Ted Underwood
@tedunderwood.com
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
Information Sciences and English, UIUC. Distant Horizons (Chicago, 2019). tedunderwood.com
Pinned
Ted Underwood
@tedunderwood.com
· Nov 4
Opinion | AI Is the Future. Higher Ed Should Shape It.
If we want to stay at the forefront of knowledge production, we must fit technology to our needs.
www.chronicle.com
Wrote a short piece arguing that higher ed must help steer AI. TLDR: If we outsource this to tech, we outsource our whole business. But rejectionism is basically stalling. If we want to survive, schools themselves must proactively shape AI for education & research. [1/6, unpaywalled at 5/6] +
Funny: bsky convo over the weekend was "Claude *Code* is really too restrictive; they should name it something broader so people grasp the breadth of possible uses" — et, voila!
January 12, 2026 at 9:57 PM
Funny: bsky convo over the weekend was "Claude *Code* is really too restrictive; they should name it something broader so people grasp the breadth of possible uses" — et, voila!
Reposted by Ted Underwood
the first time I heard someone say "this app is kind of ugly, looks like they didn't install the claude design skill" I thought they were joking
January 12, 2026 at 8:29 PM
the first time I heard someone say "this app is kind of ugly, looks like they didn't install the claude design skill" I thought they were joking
Reposted by Ted Underwood
A guy I knew had a local VLM that would watch his screen and warn him when he was losing focus / help with other metrics. Anyone care to build this and make it public?
January 12, 2026 at 9:15 PM
A guy I knew had a local VLM that would watch his screen and warn him when he was losing focus / help with other metrics. Anyone care to build this and make it public?
Reposted by Ted Underwood
Oh wow, deepseek is starting to make serious progress on LLMs that offload memory to external storage: github.com/deepseek-ai/...
github.com
January 12, 2026 at 6:44 PM
Oh wow, deepseek is starting to make serious progress on LLMs that offload memory to external storage: github.com/deepseek-ai/...
Reposted by Ted Underwood
Reposted by Ted Underwood
This is a Persian appreciation thread. Hope your folks are safe. Hope things come out better. Hope things don’t get worse. If you’re wishing you were home to be with family, I see you. If you’re wishing you were home to be out on the streets, I see you. If you’re glad you’re out of Iran, nice work.
January 12, 2026 at 6:02 PM
This is a Persian appreciation thread. Hope your folks are safe. Hope things come out better. Hope things don’t get worse. If you’re wishing you were home to be with family, I see you. If you’re wishing you were home to be out on the streets, I see you. If you’re glad you’re out of Iran, nice work.
Reposted by Ted Underwood
I'm teaching Python/computation for the humanities for the first time at #Duke this year; some thoughts but first a shoutout to @mellymeldubs.bsky.social for her great online syllabus & @laurenfklein.bsky.social + @dmimno.bsky.social for examples she builds on: melaniewalsh.github.io/Intro-Cultur...
Introduction to Cultural Analytics & Python — Introduction to Cultural Analytics & Python
melaniewalsh.github.io
January 12, 2026 at 5:55 PM
I'm teaching Python/computation for the humanities for the first time at #Duke this year; some thoughts but first a shoutout to @mellymeldubs.bsky.social for her great online syllabus & @laurenfklein.bsky.social + @dmimno.bsky.social for examples she builds on: melaniewalsh.github.io/Intro-Cultur...
Reposted by Ted Underwood
Things I miss from our custom client when I'm using Bluesky:
- avatar colors that show whether we're mutuals
- exportable bookmarks with custom folders
- feed of trending papers and articles
- safety alerts when a post goes viral
- researcher profiles with topics, affiliations, featured papers
- avatar colors that show whether we're mutuals
- exportable bookmarks with custom folders
- feed of trending papers and articles
- safety alerts when a post goes viral
- researcher profiles with topics, affiliations, featured papers
January 12, 2026 at 5:27 PM
Things I miss from our custom client when I'm using Bluesky:
- avatar colors that show whether we're mutuals
- exportable bookmarks with custom folders
- feed of trending papers and articles
- safety alerts when a post goes viral
- researcher profiles with topics, affiliations, featured papers
- avatar colors that show whether we're mutuals
- exportable bookmarks with custom folders
- feed of trending papers and articles
- safety alerts when a post goes viral
- researcher profiles with topics, affiliations, featured papers
Reposted by Ted Underwood
Excited to announce the Relative Adoption Metric a new way of studying model downloads that contextualizes it across time and model sizes.
