the purpose of a first draft is to exist the purpose of a first draft is to exist the purpose of a first draft is to exist

Excited to present a new preprint with @nkgarg.bsky.social: presenting usage statistics and observational findings from Paper Skygest in the first six months of deployment! 🎉📜 arxiv.org/abs/2601.04253

Excited to announce the publication of our article "AI Unplugged: Exploring Pathways from Physical Simulation to Conceptualization of AI Reasoning Processes" in @acmtoce.bsky.social. We present 4 activities and explore how they support conceptualization of AI reasoning. dl.acm.org/doi/10.1145/...

> We archived around 86M music files, representing around 99.6% of listens. It’s a little under 300TB. This is the largest music metadata database that is publicly available. Many interesting 📊 charts on this page that can only be made by having this scale of data. annas-archive.li/blog/backing...

another example of AI as tracer dye, exposing an already existing problem (and, of course, significantly worsening it)

that doesn’t solve the issue of students, journalists, etc. finding and citing fake academic papers, of course—but truly, it’s not an LLM’s fault when professional scholars do this—they are just efficient catalysts for a chemical reaction already underway

Doing this in my head after a big meal and too much beer, so I could be off, but I think Voyager 1 will reach one (1) light day from Earth in October or November of next year, and one light day from the Sun in January 2027. 🧪 🔭

Training LLMs end to end is hard. But way more people should, and will, be doing it in the future. The @hf.co Research team is excited to share their new e-book that covers the full pipeline: · pre-training, · post-training, · infra. 200+ pages of what worked and what didn’t. ⤵️

Halloween blog post: Italian Giallo Horror Films thisismattmiller.com/post/giallo/ - Using vision language model to analyze a 70 film corpus (🧟) / 80,000 frames - Build and plot “trope clusters” across movies Probably the longest eye acting supercut you've seen: youtu.be/cGrmkOwut6k

📣 Attention science + tech journalists! Trying to keep up with the flood of research papers that come out every day? Attempting to track what research has already been covered by others, and what may benefit from deeper exploration? You are invited to participate in our research study! [1/3]

📣 Attention science + tech journalists! Trying to keep up with the flood of research papers that come out every day? Attempting to track what research has already been covered by others, and what may benefit from deeper exploration? You are invited to participate in our research study! [1/3]

One of the coolest video game preservation stories I have ever covered arstechnica.com/gaming/2025/...

“Editing according to traffic is not editing; it’s engineering,” nymag.com/intelligence...

What are your favorite recent papers on using LMs for annotation (especially in a loop with human annotators), synthetic data for task-specific prediction, active learning, and similar? Looking for practical methods for settings where human annotations are costly. A few examples in thread ↴

Check out the camera-ready version of our ACL Findings paper ("Taxonomizing Representational Harms using Speech Act Theory") to learn more!!! arxiv.org/pdf/2504.00928

1. On Saturday, @mnlee.bsky.social will present our work on designing for agency in LLM-infused writing support tools for journalists at the Tools for Thought workshop. Lead author @sachnishal.bsky.social will be nearby at her own workshop on news + HCI! ai-tools-for-thought.github.io/workshop/

sometimes when i'm writing a sentence i really like it feels like i'm playing the piano

Remember this @neuripsconf.bsky.social workshop paper? We spent the past month writing a newer, better, longer version!!! You can find it online here: arxiv.org/abs/2502.00561

Holy moly. I'm trying to write an academic paper, and nearly every application I'm using is not only offering Generative AI as an option for writing, but *pushing it* -- pervading the design to the point where a simple misclick would make my content AI-generated. Here's why that's a problem. 🧵

Incredibly amped to co-organize this #CHI2025 workshop on the future of news and HCI 📰📺🌐 Join us in exploring how HCI can support the production, verification and dissemination of trustworthy news in a changing world! 🔗 Details: sites.google.com/view/newsfut... ⏳ Deadline: Feb 20, 2025 (AoE)

can connect playlists meant to deliver indifferent content (e.g. "bossa nova dinner") to the spurious AI-generated genres individualized for each user in Spotify Wrapped; these "genres" have no social meaning and can't be learned or shared, only experienced through more streaming

Seeing cool works on metrology and measurement modeling for NLP! So I wanted to port over the thread our ACL 2023 Findings paper (arxiv.org/abs/2305.09022) on conceptualizations of NLP tasks and measurements of performance! Work with Eric Yuan, @haldaume3.bsky.social, and Su Lin Blodgett. (1/n)

was mildly obsessed with this genre of ad all summer and took a bunch of photos for giggles. I feel they exemplify well this idea of "AI as floating signifier" from Suchman (2023), since they "suggest[s] a specific referent but work[s] to escape definition in order to maximize its suggestive power"

Feel like "fix the motion smoothing" should be replaced by "explain that ChatGPT isn't a search engine" when going home for the holidays now

PSA, your local post office probably has Ursula K LeGuin stamps or you can order them by mail.

