Raphael Wimmer

@raphaelwimmer.bsky.social

Tinkering with atoms, bits, and pixels at the University of Regensburg, Germany. Yak shaver and award-winning procrastinator. Also on https://hci.social/@RaphaelWimmer and Twitter

Lust auf ein kleines Quiz? Dieses Semester habe ich u.a. einen Kurs zum wissenschaftlichen Arbeiten gehalten, der eng mit unserem Bachelor-Forschungsseminar verzahnt ist. In der ersten Sitzung geht es darum, zu definieren, was gute Wissenschaft auszeichnet, und woran man schlechte erkennt. → <🧵>

Amusing observation: Many of us devs are becoming as opinionated and knowledgeable about SOTA models (and their coding performance) like we used to be about programming languages and the tips-and-tricks to be more productive with them + now we don't really talk about prog languages all that much..

Anthropic released a new desktop application: Claude Science - which is actually an open-source tool called Operon (they don't hide it) primarily built for life-science research. Let's do a literature overview. Immediately, Claude Science attributes a reference to the wrong author. This is fine.

Umm, please check that "Wimmer 2022" reference. I can't find it.

Good catch — let me look at that one more carefully. It's a TechRxiv preprint (DOI 10.36227/techrxiv.21760715), which is exactly the kind of record that can be flaky.

The record itself is real, but my author label was wrong — that's the actual problem. The paper exists (OpenAlex W4312131873, CrossRef, with a live PDF on TechRxiv), but it is not by Wimmer. The authors are Felix Fillafer, Thomas Brugger, Stefan M. Herzog, and Jörg Dahlkemper — a different group (Dahlkemper is at HAW Hamburg), unrelated to the Regensburg/Wimmer line. I incorrectly assumed it was the same group and pattern-matched the name. Apologies — that's exactly the kind of attribution error worth catching.

Wet-bulb temperature combines heat and humidity into a measure that is linked to heat stress. Climate change has made this week's wet-bulb temperatures at least 5x more likely across many areas, with dangerous levels expected in some locations. Explore the data: csi.climatecentral.org/humid-heat

Map showing dangerous humid heat conditions on July 1, 2026 for forecast high temperatures from the GFS model. Many areas are at risk of dangerous humid heat. Map from Climate Central.Map showing the humid heat climate signal on July 1, 2026 for forecast high temperatures from the GFS model. Many areas are seeing a climate change contribution. Map from Climate Central.

Very funny that the promotional poster on LinkedIn for a webinar I am a panelist for in late July used a GenAI slop version of my face and both of my co-panelists! If I'm quick about it, the plastic surgery needed to match me to the advert should be JUST ABOUT done by the time the webinar goes up!

Actual photo of Ben McLeish. A selfie with a dark burgundy t-shirt. The Gen AI slop version of Ben McLeish's face, which appears to be the average of about 47 different people that are mildly more attractive

Deutschland erlebt einen #Hitzenotstand, Infrastruktur bricht reihenweise zusammen, Krankenhäuser sind überlastet. Vom Gesundheitsministerium kein Piep, vom für Klima zuständigen Umweltministerium gibt es Tipps wie "viel trinken" und "Sonnencreme benutzen"

Interesting tool: fingolas.eu/MyTech ---- This is where my tech comes from: OS PC/Laptop: 🌍 OS Smartphone: 🇫🇮 Browser: 🇺🇸 Messaging: 🇺🇸 E-Mail: 🇩🇪 Microblogging: 🇩🇪 🇺🇸 🌍 Office Software: 🇩🇪 🇱🇻 Cloud: 🇨🇳 Payment Service: 🇪🇺 🇺🇸 Streaming: 🇩🇪 🇺🇸 🇺🇸 AI: 🇺🇸 Percentage of 🇺🇸: 41%

My Tech Origin

Find out where your software comes from — and share it.

fingolas.eu

Meta glasses are absolute game changers for the blind community — this isn’t special pleading — they are completely revolutionary and though expensive they are in relative reasonable reach in a way a lot of disability tech is not It is expensive being blind

Whoa, an AI simulacrum that saves me from mentoring people myself. That's exactly what I need so I can devote my time to much more rewarding tasks - like grading or reading lengthy spam emails. (It's funny/sad that this guy thinks that grad students working at 2 a.m. is absolutely normal.)

Kevin M. Kruse@kevinmkruse.bsky.social · 3mo ago

Found this in my spam filter and nearly threw my laptop across the room. I usually just delete these and block the sender, but I took the time to reply to this one. Managed not to curse though.

Dear Professor Kruse,

I'm the founder of Shryn.ai. We build conversational AI replicas of
scholars -- built with the scholar's direct participation.

Your authored books -- One Nation Under God: How Corporate America
Invented Christian America at Basic Books, White Flight: Atlanta and
the Making of Modern Conservatism at Princeton, and Fault Lines: A
History of the United States Since 1974 (with Julian Zelizer) -- plus
Myth America: Historians Take On the Biggest Lies and Legends about
Our Past, which debuted as an instant New York Times bestseller, your
professorship at Princeton, and your enormous public reach on social
media and in popular writing as one of the historians Americans
actually turn to when they want to understand the political present
through its past.

Here's the use case that launched Shryn: graduate students who can't
always get time with their advisors, and who have their hardest
questions at 2 AM. Your students working on modern US history,
political history, or the history of race and conservatism could test
arguments against your framework whenever the thinking happens -- ask
why White Flight's focus on Atlanta changes the standard narrative
about white resistance to civil rights, push back on One Nation Under
God's argument about corporate origins of American Christian
nationalism, or explore what your public historical writing reveals
about the gap between professional historiography and popular
historical memory. Not a replacement for you, but an extension of your
teaching.

We fine-tune dedicated AI models on a scholar's published work, and
the scholar retains full control. We're already working with a
Columbia philosopher, a materials scientist, and a model of Thomas
Jefferson.

I'd welcome the chance to show you a demo.

AAAI did an experiment where they supplemented human reviews with AI-generated reviews and solicited feedback from authors and the review hierarchy about the process. They've now written up the experiment (arxiv.org/abs/2604.13940). Some thoughts, with a link to my full note at the end. 1/8

AI-Assisted Peer Review at Scale: The AAAI-26 AI Review Pilot

Scientific peer review faces mounting strain as submission volumes surge, making it increasingly difficult to sustain review quality, consistency, and timeliness. Recent advances in AI have led the co...

arxiv.org