While building The ATOM Project and other tools to measure the open ecosystem at Interconnects, we are often frustrated with using downloads as a primary metric
While building The ATOM Project and other tools to measure the open ecosystem at Interconnects, we are often frustrated with using downloads as a primary metric
January 12, 2026 at 4:22 PM
Excited to announce the Relative Adoption Metric a new way of studying model downloads that contextualizes it across time and model sizes.
While building The ATOM Project and other tools to measure the open ecosystem at Interconnects, we are often frustrated with using downloads as a primary metric
While building The ATOM Project and other tools to measure the open ecosystem at Interconnects, we are often frustrated with using downloads as a primary metric
Honestly this works for everything
“I want to trap hungry 19c ghosts in jars to help us with historical research” ✅
“Please read our holiday card; we got a hungry ghost to write it this year” ❌
“I want to trap hungry 19c ghosts in jars to help us with historical research” ✅
“Please read our holiday card; we got a hungry ghost to write it this year” ❌
If you can substitute "hungry ghost trapped in a jar" for "AI" in a sentence it's probably a valid use case for LLMs. Take "I have a bunch of hungry ghosts in jars, they mainly write SQL queries for me". Sure. Reasonable use case.
"My girlfriend is a hungry ghost I trapped in a jar"? No. Deranged.
"My girlfriend is a hungry ghost I trapped in a jar"? No. Deranged.
January 12, 2026 at 1:11 PM
Honestly this works for everything
“I want to trap hungry 19c ghosts in jars to help us with historical research” ✅
“Please read our holiday card; we got a hungry ghost to write it this year” ❌
“I want to trap hungry 19c ghosts in jars to help us with historical research” ✅
“Please read our holiday card; we got a hungry ghost to write it this year” ❌
Reposted by Ted Underwood
Has any research compared the content of the knowledge graph #Wikidata with natural language #Wikipedias? Lev Manovich once contrasted databases and narratives (though they are also here clearly 'natural symbionts', as Katherine Hayles said). Presumably some kinds of information is more amenable 1/
January 12, 2026 at 10:48 AM
Has any research compared the content of the knowledge graph #Wikidata with natural language #Wikipedias? Lev Manovich once contrasted databases and narratives (though they are also here clearly 'natural symbionts', as Katherine Hayles said). Presumably some kinds of information is more amenable 1/
Reposted by Ted Underwood
I want this next generation of software development tools to run on my own computer. Vim and Emacs helped democratize software development and made it accessible. We shouldn't allow this AI wave to take us backwards.
I really hope the open source community can keep pace.
I really hope the open source community can keep pace.
January 12, 2026 at 11:47 AM
I want this next generation of software development tools to run on my own computer. Vim and Emacs helped democratize software development and made it accessible. We shouldn't allow this AI wave to take us backwards.
I really hope the open source community can keep pace.
I really hope the open source community can keep pace.
Reposted by Ted Underwood
If you can substitute "hungry ghost trapped in a jar" for "AI" in a sentence it's probably a valid use case for LLMs. Take "I have a bunch of hungry ghosts in jars, they mainly write SQL queries for me". Sure. Reasonable use case.
"My girlfriend is a hungry ghost I trapped in a jar"? No. Deranged.
"My girlfriend is a hungry ghost I trapped in a jar"? No. Deranged.
August 13, 2025 at 12:56 AM
If you can substitute "hungry ghost trapped in a jar" for "AI" in a sentence it's probably a valid use case for LLMs. Take "I have a bunch of hungry ghosts in jars, they mainly write SQL queries for me". Sure. Reasonable use case.