This, from 2015, was another longread for Flavorwire, on the importance of physical media. Another ongoing hobby horse of mine! I've really just been writing and yelling about the same things for a decade!! www.flavorwire.com/535883/the-p...

The best writers in the world. go.bsky.app/8fxDPpA

I love how I just discovered I have a folder called "READ NOW!!!!" and I just scroll pass it. Untouched since October last year 🤣

Earlier this year, I decided to try tracing character journeys from each Sherlock Holmes story on London's 1893 Ordnance Survey #maps. It's great fun. As the street structure has remained largely the same, you can sometimes use ORS tools for routing between points. This one's the Red Headed League.

Have finally coded a Small Language model. Responds to any input with “wow, that’s a lot of words” and then shuts down

📣 Attention freelance science & tech journalists! 🔬🧪💻 We're recruiting participants for a study about how generative AI could reshape journalists' work. Join us for a 60-80 mins interview and share your thoughts on generative AI tools that can support pitch writing! (1/3)

My latest at @flaminghydra.com —for our "Lost Internet" series—is a meditation on the zen of Zombocom:

📣 Attention freelance science & tech journalists! 🔬🧪💻 We're recruiting participants for a study about how generative AI could reshape journalists' work. Join us for a 60-80 mins interview and share your thoughts on generative AI tools that can support pitch writing! (1/3)

Taylorism is a management philosophy based on using scientific optimization to maximize labor productivity and economic efficiency. Here's the result of making the false Taylorist assumption that the output of scientific research is scientific papers—the more, faster, and cheaper, the better.

Here's the weird Taylorism again. The system produces work at the level of an early trainee requiring substantive supervision. This is not good ROI for producing papers. The primary output of time invested in trainee research is the development of independent scientists—not the research papers.

Rob Horning is such a great resource for making sense of what is happening with technology. This is a great piece on the tendency to add an AI generated chatbot to every website and the dream, or nightmare, of replacing human contact with AI "friends." open.substack.com/pub/robhorni...

Indignity Vol. 4, No. 129, FAN SERVICE JOURNALISM DEP'T.: What Retread Marvel Actor Should Play the Best Supervillain of All? www.indignity.net/doom-loop/

Poetry is weirdly prominent in LLM conversations. But what do models really "know" about poetry? We tested how well LLMs can recognize 20+ poetic forms in English & probed major pretraining datasets to see which poems might be memorized. New preprint: arxiv.org/abs/2406.18906

Hmmmm.

Hacks is a great comedy series that is going to lose Best Comedy to a drama (The Bear) that evades classification as a drama because it would lose Best Drama to Shōgun, a limited series that won't compete for Best Limited Series so that Fargo, a comedy series, has a chance there. Enjoy the Emmys!

*ahem* Some personal news!:

Dear #datascience, If you are still running tSNE and KNN for dimensionality reduction and clustering, you need to hear the Good Word of UMAP and HDBSCAN. umap-learn.readthedocs.io/en/latest/ Please welcome their co-creator @lelandmcinnes.bsky.social! 👋🧪

I wrote about a cool experiment we ran using GPT3 to generate news angles for jargon-y science articles! feedback and ideas appreciated!

Do crowdworkers use ChatGPT to write their responses? I'm still not sure, but when I asked on Reddit, I got a flood of fascinating responses from the workers themselves, including some practical tips for researchers looking to prevent this.

jazzed to attend #CSCW2023 this year and hang out at the workshop on community-collaborative approaches in HCI! if you're into participatory design or journalism, let's chat about how to realistically and sustainably build cool stuff with the people who will use it!

“Never tried ____ before but this is probably it” is the perfect, sentence-level articulation of the AI hype bubble.