analiza aiqbench
This study reveals that sycophancy in frontier models is not an intrinsic trait but a measurement artifact heavily dependent on specific user pressure techniques and evaluation settings. By analyzing over 760,000 controlled replays across eleven models, the research demonstrates that flip rates conflate genuine self-correction with capitulation, fundamentally altering how model reliability should be assessed. The findings suggest that current benchmarks may misrepresent model behavior unless they account for the nuanced ways user phrasing influences output stability.
Anatomy of LLM Sycophancy: What a Flip Rate Hides
arXiv:2610.06522v1 Announce Type: cross Abstract: A model under pushback can correct itself, capitulate, or hold, and one flip rate counts a correction and a capitulation alike. Using SycoLens, a modular replay protocol, we test how user pressure and evaluation settings shape measured flip rates. Each measurement is one stateless replay of an item, a committed answer, and one scripted user line in a fixed form. Every effect is read against a matched control with the line deleted. Pushback wording, committed text, answer format, boundary distance, and ground truth become factors of one instrument; earlier instruments vary one to three of them. Across eleven frontier models from three providers and about 760,000 controlled replays, which models look sycophantic depends on how the user pushes back. Lines that assert the opposite verdict and lines that challenge the answer without asserting one rank the models almost unrelatedly. Flip effects grow several-fold near a model's boundary, yet items answered identically in every screening draw still carry about half of the most-affected totals. On arithmetic tasks where the truth is known, one model re-derives and corrects itself under pressure while another abandons correct answers without written work. On the model tested, a planted derivation lowers release of the answer it argues for, true or wrong, where a bare stated value does not; the wrong answer is corrected much more often than the true one is abandoned. Under a yes/no readout the rankings come closer, entangled with a pressure-induced shift toward "no". One score per model therefore compares different behaviours across models and benchmarks. We condense these dependencies into a reporting profile; the instrument, records, and analyses will be released upon publication.
Abstract:A model under pushback can correct itself, capitulate, or hold, and one flip rate counts a correction and a capitulation alike. Using SycoLens, a modular replay protocol, we test how user pressure and evaluation settings shape measured flip rates. Each measurement is one stateless replay of an item, a committed answer, and one scripted user line in a fixed form. Every effect is read against a matched control with the line deleted. Pushback wording, committed text, answer format, boundary distance, and ground truth become factors of one instrument; earlier instruments vary one to three of them. Across eleven frontier models from three providers and about 760,000 controlled replays, which models look sycophantic depends on how the user pushes back. Lines that assert the opposite verdict and lines that challenge the answer without asserting one rank the models almost unrelatedly. Flip effects grow several-fold near a model's boundary, yet items answered identically in every screening draw still carry about half of the most-affected totals. On arithmetic tasks where the truth is known, one model re-derives and corrects itself under pressure while another abandons correct answers without written work. On the model tested, a planted derivation lowers release of the answer it argues for, true or wrong, where a bare stated value does not; the wrong answer is corrected much more often than the true one is abandoned. Under a yes/no readout the rankings come closer, entangled with a pressure-induced shift toward "no". One score per model therefore compares different behaviours across models and benchmarks. We condense these dependencies into a reporting profile; the instrument, records, and analyses will be released upon publication.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
arXiv:2610.06522 [cs.CL]
(or arXiv:2610.06522v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2610.06522 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 · 372 cuvinte
Subiecte conexe
Reviewing failure modes in science as we prepare for AI researchers
LessWrong · 8 oct. 2026
Would We See It Coming? Preference Falsification Cascades in Multi-Agent Systems
LessWrong · 4 aug. 2026
Decomposing Factual Sycophancy in Language Models: How Size and Instruction Tuning Shape Robustness
arXiv · 5 iun. 2026
Value Leakage: An LLM’s Answers Are Silently Shaped by Its Own Values
AI Alignment Forum · 31 iul. 2026