Lightnews — Scholar-powered news

Tyler Shoemaker @t-shoemaker.bsky.social · 1d

So not a token or token but a token

1 1

Tyler Shoemaker @t-shoemaker.bsky.social · 1d

Zoom in: “We thought a lot about how to represent a token, and so we chose a token”

1 1

Reposted by Tyler Shoemaker

Ted Underwood @tedunderwood.com · 1d

The full announcement is now up at this link. Deadline still 31 Oct.

Ted Underwood @tedunderwood.com · Aug 15

Pre-announcement of a funding opportunity for "artificial intelligence humanities sandpits," supported by AHRC, EPSRC, and SSHRC. Expressions of interest by 31 Oct; up to four grants will be funded.

Pre-announcement: artificial intelligence humanities sandpits: Canada, UK and US

AHRC and SSHRC plan to invite expressions of interest to attend a humanities-led, interdisciplinary research sandpit looking to put humanities insights and methodologies at the heart of artificial int...

www.ukri.org

7 13

Tyler Shoemaker @t-shoemaker.bsky.social · 1d

Right, lots of brush clearing to do, even as I do want to hold onto some specificity re: the synthetic. @mbfazi.bsky.social has been great on this front, e.g. www.pdcnet.org/jcp/content/...

The Computational Search for Unity: Synthesis in Generative AI - Volume 5, Issue 1, 2024

The outputs of generative artificial intelligence (generative AI) are often called “synthetic” to imply that they are not natural but artificial. Against that use of the term, this article focuses on ...

www.pdcnet.org

1 3

Tyler Shoemaker @t-shoemaker.bsky.social · 1d

(One place to start)

2

Tyler Shoemaker @t-shoemaker.bsky.social · 1d

Wanted: a theory of synthetic media

👁️ @nathanjurgenson.com · 1d

bc of AI's statistical cultural regurgitation, its rightful critiques are doomed to mimic old media theory, just with "but even more now" attached to the end

2 3

Tyler Shoemaker @t-shoemaker.bsky.social · 2d

The first in a (jam-packed) series of events on cultural AI this year from DTL. More incoming!

maybe: leif weatherby @leifw.bsky.social · 2d

The Digital Theory Lab at NYU will host Karen Hao for a book talk on Empire of AI on Monday, October 13, RSVP required here:
as.nyu.edu/research-cen...

Karen Hao on Empire of AI

Lunch will be available for the event. Event begins at 12pm

as.nyu.edu

1 3

Reposted by Tyler Shoemaker

Shannon Mattern @shannonmattern.bsky.social · 5d

My friend @rorshock.bsky.social — CS undergrad, humanities PhD — has built a brilliant prgm @ The New School: “Code as a Liberal Art” nurtures “code + computational thinking as tools for critical + creative inquiry,” and as forces for 👍+👎 social change. His spring Software Engineering class looks 🌟

Software Engineering Applications

Software Engineering Applications Spring 2026 Code as a Liberal Art, Eugene Lang College, The New School This course gives students the opportunity to experience and critically examine the software e...

docs.google.com

4 12 55

Tyler Shoemaker @t-shoemaker.bsky.social · 7d

Wall-clock time, user-cpu time, system-cpu time, and dealing-with-vLLM time

Tyler Shoemaker @t-shoemaker.bsky.social · 9d

What sublimating Turing’s proof does to a guy

Tyler Shoemaker @t-shoemaker.bsky.social · 9d

I’m somewhat of a mechinterp flâneur myself

1

Tyler Shoemaker @t-shoemaker.bsky.social · 11d

Models having their McLuhan moment

2

Tyler Shoemaker @t-shoemaker.bsky.social · 11d

This Q inspired by seeing a whole lot of task-specific outputs in supposedly "base" models, which is surely a dataset consideration but also made me wonder about the underlying thing we're now calling a "base" model

1 1

Tyler Shoemaker @t-shoemaker.bsky.social · 11d

Thanks! I also have gotten the sense that RL developments mess with the base/ft divide, even if only rhetorically so. Re: mid-training: change in what direction, do you think?

1

Tyler Shoemaker @t-shoemaker.bsky.social · 11d

Where is “fine-tuning” here?

1 10

Tyler Shoemaker @t-shoemaker.bsky.social · 11d

Feels like a @dorialexander.bsky.social question

1 4

Tyler Shoemaker @t-shoemaker.bsky.social · 11d

Genuine Q: do model builders still think in terms of LLM fine-tuning or have all the mid-/post-training recipes obsoleted this?

1 5

Tyler Shoemaker @t-shoemaker.bsky.social · 13d

+ `--quiet`

1

Tyler Shoemaker @t-shoemaker.bsky.social · 14d

Public humanities?

5

Tyler Shoemaker @t-shoemaker.bsky.social · 14d

Extremely good stuff from Leif here

maybe: leif weatherby @leifw.bsky.social · 14d

i was on Disintegrator podcast, a really awesome conversation
podcasts.apple.com/us/podcast/3...

3 2

Reposted by Tyler Shoemaker

Naomi Saphra @nsaphra.bsky.social · 14d

I did a QA with Quanta about interpretability and training dynamics! I got to talk about a bunch of research hobby horses and how I got into them.

Ben Brubaker @benbenbrubaker.bsky.social · 14d

I really enjoyed talking to @nsaphra.bsky.social about her thoughts on what much language model interpretability research misses. My latest in @quantamagazine.bsky.social:

To Understand AI, Watch How It Evolves | Quanta Magazine

Naomi Saphra thinks that most research into language models focuses too much on the finished product. She’s mining the history of their training for insights into why these systems work the way they d...

www.quantamagazine.org

2 13 66

Tyler Shoemaker @t-shoemaker.bsky.social · 15d

Increasingly so if you go by Köppen climate classification (dire)

1 1

Tyler Shoemaker @t-shoemaker.bsky.social · 15d

See you there

1 1

Reposted by Tyler Shoemaker

David Bamman @dbamman.bsky.social · 15d

The UC Berkeley School of Information is hiring an assistant professor in the broad field of Information--including areas of info seeking/retrieval, digital humanities, cultural analytics, info viz, & philosophy of information (among others). Deadline Nov 1! aprecruit.berkeley.edu/JPF05014

Assistant Professor - Information - School of Information

University of California, Berkeley is hiring. Apply now!

aprecruit.berkeley.edu

1 74 76

Tyler Shoemaker @t-shoemaker.bsky.social · 16d

A real one-two punch with this statement following the hallucinations discourse over the weekend

Ted Underwood @tedunderwood.com · 16d

The AHA was right to try to produce guidance on AI, but I wonder if it's even possible. This statement got dinged , of course, for being insufficiently critical. But they were working so hard to be critical that they wrote sentences (like the blue one) that are def false.+

bsky.app/profile/hist...

AI produces texts, images, audio, and video, not truths.

Generative AI is a remarkable technological achievement, but it has undeniable limitations. An awareness of these limitations is important for instructors and students alike. LLMs produce text using an algorithm to select each word from existing books, articles, images, and other media, including AI-created sources. AI texts do not reflect truth; rather, they echo and synthesize, sometimes poorly, sources on which the model has been trained. Generative AI reproduces the limitations of its own training material. By contrast, historians learn to identify and dissect author biases, experiences, social environment, and hidden motivations. Students need to learn to interpret AI-generated content with a critical lens, using their historical training to assess material rather than passively accept it as true or complete.

1 3