I build AI systems that operate in the messy reality of human behavior. Here, I write what does not fit in a paper: the intuition behind a result, the judgment calls, and the experiments that failed before something worked. Slower than a tweet, looser than a paper.
Note: The opinions expressed on this blog are my own and do not necessarily reflect the views of my employer or any affiliated organization.
-
Synthetic eye tracking: teaching NLP models how people read
Real eye-tracking is costly and privacy-sensitive. Synthetic scanpaths can still carry useful reading structure when the limits of the proxy are kept explicit.