Laura K. Nelson

@lauraknelson.bsky.social

431

Followers

175

Following

313

Posts

Associate Professor @ UBC
Computational Sociology
Machine Learning is Feminist

You only have to look at the Medusa straight on to see her. And she’s not deadly. She’s beautiful and she’s laughing.

www.lauraknelson.com

Posts Replies Media

Reposted by

Laura K. Nelson

What kind of illiberal lunacy is this? www.theguardian.com/politics/art... Think of who you don't want in power. Now imagine them enforcing this law. A society without misinformation is inherently unfree (and impossible).

Welsh government commits to making lying in politics illegalwww.theguardian.com Labour administration says ‘globally pioneering’ legislation will be brought in before next Senedd elections in 2026

I' looking for example applications of genAI/LLMs in the humanities or social sciences that would be intelligible to undergrads, and maybe ones that could be a framework for an assignment or small project. Not the "generate an essay then critique it" type. Analytical/scholarly applications?

The start of the academic fiscal year means it's time to officially update my website 🎉 www.lauraknelson.com

Reposted by

Laura K. Nelson

📢 #CHR2024 Deadline Alert! 📅 July 8 is just around the corner for paper submissions to the Computational Humanities Research conference in Aarhus. 🏃‍♀️💨 Polish those drafts or sprint to the finish line! More details at: 2024.computational-humanities-research.org/cfp/ #ComputationalHumanities

Computational Humanities Research 20242024.computational-humanities-research.org

I lived years in SF's Hayes Valley, the arts district. I was friends with an opera lighting tech who gave me comps and symphony rush tickets were $20 so those years I saw every single program at both (!). This is major. What, and I mean what, is going on with SF arts? www.nytimes.com/2024/06/24/a...

Discord at the Symphony: Losing a Star, San Francisco Weighs Its Futurewww.nytimes.com The struggles of one of the nation’s finest orchestras show the difficulties facing classical music in the United States.

My open source LLM testing continues. Llama3 is still performing great 👍 I tested Mistral - it's not terrible but it's much worse than Llama3. Of the open source models in Ollama, which other ones are worth testing? 👀 👀 ( MLSky DH CSS blueskAI 🤖 )

Reposted by

Laura K. Nelson

We’re fighting to restore access to more than 500,000 books in court on Friday. You can show your support in a number of ways - learn more: blog.archive.org/2024/06/26/w...

Reposted by

Laura K. Nelson

Interdisciplinary researchers, please take our survey! ▶️ forms.gle/XULo3xTbkm9n... Just 6 questions, should take just 2-3 minutes, and we would be so grateful for your input 🙏 Calling digital humanists, computational social scientists, and anyone doing research in multiple communities!

Writing across research communitiesforms.gle We are researchers at the Allen Institute for AI interested in studying writing cultures across research communities. A research community includes researchers with shared topics of interest, norms,...

Reposted by

Laura K. Nelson

Reply to

Frederic R. Hopp

3/3 Starting in October, I am looking for a PhD student (75% TV-L E13, 3+1 years) interested in moral psychology, machine/deep learning, and data science. Please RETWEET: leibniz-psychology.onlyfy.jobs/job/90o408mf

Three Doctoral Researchers (m/f/d)leibniz-psychology.onlyfy.jobs Become part of the open science movement

Reposted by

Laura K. Nelson

I'm looking for a good example of LLM fine-tuning + preference learning that's compelling to students but small enough to run as an in-class demo. Ideas?

Reposted by

Laura K. Nelson

I’m really impressed by llama3 in ollama for Humanities applications. Thinking of adding a prompt collection to aiforhumanists.com, would people contribute?

Update 3: A very generous colleague gave me access to his BEAUTIFUL Linux machine. Llama3:70b is blazing fast on it. Ooh boy time to play!

Update 2: Via @jwlockhart.bsky.social 's recommendation, I tried a sample on Llama3:70b. It's miserably slow on my machine, so I regret to report that the accuracy is much improved. Off to find a beefier machine! Llama 3 (70b): Accuracy: 0.79 Precision: 0.85 Recall: 0.78 F1 score: 0.82

Reposted by

Laura K. Nelson

Reply to

Laura McGrath

We've got a #DataSittersClub all about going on your first archival visit, written by Cadence Cordell! datasittersclub.github.io/site/dsc17.h...

DSC #17: Cadence’s Archives Mysterydatasittersclub.github.io by Cadence Cordell September 21, 2022 https://doi.org/10.25740/sd796vb8535 Preface (Quinn): If it weren’t for the fact that my friend Shannon Supple is the curator of the Ann M. Martin Papers at Sm...

