Ted Underwood's Avatar

Ted Underwood

@tedunderwood.com.bsky.social

Uses machine learning to study literary imagination, and vice-versa. Likely to share news about AI & computational social science / Sozialwissenschaft / 社会科学 Information Sciences and English, UIUC. Distant Horizons (Chicago, 2019). tedunderwood.com

19,892 Followers  |  5,603 Following  |  18,063 Posts  |  Joined: 10.04.2023  |  2.4507

Latest posts by tedunderwood.com on Bluesky

I'm just being a jerk

07.10.2025 19:48 — 👍 1    🔁 0    💬 0    📌 0

Yes. But in case there are children listening I prefer to express this as (cough, ₚₐᵣᵢᵃʰ ₛₜₐᵗᵉ).

07.10.2025 19:40 — 👍 3    🔁 0    💬 1    📌 0

Love this so much! Whenever a chatbot talks about "our Pleistocene ancestors" I cough and interrupt to correct the pronoun. Boundaries have to be maintained.

07.10.2025 19:24 — 👍 12    🔁 3    💬 1    📌 0

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

07.10.2025 19:06 — 👍 9    🔁 7    💬 0    📌 0

Econ and law are shameless. I thought we were bad in the humanities, because I was comparing us to CS — but I didn't even know about econ and law.

07.10.2025 15:43 — 👍 4    🔁 0    💬 1    📌 0
Preview
Evolution Strategies at Scale: LLM Fine-Tuning Beyond Reinforcement Learning Fine-tuning pre-trained large language models (LLMs) for down-stream tasks is a critical step in the AI deployment pipeline. Reinforcement learning (RL) is arguably the most prominent fine-tuning meth...

Evolution Strategies at Scale: LLM Fine-Tuning Beyond Reinforcement Learning

07.10.2025 15:07 — 👍 11    🔁 1    💬 0    📌 1

Appropriately fractal. We need to add an Ed. Mandelbrot to it.

07.10.2025 14:59 — 👍 4    🔁 0    💬 0    📌 0

Totally. I get why it happens, and it’s systemic, not just an individual moral failing.

07.10.2025 14:50 — 👍 1    🔁 0    💬 0    📌 0

Unless the article is fully generated (not likely) this is a symptom of a much older problem, which is that people aren’t reading many of the sources they cite — just gesturing at them.

07.10.2025 14:43 — 👍 30    🔁 4    💬 5    📌 1

Exactly! *that’s* how fake she is

07.10.2025 14:02 — 👍 0    🔁 0    💬 0    📌 0

Approximately no one’s seen “Tilly Norwood” in anything, but the story took off because it had the classic recipe: condense a diffuse problem into a specific woman we can judge. Is she too manipulative? Too fake? Is she the perfect symbol of what’s wrong with us?

07.10.2025 13:55 — 👍 24    🔁 1    💬 1    📌 0

👨‍🍳😘

07.10.2025 13:08 — 👍 0    🔁 0    💬 0    📌 0
Preview
MrBeast says AI could threaten creators' livelihoods, calling it 'scary times' for the industry | TechCrunch If MrBeast is openly questioning whether AI is an existential threat to his business and others like it, then it's fair to say that smaller creators are likely even more worried.

MrBeast says AI content slop threatens man-made content slop

07.10.2025 13:06 — 👍 23    🔁 3    💬 1    📌 6

To be sure, text is not literally a neural network. What I’m quibbling about is the notion that we would conceive it mainly as a source of bias or contamination. All the capacities of the network come from the data, whether the effects are positive negative or unknowable.

07.10.2025 11:08 — 👍 4    🔁 1    💬 1    📌 0

The data doesn’t influence the NN; it *is* the NN. All the capacities of the model come from the data.

07.10.2025 10:56 — 👍 5    🔁 0    💬 1    📌 0

I guess I’m a poet
And I can’t help but display it

07.10.2025 10:07 — 👍 6    🔁 0    💬 0    📌 0

Question que je me suis toujours posée : est-ce qu'avoir des problèmes de vue conduit à devenir opticien-ne, ou est-ce que tou-te-s les opticien-ne-s portent des lunettes même sans en avoir besoin pour servir de modèles publicitaires cheap à leur magasin

07.10.2025 09:08 — 👍 7    🔁 3    💬 5    📌 0

David Mamet on the other hand is great at plot :)

07.10.2025 09:49 — 👍 5    🔁 0    💬 0    📌 0

I was saying to a class yesterday that the job of a novelist is to torture their characters and then make sure that when they escape it only gets them into an even worse situation.

I actually think some of the nicer writers (like KSR) have problems with plot because they’re too nice.

07.10.2025 09:47 — 👍 7    🔁 0    💬 1    📌 0

Which I understand! my own business model is in the process of being disrupted and I’m not thrilled either. But I expect SF writers and academics to have a level of self-awareness about change and resentment of change: it’s in their job description.

07.10.2025 09:31 — 👍 5    🔁 0    💬 1    📌 0

I think what we’re actually saying when we talk about “ceding ancient things to These People” is just that there’s competition and bad blood between tech on the one hand and journalism/academia/creative media on the other. The internet messed up a bunch of business models, and ppl are still pissed.

07.10.2025 09:31 — 👍 5    🔁 0    💬 1    📌 0

I mean what does “most unlikable people on earth” actually mean here? Are the teams at Google working on LLMs actually worse people than the teams working across the hall on Maps or AlphaFold? How do we know? The CEOs may be slippery liars, but in what industry are CEOs better? +

07.10.2025 09:31 — 👍 5    🔁 0    💬 2    📌 0

It’s shocking to me, actually, that our intellectual traditions proved so powerless here.

100 years of fiction about thinking machines that blur the boundaries of … now, wait just a second! I didn’t mean you could *actually* blur those boundaries

Same story with Foucauldians

07.10.2025 04:03 — 👍 23    🔁 6    💬 2    📌 1

Video will become like other forms of testimony. It's kind of shocking how well we compensate for the fact that anyone can just say anything. People lie all the time, but specific testimony is still useful and can be correlated with other evidence.

07.10.2025 02:44 — 👍 35    🔁 3    💬 4    📌 0

Mayyybe Kim Stanley Robinson

07.10.2025 02:42 — 👍 5    🔁 0    💬 1    📌 0

Robin Sloan

07.10.2025 02:41 — 👍 4    🔁 0    💬 1    📌 0

the constraints of 300 characters to make a clear, cogent, point likely improves both your thinking and writing. to a point. that point is the set of ideas that cannot be contained in 300 characters without significant information loss to the audience. contra most, imo that set of ideas is tiny.

07.10.2025 00:42 — 👍 22    🔁 2    💬 2    📌 0
Art by Charles R. Knight

Art by Charles R. Knight

Art by Jay Matternes

Art by Jay Matternes

Screen capture from Prehistoric Park

Screen capture from Prehistoric Park

Art by Rudolf Hima

Art by Rudolf Hima

Forget King Kong, forget Spinosaurus. In honor of the 120th anniversary of the naming of Tyrannosaurus rex...we’re going to bash our cheap plastic dinosaur toys together and talk about a fight between T. rex and Triceratops.

(Artist credits in alt-text)

06.10.2025 00:50 — 👍 62    🔁 21    💬 2    📌 2

We absolutely do

07.10.2025 01:23 — 👍 2    🔁 0    💬 0    📌 0

It’s come to my attention that some of you don’t know/forgot that Bari Weiss dated Kate McKinnon. This is a crucial element of her lore.

07.10.2025 00:52 — 👍 82    🔁 10    💬 10    📌 3

@tedunderwood.com is following 20 prominent accounts