My and @max-web.bsky.social 's #rstats 📦 {rollama} just got an update with some bug fixes and several cool new features! Now using open weight models for text and image annotation in R is even easier!

Curious how you can use generative AI in your research without turning your workflow into a black box? Learn how to work with LLMs from prompting to fine-tuning—then apply them to real social science data like parliamentary speeches and open-ended survey responses. Apply now > bit.ly/4klFg3n

How do parties debate minorities? New study based on all parliament speeches in Germany and the UK 1980-2021, with @younghyunlee52.bsky.social , F. #Piñeyrúa, & @max-web.bsky.social, published in the edited volume Computational Social Science of Social Cohesion and Polarization:

Linguistic Polarization in Minority Representation: Analyzing Parliamentary Speeches in Germany and the UK (1980–2021)

Political polarization shapes how minority issues are discussed in public and parliamentary discourse, with implications for representation and policy. We examine the representation of migration and s...

link.springer.com

@jbgruber.bsky.social presenting rollama, an R package that brings gLLMs to your local R setup. Exciting possibilities for reproducible computational text analysis! #comptext

@max-web.bsky.social presenting our first try to use LLMs ({rollama}) to extract topics from a large corpus: the public reaction to the releases of Galactica and Chatgpt within weeks from each other in 2022 (which couldn't have been more different!) #comptext2025

New Course Alert! 🚀 Generative AI for Social Science Research This course explores how to apply AI in social science research, from basic interactions to fine-tuning models for large-scale data analysis. Find out more & apply: bit.ly/4l3nnGI #ESS2025 #GenerativeAI

deepseek is an extremly impressive open weights LLM. But the guardrails are absolutely wild 🤨

Aleksandra Urman@aurman21.bsky.social · 2y ago

To everyone assuming/saying it's just the web-based guardrail - apparently deepseek-r1:8b makes the same bs claim about Dokdo when run locally (just confirmed by @jbgruber.bsky.social who is as always faster in getting to run/test things than me 😀 You should totally follow him if you're not already)

Running Phi-4 and other LLMs in R using Google Colab for free through the rollama R-Package medium.com/@weber.aca/r...

Submission for #COMPTEXT2025 is still open until January 15! We are looking forward to receiving your proposals for papers, panels, and data presentations! We also appreciate it if you can spread the word and circulate shorturl.at/AmocX! See you at the University of Vienna on 24-26 April!

I'll get straight to the point. We trained 2 new models. Like BERT, but modern. ModernBERT. Not some hypey GenAI thing, but a proper workhorse model, for retrieval, classification, etc. Real practical stuff. It's much faster, more accurate, longer context, and more useful. 🧵

Nice paper showing just *how* irreprodroducible research with proprietary generative LLMs is. Luckily there are open source alternatives (and they are very easy to use too!)

Christopher Barrie@cbarrie.bsky.social · 2y ago

Pleased to share the latest version of my paper with Arthur Spirling and @lexipalmer.bsky.social on replication using LMs We show: 1. current applications of LMs in political science research *don't* meet basic standards of reproducibility...

Call for abstracts for a Special Volume on Computational Social Science (of the journal "Soziale Welt")! @marklutter345.bsky.social and I are looking for contributions presenting new research, discussing CSS methods, giving overviews of applications or ongoing research projects. Please circulate!

I read this paper a while back, and I liked it. But now it also feels relevant. www.nber.org/papers/w29724

The New York Times@nytimes.com · 2y ago

Google said its quantum computer needed less than five minutes to perform a mathematical calculation that one of the world’s most powerful supercomputers could not complete in 10 septillion years, a length of time that exceeds the age of the known universe.

Run LLMs locally in R with the rollama R package. The latest update simplifies query generation (Prompt) and adds support for Hugging Face models (@jbgruber.bsky.social) Check it out:

New #rstats 📦 version: {rollama} 0.2.0 What's new?🧵👇 https://cran.r-project.org/web/packages/rollama/index.html

If you find yourself with too much free time over the (long) weekend / holidays, I have ~3h Building an LLM from the Ground Up workshop on YouTube that may come in handy: m.youtube.com/watch?v=quh7...

Building LLMs from the Ground Up: A 3-hour Coding Workshop

YouTube video by Sebastian Raschka

m.youtube.com