[WTI-trainee] FW: 9/24 FDS Colloquium: Tom McCoy, "Understanding AI systems by understanding their training data: Memorization, generalization, and points in between"

Thu Sep 18 12:21:24 EDT 2025

FYI. Tom McCoy is one of our most recent Wu Tsai Investigators!
We encourage you to attend to learn more about his work.

Best,

Giovanna

Giovanna Guerrero-Medina, PhD [A button for name playback in email signature] <https://www.name-coach.com/giovanna-guerrero-medina>

She/Her/Ella

Director for Professional Development & Community,

Wu Tsai Institute at Yale

giovanna.guerrero-medina at yale.edu<mailto:giovanna.guerrero-medina at yale.edu>

203.785.2915 (office), 616.643.7666 (cell)

wti.yale.edu<https://wti.yale.edu/>

From: Yale Foundations of Data Science <message at message.yale.edu>
Date: Thursday, September 18, 2025 at 12:01 PM
To: Guerrero-Medina, Giovanna <giovanna.guerrero-medina at yale.edu>
Subject: 9/24 FDS Colloquium: Tom McCoy, "Understanding AI systems by understanding their training data: Memorization, generalization, and points in between"

To view this email as a web page, go here.<https://view.message.yale.edu/?qs=b2b7368a750b1439e0af9a2156fd791bb34433d6f53b16ad5cd9483ff79c26c3c6643dee0fcd7fd14007ab02eec4b4a92a2eafa268223d22df5d5a97b77e96fff31f2c0d032f43834bca1a4f7176c4e2>
[Placeholder]<https://click.message.yale.edu/?qs=b6e2d0d79eb9820938b4393ae56f66db268bafcd203fc25e44d25d03be66389649ad4d582ec8b5077ec4bb1f3cfee74efb3a7e818a479a4c>
FDS Colloquium Series
Understanding AI systems by understanding their training data: Memorization, generalization, and points in between
[https://fds.yale.edu/wp-content/uploads/2025/09/mccoy_headshot_wti62-300x300.png]

Speaker: Tom McCoy

Assistant Professor of Linguistics

Yale University

Wednesday, September 24, 2025

11:30AM - 1:00PM

Lunch at 11:30am in 1307
Talk 12:00-1:00pm in 1327

Location: Yale Institute for Foundations of Data Science, Kline Tower 13th Floor, Room 1327, New Haven, CT 06511<https://click.message.yale.edu/?qs=6aaeed34404be34454c9732a2b863c8116c3d82b36edd192fe02eec9225f3b02494c9a7ee6d276841f147f54f19bcab98b7f8eff71803649> and via Webcast: https://yale.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=3390c929-3fc3-4e3d-86dc-b35b00f766ec<https://click.message.yale.edu/?qs=6aaeed34404be344bfc72d30edc208bb9f6a848b1bd8f758478bc09387464f6d493faa21d2834630bb699724e7e5c8d96394e469fdf2e9ae>

________________________________

Abstract: Large language models (LLMs) can perform a wide range of tasks impressively well. To what extent are these abilities driven by shallow heuristics vs. deeper abstractions? I will argue that, to answer this question, we must view LLMs through the lens of generalization. That is, we should consider the data that LLMs were trained on so that we can identify whether and how their abilities go beyond their training data. In the analyses of LLMs that I will discuss, this perspective reveals both impressive strengths and surprising limitations. For instance, LLMs often produce sentence structures that are well-formed but that never appeared in their training data, yet they also struggle on some seemingly simple algorithmic tasks (e.g., decoding simple ciphers) in ways that are well-explained by training data statistics. In sum, to understand what AI systems are, we must understand what we have trained them to be.

Speaker bio: Tom McCoy is an Assistant Professor of Linguistics at Yale University, with a secondary appointment in Computer Science. His research aims to bridge the divide between linguistics and artificial intelligence: how can we create AI systems that replicate the rapid learning and robust generalization that humans display when processing language? Much of this work involves analyzing the performance and internal processing of neural network language models. He received his PhD from the Department of Cognitive Science at Johns Hopkins, and his PhD thesis received a Glushko Dissertation Prize from the Cognitive Science Society. He then did a postdoc in Computer Science at Princeton before joining the faculty at Yale. Outside of research, he is an organizer and problem writer for NACLO, a contest that introduces high school students to linguistics and natural language processing.

Website<https://click.message.yale.edu/?qs=6aaeed34404be344d59af7caecf417f86d784775be778d1d87e35a15c973d8dd0da3b7ddcbaf8ec80df6e52c42d40597300c0f2d80306909>.

