Sonia Murthy
@soniakmurthy
cs phd student @harvard · prev predoc @allen_ai, ra @cocosci_lab, undergrad @princeton · she/her
Presenting this today (5/1) at the 4pm poster session (Hall 3) at #NAACL2025! Come chat about alignment, personalization, and all things cognitive science 🐟
(1/9) Excited to share my recent work on "Alignment reduces LM's conceptual diversity" with @TomerUllman and @jennhu, to appear at #NAACL2025! 🐟 We want models that match our values...but could this hurt their diversity of thought? Preprint: arxiv.org/abs/2411.04427
NEW blog post: Do modern #LLMs capture the conceptual diversity of human populations? #KempnerInstitute researchers find #alignment reduces conceptual diversity of language models. Read more: bit.ly/4hNjtiI @soniakmurthy @tomerullman @_jennhu
The behavior of agents that ‘do what you ask, but not what you want’ has long been observed, sparking AI safety concerns. Today (12/10) @ 11am, I'll be presenting our work explicitly evaluating LLMs' capacity to engage in loophole behavior 🤖🧑➰ Come say hi! #EMNLP2023
(corrected link/title): How close are we to a machine smart-ass? "Comparing the Evaluation and Production of Loophole Behavior in Humans and Large Language Models" Lead & presented in EMNLP Findings by @soniakmurthy (support by @KempnerInst) aclanthology.org/2023.findings-…
Excited to present our ongoing work investigating loophole behavior in LLMs today at @tom_icml2023! #ICML2023 #NLProc
In ICML? --> Check out the Theory of Mind workshop! @tom_icml2023 In the Theory of Mind workshop? --> Check out Sonia Murthy's @soniakmurthy work! "Comparing the Evaluation and Production of Loophole Behavior in Children and Large Language Models" 🤖🧒➰
What makes a tutoring session more effective than a lecture? In our new preprint, we propose a hierarchical Bayesian model of pedagogy to explore the cognitive mechanisms underlying adaptive teaching.🧵 w/ Andrew Palacci @natvelali @hawkrobe @gershbrain psyarxiv.com/4u5g6/
Presenting this today at 11am in the main atrium! #EMNLP2022
Introducing ACCoRD, a system by @soniakmurthy and coauthors that can produce multiple diverse descriptions of a scientific concept by taking advantage of the many ways that concept is discussed across the scientific literature. Appearing at #EMNLP2022! blog.allenai.org/introducing-ac…