Dorothea Kolossa

Graphic combining logos and a portrait. At the top left is the Science Media Center Germany logo in a geometric gold network style. Below appears the Ubiquitous Knowledge Processing logo. On the right is a circular portrait of a woman with shoulder-length curly hair and glasses, wearing a patterned blouse and looking toward the camera. A small photo credit is visible in the corner.

⚠️ 𝗖𝗮𝗻 𝗳𝗶𝗻𝗲-𝘁𝘂𝗻𝗶𝗻𝗴 𝗹𝗲𝗮𝗱 𝘁𝗼 𝘂𝗻𝘄𝗮𝗻𝘁𝗲𝗱 𝗯𝗲𝗵𝗮𝘃𝗶𝗼𝘂𝗿𝘀 𝗶𝗻 𝗹𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗺𝗼𝗱𝗲𝗹𝘀?
A recent paper in Nature suggests that even small amounts of targeted fine-tuning data can trigger unexpected and problematic behaviour that generalises beyond the original pre-training task.

(1/🧵)

Delta-band cortical speech tracking predicts audiovisual speech-in-noise benefit from natural and simplified visual cues Humans comprehend speech in noisy environments more effectively when they can see the talker’s facial movements. While the benefits of audiovisual (AV…

The title of our new paper says it all: "Delta-band cortical speech tracking predicts audiovisual speech-in-noise benefit from natural and simplified visual cues"
Very happy about our collaboration w Tobias Reichenbach, Enrico Varano, Mike Thornton-have a look: www.sciencedirect.com/science/arti...

(repost welcome) The Generative Model Alignment team at IBM Research is looking for next summer interns! Two candidates for two topics

🍰Reinforcement Learning environments for LLMs

🐎Speculative and non-auto regressive generation for LLMs

interested/curious? DM or email ramon.astudillo@ibm.com

YouTube video by Knut Jägersberg Conscious AI and Biological Naturalism

Conscious AI and Biological Naturalism

youtu.be/O34BDDfXEUM

At @tuberlin.bsky.social we are running a coding study, to see the impact of a coder's background on their interactions with LLMs - and on their ultimate success in coding. Please join: umfragen.tu-berlin.de/llm_coding_s... 👩‍💻🧑‍💻#diversity (yes!)

Hm, yes, the world is changing super-fast. Somehow, I feel like we also need to get back to some of our idealistic roots, and try to make the world a better place. I hope we can manage to agree again, what a "good place" is!

The Rise of Silicon Valley’s Techno-Religion

So what do you think about the rationalist movement? Here's a weird and interesting article - let's discuss :)
www.nytimes.com/2025/08/04/t...

Student projects @ mtec - YouTube These videos show the final presentations of student project labs done at mtec. They were prepared and presented in either the "Projekt Elektronik" or the "P...

Elektrotechnik und Informatik können und sollen Spaß machen - das sieht man auch in in unseren Projektlaboren. Unsere neuen Videos, zum Sommersemester '25 - sind jetzt auf YouTube: www.youtube.com/playlist?lis...

📢 Upcoming Abstract Submission Deadline: May 23. ⏰
2025 ITG Conference on Speech Communication, Berlin, Germany, September 24-26.

The call for papers with all info can be found here:
www.tu.berlin/qu/itg-sc-2025

Keynotes will be given by Ivan Tashev, Sarah Verhulst and Emmanuel Vincent🎉

(what I like about those (in addition to them being flight simulators ;) is the physics model, flight dynamics and all. I think that is one effortful but effective way of bringing in some solid ground-truth :)

for sure it does! Go, flight simulators!

The Quest for A.I. ‘Scientific Superintelligence’ (Gift Article) An ambitious start-up embodies new optimism that artificial intelligence can turbocharge scientific discovery.

New York Times article on #AI4ScientificDiscovery: www.nytimes.com/2025/03/10/t...

and yet.... calling it "twitter" still is in direct opposition to those, who'd rather have propaganda than human exchange. To those who dominate the conversation, and who get to pick the name, but not the history.

Super-proud of Wentao Yu, who very successfully defended his thesis on "Multimodal Integration in Machine Learning" at TU Berlin. The thesis shows that multimodal integration is important for many tasks, from better speech recognition to detection of hate speech and misogyny.
@tuberlin.bsky.social

Dawn Song and her team @ucberkeleyofficial.bsky.social are again so quick 🙏
d2jud02ci9yv69.cloudfront.net/2025-04-28-l...

