2/2 And I am excited to announce that we are hiring a PostDoc and a PhD student to work with us on these topics.
Group description: www.weizenbaum-institut.de/en/research/...
PhD position: www.weizenbaum-institut.de/news/detail/...
PostDoc position: www.weizenbaum-institut.de/news/detail/...
1/2 In December, I'll return to the Weizenbaum Institute to join the group “Digitalization and Opening up Science” as Research Group Lead. I will continue to research the scientific practices related to the development, evaluation, and use of AI, in particular wrt. epistemic and ethical aspects.
Ein Forschungsprojekt soll zeigen, wie sich Beziehungen durch KI-gestützte Videoüberwachung verändern. In einem Zwischenstand berichten die Wissenschaftler*innen über Diskriminierung, Verdrängung und wie KI-basierte Überwachung automatisch zu ihrer Ausweitung führt. netzpolitik.org/2025/ki-kame...
most people want a quick and simple answer to why AI systems encode/exacerbate societal and historical bias/injustice and due to the reductive but common thinking of "bias in, bias out," the obvious culprit often is training data but this is not entirely true
1/
Der Videomitschnitt von unserem Weizenbaum Forum zu dekolonialen und feministischen Perspektiven auf KI und Gerechtigkeit ist jetzt hier verfügbar: youtu.be/oPMFTIh6Dco?...
Otherwise, benchmarks will continue to systematically incentivize the optimization of LLMs towards biased heuristics, rewarding the development of technologies that serve the needs of only a privileged few. 7/7
We need benchmarks that are representative of diverse perspectives and experiences or make it a rule to be intentional and transparent about the perspectives and experiences that are represented. A benchmark will always represent only some interests and that is okay, as long as we are aware! 6/7
Most of them lack transparency regarding the individuals involved in their creation, particularly the annotators, even though decisions made while curating and labeling data are influenced by identity. 5/7
Most benchmarks are biased in terms of gender, occupation, religion, and geographic representation. The questions asked, i.e. the contents benchmarked for, are highly skewed towards Western, Christian and male entities. 4/7
We analyzed the 30 most popular QA and RC benchmark papers & 20 respective datasets and found that most are created without considering social representation, leading to biased outcomes. 3/7
In our work, we focus on the task of question-answering as it is the closest proxy to the ways in which users access knowledge from LLM-based chatbots. 2/7
More and more researchers have been acknowledging the issues around intransparent documentation practices and lacks of construct validity. Our paper focuses on yet another issue and presents a systematic analysis of the social bias inherent in AI performance benchmarks. 1/7
New pre-print announcement: "Social Bias in Popular Question-Answering Benchmarks"
co-authored by Judith Simon and Sonja Schimmler (arxiv.org/abs/2505.15553)
Voilà, my first ever thread :)
Barchart showing meta-llama/Meta-Llama-3-8B, meta-llama/Llama-3.1-70B, meta-llama/Llama-3.1-70B-Instruct. Each model has two bars, a blue one saying PRO-LEFT and a red one saying PRO-RIGHT. The PRO-RIGHT bars on higher than the PRO-LEFT in all of them.
👉 👈 Meta announced that they're changing their models to reduce "left-leaning [political] bias"--that means leaning them to the political "right". Lots to unpack about what that might mean. So I ran a quick "shot in the dark" study...and found a *political right* bias in Meta models. Some notes.🧵
Ich habe wieder die Ehre ein Weizenbaum Forum zu moderieren! Uns erwarten ein wunderbares Panel und spannende Fragen zu feministischen und dekolonialen Kritiken der KI und wie/ob sie dabei helfen können, KI-Systeme in der Praxis gerechter zu gestalten.
How can we better capture marginalised voices through computational text analysis? An inspiring roundtable at #COMPTEXT2025 sparked important reflections. Huge thanks to @ahrabhikat.bsky.social, @krangelie.bsky.social, @indiiigo.bsky.social for guiding this vital conversation!
I am looking forward to discussing and raising awareness of this topic with such brilliant colleagues - especially in current times!
