Rich Sutton warned that computational scale ultimately outperforms human expertise. Does his "bitter lesson" hold true when reproducing our own custom methods with LLMs? We find: the bitter lesson is sometimes sweet, sometimes bitter, and sometimes inapplicable. deep-culture.org/the-occasion...

It’s not just training data. It’s what gets incentivized — that’s why LLMs often write exactly like this. This isn’t common Internet speech. It’s human response ratings — and it’s rewarding the system for the appearance of nuance. ✨⬇️

Appreciate the distinction between "representational" and "operational: readings laid out by @ddzeeuw.bsky.social + @hetsalwel.bsky.social here. Resonates with what @bentarnoff.com + I have been groping towards in Musk's relationship to social media and genAI. podcasts.apple.com/us/podcast/n...

What’s the difference between latent space, vector space, feature space, and embedding space? @louravn.bsky.social spelled it out for you in this short blog post deep-culture.org/whats-the-di...

🐈‍⬛🐈New 1.54 4CAT update 🐈🐈‍⬛ github.com/digitalmetho... 📽️ Show downloaded media in the Explorer. This works for media uploads and social media datasets if images/videos have been downloaded through 4CAT. ✏️ New "Upload annotations" processor that lets you add new data fields to existing datasets. (1/2)

Social media is no longer social. Most of it is passive viewing of videos and pictures from people we've never met. But we're still studying social media like it's 2010. We've entered the post-social media era — and research needs to catch up. osf.io/preprints/so... preprint w/ Richard Rogers

enjoyed this "wayback studio" moment at @internetarchive.eu with @anthbrtn.com @iiil.li @idilg.bsky.social @hetsalwel.bsky.social @esthrr.bsky.social @digitalmethods.net @d-eepculture.bsky.social @publicdatalab.bsky.social @digitalculturekcl.bsky.social & many more! ✨

Together with Bastian August and Brogan Latil, I spent decidedly too much time trying to understand how Andrew Tate remains present on YouTube after his two channels were suspended. Full paper in New Media & Society: journals.sagepub.com/doi/10.1177/...

A small 4CAT update just in time for the holidays - v1.52! 📊 Processors for regression analysis and dataset statistics 🧭 Test LLM prompts from our prompt library with the PromptCompass processor 🤖 Use Deepseek and Gemini 3 APIs with LLM processors 🐛 Many bug fixes! ➡️ github.com/digitalmetho...

How to move along with the audiovisual, vibey, and AI-infused feeds of today's platforms? In this new article with @ddzeeuw.bsky.social and Tommaso Venturini, we theorize 'digital rhythmanalysis' as a potential new quali-quantitative approach to measure and 'feel' rhythms online: bit.ly/pns-dra

4CAT now contains everything for quite sophisticated classification pipelines with humans and (local) LLMs! I myself used it to classify what types of migration were mentioned in Dutch news articles (see www.groene.nl/artikel/onde...). See all changes below 👇

Voor @groene.nl onderzochten we hoe het migratiedebat de laatste jaren is gegroeid, verhard en verbreed. We analyseerden 150.000 nieuwsartikelen en 20.000 Tweede Kamer-spreekbeurten en gebruiken een (lokale!) LLM om te zien welke migratietypen de meeste aandacht kregen. Lees het artikel hieronder 👇

Mijn analyse over een vrolijke liberaal in een steeds rechtser land staat nu alvast online. www.groene.nl/artikel/een-....

gefeliciteerd schier

On election day, AI Overviews are still erratic, shown in ~10% of our queries. Some issue-related queries are unanswered but "which party is for defence" recommends VVD and JA21 at the top and includes outdated stances on JSF investments and the PvdA as a separate party. Nothing on NATO norms.

Genuinely what the fuck is going on in Albania

With @best.wel.moe and @nieuwsuur.nl, we studied Google's AI Overviews for political searches in the NL. We found that larger parties often benefit and that Google changed its use of AI Overviews *three times* during the run-up to the Dutch elections. Read the article here: nos.nl/nieuwsuur/ar...

Keywords are not neutral. With Tobias Blanke, we investigate the power of keywords as interpretive tools in our work "Keywords in Digital Humanities: A critical assessment of computational techniques for mapping security and freedom in historical debates". You can read the article now on DHQ!

Super fun to contribute to this brief interview with Terminally Online, a student-led organization on New Media at the University of Amsterdam. We spoke about my wonderful team‘s research (@d-eepculture.bsky.social), who inspires me, and some of my own work. 🔗: www.linkedin.com/pulse/interv...

Dit is de omgekeerde wereld. De academische vrijheid wordt bedreigd doordat onze onderzoekers die zich uitspreken in het publieke debat over “omstreden” onderwerpen als snel het doelwit worden van online bedreigingen, met name uit extreemrechtse hoek.

