Stephan

An analysis of ~10,000 PhD students finds that: Having an advisor with a strong publication record + past students who have been successful predicts success, along with having a close student-advisor relationship and publications prior to starting one's PhD: nber.org/system/files/w…

Agreed!

An amazing achievement!

Your LLM isn't doing math - it's using clever pattern matching tricks. LLMs perform arithmetic using neither robust algorithms nor memorization; rather, they rely on a “bag of heuristics”, as proposed in this paper. 🤔 Original Problem: Do LLMs solve reasoning tasks using actual algorithms or just memorize training data? This fundamental question impacts how these models truly learn and generalize. ----- 🛠️ Solution in this Paper: Using arithmetic as a test case, researchers analyzed individual neurons in LLMs to understand their reasoning mechanism. They identified a circuit of ~200 neurons per layer (1.5%) that handles arithmetic operations. Each neuron implements simple pattern-matching rules called "heuristics" - like recognizing when operands fall within certain ranges or have specific patterns. ----- 💡 Key Insights: → LLMs don't use true mathematical algorithms or pure memorization → They combine multiple simple pattern-matching rules (heuristics) to solve arithmetic → Different neurons activate for different numerical patterns and boost corresponding answer tokens → This explains both their capabilities and limitations in arithmetic reasoning → The mechanism appears early in training and remains consistent throughout ----- 📊 Results: → The identified circuit achieves 96% faithfulness in reproducing full model behavior → Only 200 neurons per layer (1.5%) are needed for accurate arithmetic → 91% of top neurons implement identifiable heuristic patterns → Each heuristic causes 29% accuracy drop when ablated

📢Interested in the (#legal, #economic, and #technical) #dynamics behind #GenerativeAI?? 🚀🚀 💡The recordings of our #Symposium on the “Dynamics of Generative #AI” (by @JWI_Berlin & @alti_VU) are now available! 🎥Watch it now & spread the word👉 youtube.com/playlist?list=…

Bis unsere Forderung nach dem grundständigen Pflichtfach Informatik deutschlandweit etabliert sein wird, ist es noch ein weiter Weg. Über den Stand der Dinge informiert jährlich der Informatik-Monitor, der lesenswert ist: kurzlinks.de/infomonitor @informatikradar @stifterverband

already exists

Seeing increasing numbers of work from LLM research community that gets really close in nature to process mining. Especially in LLM subfield called “agentic workflows”, folks try to improve an AI’s ability to solve tasks by executing a workflow of several prompts.

+++Job in Berlin+++ 📢 For our research group "Well-being in the Digital World" at the Weizenbaum Institute we are looking for a Research Associate. 📆 starting from 15.12.2024 with 75% of the regular working hours (29.25 hours per week). ✅ Apply now 👉 weizenbaum-institut.de/en/news/detail…

On the one hand a very big success for Weizenbaum. On the other hand, we loose some great minds!

Vier Professuren für WI-Forschende: Wir gratulieren Bianca Herlo, Lena Ulbricht, Annett Heft und Florian Butollo zu ihren neuen Aufgaben! ➡️ weizenbaum-institut.de/news/detail/vi… @bherlo @akheft @lena_ulbricht @flobut

Very nice work! I really like the idea of using genetic algorithms! Big step towards fairness-aware process mining!

@profBPM I understand the decision, but it also has quite a few negative consequences. Just as an example, the UK and Hong Kong give out talent visas to graduates from top-100 universities. Graduates from universities that do not participate in rankings are at a disadvantage here.

Proud of Utrecht University's decision to withdraw from the Times Higher Education ranking:

@finnkless zeigt in diesem Interview auf, wie wir die EU-Taxonomie zu Nachhaltigkeit für Geschäftsprozesse operationalisieren könnten. Fantastische praxisrelevante Forschung! Danke an @luise_pufahl und @janmendling für die Zusammenarbeit!

Unternehmen sollen im Sinne der #EU-#Taxonomie-Verordnung nachhaltig wirtschaften. Was besagt diese und wie können Unternehmen überprüfen, ob ihnen das gelungen ist? WI-Wissenschaftler:innen haben eine Methode entwickelt, die helfen könnte 👉weizenbaum-institut.de/news/detail/na… @finnkless

🎉🎊 Die #HumboldtUniversität erreicht im aktuellen @timeshighered-#Ranking den Platz 84 von 2.092 bewerteten Universitäten weltweit! Deutschlandweit belegt sie den vierten Platz. Mehr Infos 👉 hu-berlin.de/de/pr/nachrich…

#ComputerScience is coming of age.

We are still welcoming applications for our Postdoctoral Research position in Political Science! 📩

@lena_ulbricht @info_uni_hi @WZB_Berlin @JWI_Berlin @techdo_berlin Congratulations!

An opportunity to follow the new minds of #BPM