Mountain scenery with road, small houses and the sun in the sky.

Spending a few days with my best friend and her son in Austria. Out hiking today 😀

Soziale Netzwerke als demokratische Kraft retten Rund 100 Akteur*innen aus Kultur, Wirtschaft und Medien haben sich zur Initiative Save Social zusammengetan. [1] Gemeinsam schlagen sie zehn konkrete Schritte vor, um das Internet von der Dominanz ...

Save social media - and democracy - from undue influence of intransparent algorithms that push hostility and disinformation to maximize reach: weact.campact.de/petitions/sa...

CDC removes gender, equity references in public health material Large sets of data are being scrubbed of references to transgender and LGBTQ+ people, among others, which could compromise their use in research.

What's happening to academic integrity and freedom of expression in the States?
Am I falling for disinformation here? #cantbereal
www.washingtonpost.com/health/2025/...

Podcast: Fakten statt Fiktion - Gemeinsam gegen Desinformation Verschiedene Formen von Desinformationen wie Fake News, manipulierte Videos und Deepfakes stellen unsere Demokratie vor neue Herausforderungen. Jörg Thadeusz begrüßt seine Gäste zur Podiumsdiskussion.

🎧 Hörtipp: Fakten statt Fiktion!
Wie beeinflusst #Desinformation unsere #Demokratie? Dr. Vera Schmitt von der #TUBerlin diskutiert bei @rbb24.de u.a. mit @alexandrageese.bsky.social über Deepfake-Erkennung & #KI- Regulierung. Jetzt reinhören ⬇️
www.ardaudiothek.de/episode/foru...
@bifold.berlin

Together with the idea that we should work towards 'uploading' our consciousness

I agree, not the proposition itself is dangerous. As long as it's just a hyothesis, and we're willing to put it to the test scientifically, that's safe & important :)
But if decisions are made, based on assuming it to be true, it's very worrisome to me. What do you think about longtermism?

And so what is your take on this? I see this as pure speculation, without a proposed way of generating evidence.
Dangerous speculation, imho, as it feeds the narrative that we'll be able to migrate to computers - so this plan may take precedence over saving life as we know it. But that's just me :)

YouTube video by Yuval Noah Harari Jon Kabat-Zinn & Yuval Noah Harari In Conversation

@yuvalharari.bsky.social and Jon Kabat-Zinn on the future of #consciousness and #AI (that's my take at least): www.youtube.com/watch?v=7r5l...

Sieben Wissenschaftler zeigen lächelnd auf den Kleinsatelliten InnoCube und den dazugehörigen Auswurfcontainer. Foto: Exolaunch

🛰️ Heute Abend hebt der #Satellit InnoCube der #TUBerlin ab und schreibt #Raumfahrt- Geschichte: Er ist der 1. Satellit mit kabellosem Datenverkehr.

Den Start im Livestream mitverfolgen:
📺 www.youtube.com/live/2m-KFet...
📅 14.01.2025
⏰ 17:30 Uhr (Beginn der Übertragung)

Call for applicants - PhD students (12 positions)

“Privacy for Smart Speech Technology” (PSST) is a joint doctoral training programme and Horizon Europe Marie Skłodowska-Curie Action.

See psst-doctoralnetwork.eu/news/call-fo...

#doctoralnetwork #openposition #phd #PSST #mariecurie

YouTube video by Lex Fridman Dario Amodei: Anthropic CEO on Claude, AGI & the Future of AI & Humanity | Lex Fridman Podcast #452

My holiday content suggestion. Such a fascinating discussion on what #LLMs should be like. Plus (sorry about the clickbait):
Hear @lexfridman.bsky.social on something 'good for the soul' (I fully agree, not that that matters ;-)
www.youtube.com/watch?v=ugvH...

Very interesting takeaways from consciousness theories for AI system design:

Wie hilft KI beim Faktenchecken? Event am Dienstag, 10.12., an der @TUBerlin: www.tu.berlin/nachrichtend...

so happy to be here, migrated from twitter, among what feels like a crowd of mostly sane people and regular nerds like me🥳