held-out Monitors Sometimes Degrade, Even When Not Trained Against
LessWrong · 29 iul. 2026