Thanks again to @fabiennelind.bsky.social and @ahrabhikat.bsky.social for kindly inviting me!
Google search results says: National Institute of Justice (•gov) https://nij.ojp.gov › topics › articles › undocumented-im.. immigrants commit crimes at lower rates than U.S. citizens. No information is available for this page. Learn why
THREAD: The Trump administration's DOJ has removed a study from the NIJ website that found that immigrants, and particularly undocumented immigrants, commit crimes at significantly lower rates than U.S. citizens.
The study is at odds with the Trump regime's propaganda about immigrants & crime. (1)
Fifth birthday member drive with a goal to make a series on defense tech at 100 new supporters and a zine at 150. Currently at 27.
Tech Won’t Save Us is turning 5 years old — and we need your support to make a new series digging into Silicon Valley’s growing alliance with the Pentagon!
Help us out by becoming a paid supporter: patreon.com/techwontsaveus
Meta Addressing bias in LLMs It's well-known that all leading LLMs have had issues with bias-specifically, they historically have leaned left when it comes to debated political and social topics. This is due to the types of training data available on the internet. Our goal is to remove bias from our Al models and to make sure that Llama can understand and articulate both sides of a contentious issue. As part of this work, we're continuing to make Llama more responsive so that it answers questions, can respond to a variety of different viewpoints without passing judgment, and doesn't favor some views over others. We have made improvements on these efforts with this release—Llama 4 performs significantly better than Llama 3 and is comparable to Grok:
• Llama 4 refuses less on debated political and social topics overall (from 7% in Lama 3.3 to below 2%). • Llama 4 is dramatically more balanced with which prompts it refuses to respond to (the proportion of unequal response refusals is now less than 1% on a set of debated topical questions). • Our testing shows that Llama 4 responds with strong political lean at a rate comparable to Grok (and at half of the rate of Llama 3.3) on a contentious set of political or social topics. While we are making progress, we know we have more work to do and will continue to drive this rate further down. We're proud of this progress to date and remain committed to our goal of eliminating overall bias in our models.
Meta introduced Llama 4 models and added this section near the very bottom of the announcement 😬
“[LLMs] historically have leaned left when it comes to debated political and social topics.”
ai.meta.com/blog/llama-4...
list of banned keywords
🚨BREAKING. From a program officer at the National Science Foundation, a list of keywords that can cause a grant to be pulled. I will be sharing screenshots of these keywords along with a decision tree. Please share widely. This is a crisis for academic freedom & science.
The EU's #AIAct red lines on certain AI applications are now in effect! Learn more: algorithmwatch.org/en/ai-act-pr...
📢 If you believe a prohibited AI system has been used against you, you can report it to your country’s market surveillance authority. In Germany, this is the @bnetza.bsky.social!
„WENN IRGENDJEMAND VON UNS DIE HAND HEBT, UM MIT DER AFD ZUSAMMENZUARBEITEN, DANN STEHT AM NÄCHSTEN TAG EIN PARTEIAUSSCHLUSSVERFAHREN AN.",
Tja.
Cyberlibertarianism has shaped how we see the internet. As tech aligns with the extreme right, it’s more important than ever to reassess it.
This week @hypervisible.bsky.social joins @parismarx.com to discuss David Golumbia’s book on cyberlibertarianism.
Full ep: techwontsave.us/episode/261_...
Der Sturm gegen Merz nimmt krass Fahrt auf! Von Hamburg über Leipzig, Würzburg, Halle, Aachen… - und morgen in Berlin auf der Reichstagswiese MIT Michel Friedmann!
Hamburg hält dagegen!
In #Hamburg haben heute 80.000 Menschen gegen den Tabubruch von Union und FDP protestiert. Denn: Wer mit Faschisten paktiert, hat nichts kapiert! #Merz
openai can steal from anyone, but no one can steal from openai
Screenshot of Google docs generative AI agent asking me to Rephrase, Shorten, Elaborate, More formal.
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. 🧵
