The case for training AI on biological signals, not just text
An interview summary presents Stanford professor James Zou’s argument that text captures only a fraction of thought—and that superintelligence needs training on physiological signals, not just words.
TLDR
A post summarizing an interview with Stanford professor James Zou quotes him calling written language a “downstream artifact” representing “less than 1% of the actual thoughts.” The summary presents his case for training AI on physiological signals. It also describes projects from his lab, including Paper2Agent, which turns papers into interactive “virtual authors,” and SleepFM, which it says predicts risk for about 130 diseases from one night of sleep data.
Combined views
44
1 Source, first seen ago