I’m Gabriel Recchia, a cognitive scientist at the University of Cambridge’s Winton Centre for Risk and Evidence Communication, where I work on how to communicate information in ways that support comprehension and informed decision-making; I also lead on user testing research and evaluation of patient-friendly genetic reports and the NHS: Predict family of prognostic tools. I have spent much of my career involved in research investigating the capabilities, properties, and applications of distributional models trained on large volumes of text, and have continued this while at the Winton Centre to explore their applications in characterizing how risk is communicated and perceived. See my Google Scholar profile for a list of my most cited papers.
My current research interests are on the “human side” of aligning modern language models with human preferences, with a particular focus on ensuring that such models remain aligned even in cases where naïve approaches to eliciting these preferences do not yield a reliable and accurate training signal. One promising class of approaches involves making it easier for people to accurately evaluate the outputs of language models, e.g., by training language models to critique their own outputs, or by training them to produce justifications of their outputs in a form that is easy to evaluate. Finding out how successful these approaches are, and how they might be tweaked to be more successful, requires experiments with human participants. I’m especially curious about whether ways of making a language model’s “reasoning process” more intelligible–e.g., “showing one’s work”–might have a role to play in making it easier for people to evaluate model behaviour accurately.
Previously, I was at the Centre for Research in the Arts, Social Sciences and Humanities, where I worked with distributional approaches to the analysis of large corpora of historical texts, and investigated conceptual change by attending to shifting statistical associations between words over time. This position also involved the development and testing of user interfaces for the display of complex quantitative information to individuals of various backgrounds.
I received my bachelor’s degree in Symbolic Systems from Stanford University in 2007, and my doctorate is in Cognitive Science at Indiana University, with a minor in computational linguistics and with language modelling as my content specialization. Publications, skills, and previous employers are listed on my CV; Google Scholar may be more up-to-date.