"My girlfriend is a hungry ghost I trapped in a jar"? No. Deranged.
"My girlfriend is a hungry ghost I trapped in a jar"? No. Deranged.
Reposted by Ted Underwood
Do you have ideas for the future of reading?
Submit a 2-4 page paper to the CHI workshop I am co-organising! (deadline Feb 12) “Science and Technology for Augmenting Reading"
chi-star-workshop.github.io
Submit a 2-4 page paper to the CHI workshop I am co-organising! (deadline Feb 12) “Science and Technology for Augmenting Reading"
chi-star-workshop.github.io
January 12, 2026 at 3:46 AM
Do you have ideas for the future of reading?
Submit a 2-4 page paper to the CHI workshop I am co-organising! (deadline Feb 12) “Science and Technology for Augmenting Reading"
chi-star-workshop.github.io
Submit a 2-4 page paper to the CHI workshop I am co-organising! (deadline Feb 12) “Science and Technology for Augmenting Reading"
chi-star-workshop.github.io
Reposted by Ted Underwood
One of my favorite findings: Positional embeddings are just training wheels. They help convergence but hurt long-context generalization.
We found that if you simply delete them after pretraining and recalibrate for <1% of the original budget, you unlock massive context windows. Smarter, not harder.
We found that if you simply delete them after pretraining and recalibrate for <1% of the original budget, you unlock massive context windows. Smarter, not harder.
Introducing DroPE: Extending Context by Dropping Positional Embeddings
We found embeddings like RoPE aid training but bottleneck long-sequence generalization. Our solution’s simple: treat them as a temporary training scaffold, not a permanent necessity.
arxiv.org/abs/2512.12167
pub.sakana.ai/DroPE
We found embeddings like RoPE aid training but bottleneck long-sequence generalization. Our solution’s simple: treat them as a temporary training scaffold, not a permanent necessity.
arxiv.org/abs/2512.12167
pub.sakana.ai/DroPE
January 12, 2026 at 4:12 AM
One of my favorite findings: Positional embeddings are just training wheels. They help convergence but hurt long-context generalization.
We found that if you simply delete them after pretraining and recalibrate for <1% of the original budget, you unlock massive context windows. Smarter, not harder.
We found that if you simply delete them after pretraining and recalibrate for <1% of the original budget, you unlock massive context windows. Smarter, not harder.
Reposted by Ted Underwood
yeah this feels correct. it will be net negative for us but net positive for the world (I hope)
I think a tipping point is coming for bsky — but I also think, after it tips, we will miss the candor of a platform that was too small to be worth engagement farming
The Telegraph is reporting that the UK is seeking Canada and Australia to join together to create a unified ban on X dot com
January 11, 2026 at 9:03 PM
yeah this feels correct. it will be net negative for us but net positive for the world (I hope)
clapping gloved hands belong to your humble narrator
At least 25-minute long car rally against ICE organized yesterday on FB by the Champaign-Urbana Resistance Effort
They organized the theme: yellow, with streamers, driving across town & up to the Lowe’s parking lot
They organized the theme: yellow, with streamers, driving across town & up to the Lowe’s parking lot
January 11, 2026 at 8:33 PM
clapping gloved hands belong to your humble narrator
The psychological drain of using AI, for me, is that its intellectual successes are, emotionally, empty calories.
Whenever the model solves a big problem I want to stop working and go tell someone (human) "look how good this turned out!" With a human collaborator, we'd just tell each other.
Whenever the model solves a big problem I want to stop working and go tell someone (human) "look how good this turned out!" With a human collaborator, we'd just tell each other.
one thing i don’t see talked about a lot is the psychological aftereffects of exposure to AI “for work”. i actually don’t know if we know what they are.
i don’t mean psychotic episodes it triggers in some or the delusions etc.
i feel like i get some kind of… hangover of uncanny. like i ate plastic
i don’t mean psychotic episodes it triggers in some or the delusions etc.
i feel like i get some kind of… hangover of uncanny. like i ate plastic
January 11, 2026 at 7:08 PM
The psychological drain of using AI, for me, is that its intellectual successes are, emotionally, empty calories.
