Germans Savcisens (Savčišens)
Postdoc @nunetsi.bsky.social (Northeastern Uni) 🎓 Computational Social Science 👾 ✨ work on stability of belief in LLMs & Human-AI Collaboration 🌿 he/him 🇱🇻 🇺🇦 | www.savcisens.com
- Reposted by Germans Savcisens (Savčišens)A feature in Nature reports how research shows that ‘exercise snacks’ and other forms of everyday movement can greatly reduce the risk of heart disease and death. #medsky 🧪
- It’s been two years since we published the #life2vec paper, and it’s still circulating widely. People keep discovering it but much of what circulates online is misleading. Agter a long time, I finally wrote a short explainer to clear up a few things: converges.medium.com/is-life2vec-...
- Attending @neuripsconf.bsky.social this week! If you want to chat about LLMs for behaviour / health / labour modeling... or about beliefs and opinions of LLMs, hit me up. I’ll also be presenting a poster on truth tracking at the Mechanistic Interpretability workshop later. Come say hi!
- My 2 cents: If you exploited the #openreview bug or are actively searching for the leaked data, you should seriously reconsider your place in research. If you cannot uphold the basic principle of double-blind review, how can we trust anything you publish?
- Truthfulness isn’t always binary. Sometimes it’s… neither 🤔 Our Trilemma of Truth paper is headed to the @neuripsconf.bsky.social Mechanistic Interpretability workshop 🚀 Let’s connect in San Diego! 🌴 Preprint: arxiv.org/abs/2506.23921 Code and data: github.com/carlomarxdk/...
- Had the pleasure of presenting our work on Three-valued veracity probes for LLMs at #NEMI Workshop! Mechanistic Interpretability has such a great and welcoming community. If we crossed paths - let’s connect! 🚀 Poster: zenodo.org/records/1690... Preprint: arxiv.org/abs/2506.23921
- @nunetsi.bsky.social had a great week at @ic2s2.bsky.social! Looking forward to the next #IC2S2 in Vermont 🔬⛰️
- Reposted by Germans Savcisens (Savčišens)
- All the keynote recordings are available now, enjoy! www.youtube.com/playlist?lis...
- Colleagues are making sure that I stay focused at #IC2S2 🤓
- Presented our work on veracity-tracking in LLMs at #IC2S2 today! Now looking forward to the next few days of great talks and conversations ✨️🎓
- Perfect weather, charming streets, and a poster so big it almost needed its own boarding pass 🧳✨ Excited to attend #IC2S2 in Norrköping 🇸🇪 Find me at the Poster Session on Tuesday: "Improving Probes that Track Veracity in Large Language Models" (Poster ID: 39) 🧪
- Little wins: our "Trilemma of Truth" dataset just hit 150 downloads. It contains true, false, and neither-valued statements (inspired by the three-valued logic) used to stress-test LLMs for fact-checking, veracity tracking, and uncertainty handling. Dataset📚: huggingface.co/datasets/car...
- Perfect weather, charming streets, and a poster so big it almost needed its own boarding pass 🧳✨ Excited to attend #IC2S2 in Norrköping 🇸🇪 Find me at the Poster Session on Tuesday: "Improving Probes that Track Veracity in Large Language Models" (Poster ID: 39) 🧪
- I’m presenting a poster on my latest project: “The Trilemma of Truth.” Drop by to see how LLMs leverage three‑valued logic to model truth 🔢🤖 And hey, if you fancy grabbing a coffee ☕, DM me! 📄 Poster: zenodo.org/records/1605... 📖 Preprint: arxiv.org/abs/2506.23921
- #IC2S2 2025 is just around the corner! July 21–24 in Norrköping, Sweden. Bookmark tutorials, keynotes, and must-see sessions and connect with fellow attendees using #IC2S2. Full program here: www.ic2s2-2025.org/program/ See you in Norrköping! 🇸🇪
- 🚨 New preprint! Do LLMs really know what’s true? In our paper, @eliassi.bsky.social and I introduce sAwMIL: a probing method that distinguishes between true, false, and neither—capturing what LLMs actually “retain.” We evaluated 16 open models across 3 new datasets. 📄 arxiv.org/abs/2506.23921
- Dataset with statements related to City Locations, Medical Indication, and Word Definitions is available on 🤗 huggingface.co/datasets/car...
