BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:You Know It or You Don’t: Categorical Differences in Language Mo
 del Behavior - Naomi Saphra (Harvard University &amp\; Boston University)
DTSTART:20260212T163000Z
DTEND:20260212T173000Z
UID:TALK244528@talks.cam.ac.uk
CONTACT:Lucas Resck
DESCRIPTION:Abstract: While years of scientific research on model training
  and scaling assume that learning is a gradual and continuous process\, br
 eakthroughs on specific capabilities have drawn wide attention. Why are br
 eakthroughs so exciting? Because humans don’t naturally think in continu
 ous gradients\, but in discrete conceptual categories. If artificial langu
 age models naturally learn discrete conceptual categories\, perhaps model 
 understanding is within our grasp. I will describe what we know of categor
 ical learning in language models\, and how discrete concepts are identifia
 ble through empirical training dynamics and through random variation betwe
 en training runs. These concepts involve syntax learning\, weight mechanis
 ms\, and interpretable patterns---all of which can predict model behavior.
  By leveraging categorical learning\, we can ultimately understand a model
 's natural conceptual structure and evaluate our understanding through tes
 table predictions.\n\nBio: Naomi Saphra is a research fellow at the Kempne
 r Institute at Harvard University and incoming faculty at Boston Universit
 y in 2026. Naomi is interested in empirically understanding training in NL
 P and language models: how models learn to encode linguistic patterns or o
 ther structure and how we can encode useful inductive biases into the trai
 ning process. Recently\, she has begun collaborating with natural and soci
 al scientists to use interpretability to understand the world around us. S
 he is particularly interested in fish. Previously\, she earned a PhD from 
 the University of Edinburgh on Training Dynamics of Neural Language Models
 \; worked at NYU\, Google\, MosaicML\, and Facebook\; and attended Johns H
 opkins and Carnegie Mellon University. Outside of research\, she plays rol
 ler derby under the name Gaussian Retribution and performs standup comedy.
LOCATION:https://cam-ac-uk.zoom.us/j/86890624365?pwd=oYGWpY7d5r3JOaUCaJXTD
 0sRECFxab.1
END:VEVENT
END:VCALENDAR
