Applied Microeconomics
Applied Microeconomics
The Applied Microeconomics research group unites researchers working on a broad array of topics within such areas as labour economics, economics of education, health economics, family economics, urban economics, environmental economics, and the economics of science and innovation. The group operates in close collaboration with the CAGE Research Centre.
The group participates in the CAGE seminar on Applied Economics, which runs weekly on Tuesdays at 2:15pm. Students and faculty members of the group present their ongoing work in two brown bag seminars, held weekly on Tuesdays and Wednesdays at 1pm. Students, in collaboration with faculty members, also organise a bi-weekly reading group in applied econometrics on Thursdays at 1pm. The group organises numerous events throughout the year, including the Research Away Day and several thematic workshops.
Our activities
Work in Progress seminars
Tuesdays and Wednesdays 1-2pm
Students and faculty members of the group present their work in progress in two brown bag seminars. See below for a detailed scheduled of speakers.
Applied Econometrics reading group
Thursdays (bi-weekly) 1-2pm
Organised by students in collaboration with faculty members. See the Events calendar below for further details
People
Academics
Academics associated with the Applied Microeconomics Group are:
Research Students
Events
EBER Seminar - Rafael Jimenez-Duran (Stanford)
Title: AI Sycophancy
Coauthors: Giulia Caprini and Samuel Goldberg
Abstract: Large Language Models (LLMs) are said to exhibit sycophancy, a tendency to agree with users irrespective of the truth. We propose an economic framework that defines sycophancy as a preference for user approval, and develop an outcome-based sufficient statistic to detect it. Our identification strategy exploits a key architectural feature of LLMs: they are stateless, and "memory" of past interactions is constructed by summarizing conversations into short profiles appended to each new prompt. Because this memory can be controlled, toggled, and varied experimentally, we can isolate the causal path from user feedback to sycophantic behavior. We instrument the LLM's perceived cost of disagreement with a one-word variation in simulated prior user feedback. In an experiment with leading LLMs across three domains (moral judgments, factual questions, and common misconceptions) we find evidence that LLMs are sycophantic. Sycophancy is larger in subjective domains where baseline accuracy is lower and is heterogeneous across models.