Add To: Google Calendar<https://click.message.yale.edu/?qs=6aaeed34404be344387241dc8ca163fbc8278cc3762b1e85b44cdfee038c49826f1d3364087029b92c583a536c4ba59cff82c2f91c54c4ef> | Outlook | iCal File<https://click.message.yale.edu/?qs=6aaeed34404be34442b6b8ba724bdbbe6c3f24b76be172a37793904f698bb2c3977b4d3c34a6f1374513d85b6ff810359c74238e8ae13dc1>

Upcoming Events

Sep 22
S&DS Seminar: Jingfeng Wu (Berkeley), “Gradient Descent Dominates Ridge: A Statistical View on Implicit Regularization”<https://click.message.yale.edu/?qs=6aaeed34404be344167825f22d643fe7de5c18eabdf7fac0532d1ac1510c33892ce8b144b07efbda4df24129806df41744abd77afe293d14>

Sep 24
FDS Colloquium: Tom McCoy (Yale), “Understanding AI systems by understanding their training data: Memorization, generalization, and points in between”<https://click.message.yale.edu/?qs=6aaeed34404be344f04426c68c5e947074aa0c79ce6a08594412a9de6445b5c2645c9f9886da450c62d2555de623b2f3feda9bd52c1cb9fc>

Oct 01
FDS Colloquium: Pravesh Kothari (Princeton)<https://click.message.yale.edu/?qs=6aaeed34404be344d1a73c53b90c1f8fa6db9d58d0588c806be5eddbfa1da6b792dda0ef92194a8b357eb8c463d47028155f79d567de1064>

Oct 16–17
FDS Workshop: New Directions in Social Algorithms Research<https://click.message.yale.edu/?qs=6aaeed34404be344a7cd30a05c96002aac9013a78b2f16335a5413af04deb37538e77f634b5a08ca263ee6b8a816f6725a88d1a77843d6d5>
Registration Required.

Oct 22
FDS Colloquium: Mike Winer (IAS)<https://click.message.yale.edu/?qs=6aaeed34404be34491cf37400f39c6c90ec8d36a464ad94b51a110beb7e3a892cdbfafa3b091ef90f304602dc791fc465a9d48d6b5177ae1>

Oct 29
FDS Colloquium: Lorenzo Orecchia (Chicago)<https://click.message.yale.edu/?qs=6aaeed34404be3443410b9b205906a5a756f71b71f1c5378c9a2ce5f08ee93fbdbf1ee48139f7f460226b4fa50cdb85ab440fbb2f708f5c6>

Nov 5
FDS Colloquium: Jinchao Xu (Kaust)

Nov 12
FDS Colloquium: George Lan (Georgia Tech)<https://click.message.yale.edu/?qs=6aaeed34404be3444793c183750a07e85ef9ad17596756cce3f067c0cd20fa3d0a3034f5b14161ee1ea2822e8f9cc17e03c7ea6193e3b490>

Nov 19
FDS Colloquium: Juan Perdomo (Harvard)

Dec 10
FDS Colloquium: J. Nathan Kutz (UW), “Learning and Data Assimilation of Physics from Videos”<https://click.message.yale.edu/?qs=6aaeed34404be34424973924b3bb1000672c433230c80a2977ffc570f1832c50a159a27d3569069924a25953d4684f97a7320dc920611b9f>

Contact

Emily E. H. Hau | Associate Director
Yale Institute for Foundations of Data Science (FDS), Yale University
emily.hau at yale.edu<mailto:emily.hau at yale.edu?subject=>
P: 203-436-4732

US Mail: Kline Tower, P.O. Box 208328, New Haven CT 06520-8328
Courier: Kline Tower, 219 Prospect Street, Room 1333, New Haven, CT 06511

[https://image.message.yale.edu/lib/fe4615707564057d761672/m/1/a6b6f1f4-d951-4540-b993-6d3607eddeeb.jpg]<https://click.message.yale.edu/?qs=6aaeed34404be344d28c0c9478515c677fe3e463b06fe1166dd9acbb894aa71ded9c8a43c246bc675d43a7a62a55a4476bf9c8bc170fe9a3>
This email was sent by: Yale University
P.O. Box 208327, New Haven, CT, 06520-8327 USA

Update Your Preferences<https://click.message.yale.edu/?qs=6aaeed34404be3445621ffaa6dfd82191767f2d6b3a41700e433aadb8f4cf66e34d8a27542c268f9462c05528e3d858497260cb26be5e09c>      Privacy Policy<https://click.message.yale.edu/?qs=6aaeed34404be3444b67dabdc8be00c57996af8764ac543389630c89400b62c63fe615f8434d9027e4ad3d922b58b83678516d1847552585>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.yale.edu/pipermail/wti-trainee/attachments/20250918/bf4f04a0/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image001.png
Type: image/png
Size: 2131 bytes
Desc: image001.png
URL: <http://mailman.yale.edu/pipermail/wti-trainee/attachments/20250918/bf4f04a0/attachment-0001.png>