- Reposted by Germans Savcisens (Savčišens)Germans Savcisens, Tina Eliassi-Rad: The Trilemma of Truth in Large Language Models arxiv.org/abs/2506.23921 arxiv.org/pdf/2506.23921 arxiv.org/html/2506.23921
- That happens way too often to me 🥲
- Everyday struggle 🧪 plentyofroom.beehiiv.com
- What's the coolest guide/source on "Complex Data Visualization"? I am looking for some inspiration to visualize graphs and high dimensional data.
- Reposted by Germans Savcisens (Savčišens)Great read from 2019 about abandoning the use of p-values in a dichotomous way and what we can do instead. More thinking, and less relying on significance tests to decide things for us! www.nature.com/articles/d41...
- Reposted by Germans Savcisens (Savčišens)Sometimes we need a reality check 😉 @serge.belongie.com
- Amazing time with the folks from @tint-philosophy.bsky.social at the retreat on Predictability of Human Lives. Great people & discussions, and so much to reflect on—especially around integrative modeling and how neural networks can help us get there. Plus, a relaxing sauna to top it off!
- Visiting @mpidr.bsky.social this week—super excited to see what’s happening in Demographic Studies (don’t miss my talk!). Also, I’ll be in Berlin on Feb 2, Helsinki from Feb 3-6, and Copenhagen from Feb 10-12. Let me know if you’re around and up for a coffee 🧪🔬☕️
- Reposted by Germans Savcisens (Savčišens)For those interdisciplinary students/scholars who are having identity crisis, this is for you (from 2018): "How to survive as an interdisciplinary being" www.slideshare.net/slideshow/ho... #NetSciX2025
- Reposted by Germans Savcisens (Savčišens)New tool to estimate the level of participation in collective action expressed in natural language. Applied to social media, it can produce large-scale and granular estimates of behavior change wrt collective action. github.com/ariannap13/e... @nerdsitu.bsky.social @itu.dk @carlsbergfondet.dk
- I once asked ChatGPT how it thinks my life would look like in 20 years. And "Visionary Multidimensional Social Scientist" sounds like a great job title 😅 I guess it captured my love for the "His Dark Materials" trilogy.
- Reposted by Germans Savcisens (Savčišens)📢 @savcisens.com discusses a recent study that shows that LLMs exhibit social identity biases similar to humans. www.nature.com/articles/s43... 🔓https://rdcu.be/d5owe
- Happy to write this News & Views piece on the recent audit showing LLMs picking up "us versus them" biases: www.nature.com/articles/s43... (Read-only version: rdcu.be/d5ovo) Check out the amazing (original) paper here: www.nature.com/articles/s43...
- Reposted by Germans Savcisens (Savčišens)I don't like the way many CS papers are written, even the supposedly good ones, but these tips are very generically applicable and useful. Just ignore the bit about the acronyms...
- If you want to follow what happens with #ML, #AI, #LLM research in Denmark - here is a great Starter Pack 😋 #researchers
- #YouTube has so far come up with the best application of GenAI (in recent months): summarization of videos and the ability to chat with them -- a great way to check the contents of 60+ min talks.
- Had the pleasure of presenting how transformers can help us decode/explore complex life patterns in labour and health trajectories. If you're curious about the intersection of machine learning, social science, and health, check it out here: youtu.be/szSY7MSkLtk?...
- Reposted by Germans Savcisens (Savčišens)Network Science Institute starter pack in BlueSky! A list of members and alumni (to be updated) go.bsky.app/N3PQTdwat://did:plc:evquxlulwq744n6xh5daezyd/app.bsky.graph.starterpack/3lan2m3vgnh2j
- Reposted by Germans Savcisens (Savčišens)We're hiring a postdoc in the &-Lab at Northeastern's Network Science Institute! Looking for a curious, collaborative scholar to work on computational social science questions, at the intersection of data justice + network science. northeastern.wd1.myworkdayjobs.com/careers/job/...
- Join the talk to hear more about transformers for social science: www.soc.cuhk.edu.hk/event/nov-14...
- A simple implementation of the #life2vec model is now available at github.com/carlomarxdk/.... You can try out the workflow with dummy data, + we are planning to add more examples (including finetuning) in the near future.
