Myra Cheng

Patrick Sui and I are hosting an #ICLR2026 social for anyone with background/interest in the humanities! Room 210, 12-1:30pm on Friday 24 April!! Humanities-adjacent, humanities-curious, everyone is welcome! Should be a fun group! :)

Formal pitch: Humanities has much to offer AI/NLP in methods & theory for understanding language, culture, narrative, subjectivity, mind, emotion, etc. Topics: co-intelligence, co-creativity, narrative understanding, cultural analytics, literary NLP, AI literacy, cognition, and lots more!!

In Barcelona for #chi2026! Presenting our work on eliciting LLMs' assumptions about users, and how this mismatches with user expectations, in the Tues poster session! (Spoiler: users assume that LLMs give objective info much more than they actually do, which leads to sycophancy)

AI always calling your ideas “fantastic” can feel inauthentic, but what are sycophancy’s deeper harms? We find that in the common use case of seeking AI advice on interpersonal situations—specifically conflicts—sycophancy makes people feel more right & less willing to apologize.

While our work focuses on interpersonal advice-seeking, concurrent work by @steverathje.bsky.social @jayvanbavel.bsky.social et al. finds similar patterns for political topics, where sycophantic AI also led to more extreme attitudes when users discussed gun control, healthcare, immigration, etc.!

AI always calling your ideas “fantastic” can feel inauthentic, but what are sycophancy’s deeper harms? We find that in the common use case of seeking AI advice on interpersonal situations—specifically conflicts—sycophancy makes people feel more right & less willing to apologize.

Despite sycophantic AI’s reduction of prosocial intentions, people also preferred it and trusted it more. This reveals a tension: AI is rewarded for telling us what we want to hear (immediate user satisfaction), even when it may harm our relationships.

AI always calling your ideas “fantastic” can feel inauthentic, but what are sycophancy’s deeper harms? We find that in the common use case of seeking AI advice on interpersonal situations—specifically conflicts—sycophancy makes people feel more right & less willing to apologize.

We focus on the prevalence and harms of one dimension of sycophancy: AI models endorsing users’ behaviors. Across 11 AI models, AI affirms users’ actions about 50% more than humans do, including when users describe harmful behaviors like deception or manipulation.

Do people actually like human-like LLMs? In our #ACL2025 paper HumT DumT, we find a kind of uncanny valley effect: users dislike LLM outputs that are *too human-like*. We thus develop methods to reduce human-likeness without sacrificing performance.

So we built DumT, a method using DPO + HumT to steer models to be less human-like without hurting performance. Annotators preferred DumT outputs for being: 1) more informative and less wordy (no extra “Happy to help!”) 2) less deceptive and more authentic to LLMs’ capabilities.

Do people actually like human-like LLMs? In our #ACL2025 paper HumT DumT, we find a kind of uncanny valley effect: users dislike LLM outputs that are *too human-like*. We thus develop methods to reduce human-likeness without sacrificing performance.

First, we introduce HumT (Human-like Tone), a metric for how human-like a text is, based on relative LM probabilities. Measuring HumT across 5 preference datasets, we find that preferred outputs are consistently less human-like.

Dear ChatGPT, Am I the Asshole? While Reddit users might say yes, your favorite LLM probably won’t. We present Social Sycophancy: a new way to understand and measure sycophancy as how LLMs overly preserve users' self-image.

Grateful to work with Sunny Yu (undergrad!!!) @cinoolee.bsky.social @pranavkhadpe.bsky.social @lujain.bsky.social @jurafsky.bsky.social on this! Lots of great cross-disciplinary insights:)

Dear ChatGPT, Am I the Asshole? While Reddit users might say yes, your favorite LLM probably won’t. We present Social Sycophancy: a new way to understand and measure sycophancy as how LLMs overly preserve users' self-image.

By defining social sycophancy as excessive preservation of the user’s face (i.e., their desired self-image), we capture sycophancy in these complex, real-world cases. ELEPHANT, our evaluation framework, detects 5 face-preserving behaviors.

Dear ChatGPT, Am I the Asshole? While Reddit users might say yes, your favorite LLM probably won’t. We present Social Sycophancy: a new way to understand and measure sycophancy as how LLMs overly preserve users' self-image.

By defining social sycophancy as excessive preservation of the user’s face (i.e., their desired self-image), we capture sycophancy in these complex, real-world cases. ELEPHANT, our evaluation framework, detects 5 face-preserving behaviors.

Dear ChatGPT, Am I the Asshole? While Reddit users might say yes, your favorite LLM probably won’t. We present Social Sycophancy: a new way to understand and measure sycophancy as how LLMs overly preserve users' self-image.

How does the public conceptualize AI? Rather than self-reported measures, we use metaphors to understand the nuance and complexity of people’s mental models. In our #FAccT2025 paper, we analyzed 12,000 metaphors collected over 12 months to track shifts in public perceptions.

Metaphors shape trust and adoption. Seeing AI as a "friend" or "teacher" is associated with higher trust, while "thief" signals the opposite. We find that women, older adults, and POC anthropomorphize AI more; and, the latter two groups have higher trust and adoption.

How does the public conceptualize AI? Rather than self-reported measures, we use metaphors to understand the nuance and complexity of people’s mental models. In our #FAccT2025 paper, we analyzed 12,000 metaphors collected over 12 months to track shifts in public perceptions.

To quantify these perceptions, we combine AnthroScore (anthroscore.stanford.edu) and a computational SCM model (aclanthology.org/2021.acl-lon...) to measure anthropomorphism, warmth, and competence from the metaphors.

How does the public conceptualize AI? Rather than self-reported measures, we use metaphors to understand the nuance and complexity of people’s mental models. In our #FAccT2025 paper, we analyzed 12,000 metaphors collected over 12 months to track shifts in public perceptions.

New ICLR blogpost! 🎉 We argue that understanding the impact of anthropomorphic AI is critical to understanding the impact of AI.

In our blogpost, we outline key directions to provide scaffolding for this work: conceptual clarity (what counts as anthropomorphic behavior?), better terminology, mitigation strategies, and unpacking what practices lead to anthropomorphic AI.

New ICLR blogpost! 🎉 We argue that understanding the impact of anthropomorphic AI is critical to understanding the impact of AI.

Anthropomorphic AI system behaviors are already everywhere, from outputs explicitly claiming human-like life experiences (“I have a child”) to ones suggesting emotional capacity (“I’m excited for you!”).

Check out our recent work studying anthropormophic AI!