🚨The Critical AI Seminar Series returns in 2025/2026 ✨ This year we will have four talks and are beyond excited to welcome @amoorelouise.bsky.social & @alexcampolo.bsky.social, Fabian Offert, Helene Ratner & @nannathylstrup.bsky.social and @thaophan.bsky.social ✨ Sign up: bit.ly/uvacriticala...

The last few weeks I worked 4CAT's LLM features. We're not free from Silicon Valley APIs yet, but you have more options in terms of providers (including Mistral) and it's also possible to use local LLMs through LM Studio and Ollama. Check it out:

🔎New paper! Guillén Torres and I study "enrichment" on Google and Bing as a form of content moderation. Main finding: enrichment can be erratic, with components like "AI Overviews" even being banned for certain political queries like those including "trump". Pre-print: ssrn.com/abstract=528...

Op nos.nl ging de laatste vijftien jaar even vaak over klimaatverandering als over Erik ten Hag. (2853 vs. 2766 artikelen/updates) @xrnl.bsky.social @nieuws.nos.nl

Tired of looking at Excel rows when investigating social media data? 4CAT's new Explorer feature is here to help!

yes een nieuwe kans voor Schier

Na ons artikel uit 2021 over klimaatverslaggeving (www.groene.nl/artikel/we-h...) was het tijd voor een data-update. Gelukkig is meer aandacht voor de klimaatcrisis in de vier grootste kranten. Maar er is nog steeds een golfbeweging gaande en aandacht is de laatste maanden weer afgenomen.

We’re hiring: Lecturer in Cultural Data & AI and Archival & Information Studies (Media Studies, @uvahumanities.bsky.social)! Join @tpoell.bsky.social, @riederb.bsky.social, myself, & other wonderful colleagues in the Critical Data & AI team. Applications due 28 May. werkenbij.uva.nl/en/vacancies...

🗞️ A great deep dive in Dutch newspaper @nrc.nl, in which Elon Musk's tweets over the past few months are analysed after capturing them with our tool Zeeschuimer (tools.digitalmethods.net/zeeschuimer/)! www.nrc.nl/nieuws/2025/...

De afgelopen maanden werkte ik als AI-trainer om chatbots als ChatGPT nuttiger, vriendelijker en 'menselijker' te maken. Alles over mijn avonturen in deze 'droombaan' lees je in de @groene.nl van deze week. Met heerlijke illustraties van @kamagurka.bsky.social. www.groene.nl/artikel/heea...

Waarom gaan extreem-rechts en memecultuur zo goed samen? In deze nieuwe aflevering verwonderen ons met @hetsalwel.bsky.social over 4chan ondermediadoctoren.nl/afl-202-4chan/

New search engine scraper tool by @best.wel.moe and @digitalmethods.net! I added the feature to collect widgets from Google and Bing (it was a lot of work, please use it 😂). Feedback welcome.

we are hiring a PhD candidate for the project "Interface of Intelligence: How Generative AI Reshapes Knowledge, Media, and Interpretation"; details here: werkenbij.uva.nl/en/vacancies...

Lots of new stuff!

Throwback to my proudest citation, from "Exhaustive Survey of Rickrolling in Academic Literature" arxiv.org/pdf/2204.06826

🔍 How can creative coding techniques like soundscaping and video recomposition help us unpack political issues + influencer practices on TikTok? 🎵📹 📖 Insights from explorations w/ @publicdatalab.bsky.social around the 2024 Romanian elections at @digitalmethods.net winter school:

they're finally adding the 25th hour

Door @coenvdven.bsky.social, @rasitelibol.bsky.social en @hetsalwel.bsky.social in @groene.nl

damn

His last number of the day was #1 youtu.be/NfDniBBowyU

'subreddit tapestries' #mildlyinterestinggraphs

"Chatbots and LLMs for Internet Research?" - opening of @digitalmethods.net #dmi25 at University of Amsterdam ✨🧵https://www.digitalmethods.net/

Have to finish setting this up later but for now: new publication! Using the Birmingham School's subcultural theory to look at political compass memes, we attempt to theorize how "digital political subcultures" are refashioning political subjectivity www.tandfonline.com/doi/full/10....

ik maakte hier al wrapped grollen voordat het cool was, om het maar even gezegd te hebben

good album: open.spotify.com/album/4mZgIo...

To all the researchers here on @bsky.app for whom the topic data access from online platforms is relevant, please consider writing a short feedback on the draft of the delegated acts for the DSA: ec.europa.eu/info/law/bet...

Follow @digitalmethods.net for your healthy dose of tool updates 🦋

🍂 Release v1.46: Autumn Additions 🍂 - Add custom data sources and processors as 'extensions' (see t.ly/WQc4Z) - Import Threads data from Zeeschuimer - Import and export datasets as zip files - Crawl Telegram based on mentions and links - LLM processor for text classification And more: t.ly/psaqe

Join our cool new reading group on digital media and rhythm: asca.uva.nl/programme/re...