Whenever the model solves a big problem I want to stop working and go tell someone (human) "look how good this turned out!" With a human collaborator, we'd just tell each other.
Whenever the model solves a big problem I want to stop working and go tell someone (human) "look how good this turned out!" With a human collaborator, we'd just tell each other.
Reposted by Ted Underwood
The combo of improvements in reasoning efficiency (fewer tokens per answer, still very new research area) and faster chips is going to make coding agents so so much faster in 6-12 months.
The products in 2+ years will feel approx instantaneous relative to today.
The products in 2+ years will feel approx instantaneous relative to today.
January 11, 2026 at 6:04 PM
The combo of improvements in reasoning efficiency (fewer tokens per answer, still very new research area) and faster chips is going to make coding agents so so much faster in 6-12 months.
The products in 2+ years will feel approx instantaneous relative to today.
The products in 2+ years will feel approx instantaneous relative to today.
Reposted by Ted Underwood
yeah, i'm not sure I share Dan's exact symptoms but i do think this is a thing and we should better understand it
one thing i don’t see talked about a lot is the psychological aftereffects of exposure to AI “for work”. i actually don’t know if we know what they are.
i don’t mean psychotic episodes it triggers in some or the delusions etc.
i feel like i get some kind of… hangover of uncanny. like i ate plastic
i don’t mean psychotic episodes it triggers in some or the delusions etc.
i feel like i get some kind of… hangover of uncanny. like i ate plastic
January 11, 2026 at 5:34 PM
yeah, i'm not sure I share Dan's exact symptoms but i do think this is a thing and we should better understand it
Reposted by Ted Underwood
I think a tipping point is coming for bsky — but I also think, after it tips, we will miss the candor of a platform that was too small to be worth engagement farming
The Telegraph is reporting that the UK is seeking Canada and Australia to join together to create a unified ban on X dot com
Starmer rallies international support to take on Musk
Tech billionaire says threat to ban social media site over sexualised AI images is ‘censorship’
www.telegraph.co.uk
January 11, 2026 at 3:35 PM
I think a tipping point is coming for bsky — but I also think, after it tips, we will miss the candor of a platform that was too small to be worth engagement farming
Reposted by Ted Underwood
New research on Gifted dogs is out, in @science.org!
Huge congrats to SHANY DROR for her effort and this incredible achievement.
📄 www.science.org/doi/10.1126/...
Huge congrats to SHANY DROR for her effort and this incredible achievement.
📄 www.science.org/doi/10.1126/...
Dogs with a large vocabulary of object labels learn new labels by overhearing like 1.5-year-old infants
Children as young as 18 months can acquire novel words by overhearing third-party interactions. Demonstrating similar learning processes in nonhuman species would indicate that the social-cognitive sk...
www.science.org
January 9, 2026 at 1:06 PM
New research on Gifted dogs is out, in @science.org!
Huge congrats to SHANY DROR for her effort and this incredible achievement.
📄 www.science.org/doi/10.1126/...
Huge congrats to SHANY DROR for her effort and this incredible achievement.
📄 www.science.org/doi/10.1126/...
Reposted by Ted Underwood
Trying to persuade your institution, employer, news source, school, etc., to quit X? Looking for a list of organizations & individuals who've already done exactly that to help with the persuasion? Well, here you go! Excellent and useful initiative from @eloquence.bsky.social 🙏
Spread it around:
Spread it around:
January 11, 2026 at 1:36 PM
Trying to persuade your institution, employer, news source, school, etc., to quit X? Looking for a list of organizations & individuals who've already done exactly that to help with the persuasion? Well, here you go! Excellent and useful initiative from @eloquence.bsky.social 🙏
Spread it around:
Spread it around:
Reposted by Ted Underwood
the world is so pretty
January 11, 2026 at 12:39 PM
the world is so pretty