Connectionists: Non-technical Workshop on LLMs

[David H Kirshner]

Ashwath Shetty, Paula Kuokkanen, and James Stovald,

My appreciation to each of you for the resources you've identified that address LLMs in a not-too-technical way.
I'm posting them for the benefit of anyone else who is interested.

David
_____________________

>From Paula:
I found this text helpful for understanding the internal working logics of LLMs, despite knowing the technicalities too:
https://medium.com/@colin.fraser/who-are-we-talking-to-when-we-talk-to-these-bots-9a7e673f8525 .
The text is from 2023 but its main messages hold with newer versions of ChatGTP.

>From James:
I can recommend this tutorial by Brandon Rohrer: https://www.brandonrohrer.com/transformers which explains the underlying model using easy-to-understand language.

>From Ashwath:
I would really recommend this series from 3blue1brown
https://youtube.com/playlist?list=PLZHQObOWTQDNU6R1_67000Dx_ZCJB-3pi&si=_cmDDGb10ww0uxed
And this series from andrej karpathy
https://youtu.be/7xTGNNLPyMI?si=D38_teACvbsfAaFE

From: Ashwath Shetty <shettyashwath2010 at gmail.com>
Sent: Tuesday, March 18, 2025 5:51 AM
To: David H Kirshner <dkirsh at lsu.edu>
Cc: connectionists at mailman.srv.cs.cmu.edu
Subject: Re: Connectionists: Non-technical Workshop on LLMs

You don't often get email from shettyashwath2010 at gmail.com<mailto:shettyashwath2010 at gmail.com>. Learn why this is important<https://aka.ms/LearnAboutSenderIdentification>

I would really recommend this series from 3blue1brown
https://youtube.com/playlist?list=PLZHQObOWTQDNU6R1_67000Dx_ZCJB-3pi&si=_cmDDGb10ww0uxed

And this series from andrej karpathy
https://youtu.be/7xTGNNLPyMI?si=D38_teACvbsfAaFE

-Ashwath

On Tue, 18 Mar, 2025, 08:28 David H Kirshner, <dkirsh at lsu.edu<mailto:dkirsh at lsu.edu>> wrote:
I am an educator interested in the implications of LLMs for the teaching profession. I've studied the basics of ANNs to gain an understanding of how connectionist models simulate intelligent behavior adaptively, without a guiding mental model or rule system. But my exposure to ChatGPT leaves me baffled. If the output layer is focused on identifying the appropriate next word in a string of text, it shouldn't be able to perform nearly as effectively as it does.
Does anyone know of a workshop, seminar, or piece of writing that explains LLMs in a relatively non-technical way?
David Kirshner
________________________
David Kirshner, Professor
Mathematics Education
103 Peabody Hall
School of Education
Louisiana State University
Baton Rouge LA 70803-4728
Office: (225) 578-2332    Fax: (225) 578-9135
https://www.lsu.edu/chse/education/faculty_and_staff_directory/dkirshner.php

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20250318/f13ad8b8/attachment.html>