Last night I added an event to my calendar to remind myself to do a thing today. I forgot to add what the thing is but thought I'll easily remember what it is, I just need the ping. So I didn't go back and add details. Today I have absolutely no idea what the thing I'm supposed to do is. Whomp whomp

#BookChallenge The challenge is to choose 20 books that greatly influenced you. One book per day, for 20 days. No explanations, no reviews, just covers. Join in! 💙📚👀 Day 1

Update: My first try on the 8b model achieved roughly comparable results to GPT-4. Llama 3 (8b): Accuracy: 0.69 Precision: 0.81 Recall: 0.66 F1 score: 0.73 GPT-4: Accuracy: 0.72 Precision: 0.83 Recall: 0.69 F1 score: 0.76

Got Llama 3 running on my machine this morning and started annotating text with it. Y'all, this is my qualitative coding dream. A machine that's 100% reliable and I can iterate with in English? And then it codes text as I work? You can't convince me this isn't the future of qualitative coding.

Reposted by

Laura K. Nelson

Laura’s post prompted me to try Llama 3 for tagging Viral Text cluster genres (like I described with GPT-4 in this piece ryancordell.org/research/sci...) & I’m pretty impressed with the results. Given that I can run this model locally (& free) it’s an appealing option

The Scissors, the Paste-Pot, and the Large Language Model | Ryan C. Cordellryancordell.org Book history, digital humanities, old newspapers, and information sciences

Reposted by

Laura K. Nelson

Modern generative models are trained to imitate human experts, but can they actually beat those experts? Our new paper uses imitative chess agents to explore when a model can "transcend" its training distribution and outperform every human it's trained on. arxiv.org/abs/2406.11741

You guys. The "Trump Critic Troll". It's Tim Gill. I'm wheezing, dying, dead. Tim Gill is an absolute master of the craft at this point 💀💀💀

Reposted by

Laura K. Nelson

NEWS: After sustained attacks from House Republicans, the Stanford Internet Observatory is being dismantled. A huge blow to academic freedom and our ability to understand platforms and influence operations www.platformer.news/stanford-int...

Reposted by

Laura K. Nelson

This is one of the projects I'm most proud about. We often talk about culture as a rugged landscape. Using longitudinal data, we built that landscape and tallied up how folks moved across it. Turns out the organization of the landscape-- its ruggedness --is predictive of the transitions people make!

we updated our fun paper with @nicolasrestrepo.bsky.social, that formalizes a "cultural landscape" as a feature space, or simply, a string of 1s and 0s---a simple protocol that maps religious groups, political beliefs, or writing instruments in one basic framework:

OSFosf.io

Reposted by

Laura K. Nelson

An Angus Reid survey shows lots of people are thinking of leaving BC over housing costs. @jensvb.bsky.social and I work through some of the - at times surprising - implications, including why fewer people leave Metro Vancouver than you might expect. homefreesociology.com/2024/06/11/m...

Migration and Housing Costshomefreesociology.com Co-authored with Jens von Bergmann and cross-posted at MountainMath New data supports a common theme: Housing costs seem to be increasingly important as a determinant of long-distance migration, addin...

Wow this issue looks excellent! 👀

For those asking bout digital access to the new POETICS TODAY issue on AI, with excellent pieces by @hannesbajohr.de, @ritaraley.bsky.social, and numerous others, including 👀 myself: read.dukeupress.edu/poetics-toda...

Volume 45 Issue 2 | Poetics Today | Duke University Pressread.dukeupress.edu

Reposted by

Laura K. Nelson

A team centered at McGill and Harvard argues that LLM "hallucination" is better understood as confabulation, and resembles human habits of using narrative for sense-making. #machinelearning 🤖 arxiv.org/abs/2406.04175

Confabulation: The Surprising Value of Large Language Model Hallucinationsarxiv.org This paper presents a systematic defense of large language model (LLM) hallucinations or 'confabulations' as a potential resource instead of a categorically negative pitfall. The standard view is...

Undergrads hate email - they MUCH prefer, as do I, to communicate over chat (Teams, as we're a Microsoft campus [bleh]). Staff/admin, all on Teams. But I can not, for the life of me, get my fellow faculty to do *anything* on Teams. WHY? Is this a my uni thing, or a universal faculty thing?

Four year postdoc at the University of Zurich 👀 That's an amazing gig for the right person

🚨 4-year postdoc position in computational social science at University of Zurich. Come work with me and @fgilardi.bsky.social at @ipz.bsky.social ! Application deadline is June 30. Job ad & details: ⬇️ www.ipz.uzh.ch/de/ueber-uns... polisky csssky socsky commsky

Reposted by

Laura K. Nelson

Are you interested in learning about: - doing digital history with British newspapers & maps? - thinking about what it means for historians to re-use data & research software? - how to do open & reproducible historical research? Join us for our final 2024 workshop 30 Sept-2 Oct in London.

Autumn Data/Culture Workshopwww.turing.ac.uk This workshop aims to bring together historians and others with an interest in using big, digitised historical map and newspaper collections as primary sourc

EXCELLENT new paper by Ed Amenta, Neal Caren (@haphazardsoc.bsky.social), and Weijun Yuan. The evidence is mounting: social movement organizations are mentioned in the news much more often doing things that are *not* protests. We gotta shift our paradigm! onlinelibrary.wiley.com/share/TBTAW7...

Beyond the protest paradigm: Four types of news coverage and America's most prominent social movement organizationsonlinelibrary.wiley.com You have to enable JavaScript in your browser's settings in order to use the eReader.