Daniel Oberski

@daob
410 Followers
234 Following
22 Posts

Professor of Social & Health Data Science at Utrecht University & University Medical Center Utrecht, The Netherlands.

Statistics; Latent variables; Structural Equation Models; Methodology of data science; Applications of machine learning to research in social and health domains.

Group website @UUhttps://hds.sites.uu.nl
Publicationshttps://daob.nl/publications

An interesting phenomenon in asking chatGPT to generate R, Julia, Lean, and Python code: it invents syntax and facilities that *should* exist.

For example, if I ask for a survey SEM analysis in R, it expects there to be an svysem function. This function does not exist (it's called lavaan.survey), but maybe it should!

So ChatGPT can illustrate how users of software might reasonably expect the software to work, potentially helping us design it better.