Bernhard Jaeger

Tired Europe: Let's do tons of AI regulations
Wired Europe: Let's do tons of AI open source
#aiPULSE2025

This essay, roughly on dual use, has been haunting me for a while now:
dl.acm.org/doi/pdf/10.1...

Excited to be at #Neurips2025 this week to present our paper "Monoculture or Multiplicity: Which is it?", joint work with Moritz Hardt.

📄 Paper #1000: openreview.net/pdf?id=DO5Lt...
📍 Wed, Dec 3, 2025 • 4:30 PM – 7:30 PM

Feel free to come by and reach out!

A short 🧵.

Attending #Neurips2025? Get your personalized Scholar Inbox conference program now to easily navigate the poster sessions and find what you are looking for:
www.scholar-inbox.com/conference/n...

Scholar Inbox for NeurIPS is live now.

Preprint site arXiv is banning computer-science reviews: here’s why The repository is taking steps to tackle a surge in low quality, AI-generated content.

www.nature.com/articles/d41...

ArXiv banned surveys due to AI slop spam.
Now we need to wait for them to be peer-reviewed.
Bad development, we need to find better solutions to AI slop than banning unreviewed papers.
Getting a survey reviewed at a good journal can take over a year. :(

Quick reminder about the EPFL PhD program deadline (EDIC) on Dec 15.

no this work focusses on IL.
I would personally be interested whether RL models habe similar failures but it is much harder to do this type of analysis when the model predicts actions not waypoints. (Can't do it offline anymore)

🚀 Introducing TMLR Beyond PDF!

🎬 This is a new, HTML-based submission format for TMLR, that supports interactive figures and videos, along with the usual LaTeX and images.

🎉 Thanks to TMLR Editors in Chief: Hugo Larochelle, @gautamkamath.com, Naila Murray, Nihar B. Shah, and Laurent Charlin!

Congratulations to @cworthy.org on their announcement today!

Learn more about this wonderful FRO here: www.youtube.com/watch?v=DA-e...

Come be our colleague in the robotics and embodied intelligence center at NYU!
🔷 Professor in Robotics / Embodied AI (Open Rank)
apply.interfolio.com/176977
🔷 Faculty Fellow in Robotics / Embodied AI
apply.interfolio.com/177077

I think research that hits dead ends is not the useless research people are talking about.

there is great work that hit dead ends:
icbinb.cc

publication has a cost, wastes other peoples time to read, verify.
Also, only works in small science fields.
In ML your reputation outside your core community is your google scholar.
Know several people I would never hire but have compelling profiles. people don't call out sloppy work publicly.

🚀 New paper: ConeGS Error-Guided Densification Using Pixel Cones. We improve 3D Gaussian Splatting by placing Gaussians where they matter most: ConeGS adds primitives along pixel-view cones guided by image error, boosting quality with fewer Gaussians. baranowskibrt.github.io/conegs/

New SOTA in imitation learning (IL) for planning in CARLA, and an in-depth analysis of all the overfitting that is going on with IL policies.

arxiv.org/abs/2511.07292

GitHub - autonomousvision/Bench2Drive-Leaderboard: Bench2Drive Leaderboard Bench2Drive Leaderboard. Contribute to autonomousvision/Bench2Drive-Leaderboard development by creating an account on GitHub.

Bench2Drive has become a popular autonomous driving benchmark with over 40 methods. Papers only report a few baselines due to space constraints, making it hard to keep an overview. To fix this, we made a leaderboard that contains every published result:

github.com/autonomousvi...

📣 JobAlert:
12 days left, hit the Apply button now:
Join Tübingen AI Center as a Scientific Project Coordinator for OpenEuroLLM - support, building a series of open-source foundation models for transparent AI in Europe!
👉 lnkd.in/eTptRaFy
📅 Deadline: November 19, 2025
@unituebingen.bsky.social

Personally I would not want to be in the camp of arguing that we simply *must* have 40k vehicular deaths a year. Status quo bias is an insane drug

Here is a formal impossibility result for XAI: Informative Post-Hoc Explanations Only Exist for Simple Functions. I'll give an online presentation about this work next tuesday in @timvanerven.nl 's Theory of Interpretable AI Seminar:

arxiv.org/abs/2508.11441

tverven.github.io/tiai-seminar/

Launching the Physical AI AV Dataset! 🚀

huggingface.co/datasets/nvi...

One of the largest, most diverse & commercially usable open-source datasets for AVs.
- 1727 hours of driving data
- Camera, LiDAR, & radar
- 25 countries, 2500+ cities

This is just the beginning, more features to come!

Qualitative difference in data quality. OpenDV are youtube dash cams. This is data from real self-driving vehicle sensor stacks.
This should be around 10x the size of prior biggest dataset (the nuPlan split that has sensor data).

Strange, that is a bug in the bluesky preview. The blue link I posted works perfectly fine.

Big day for autonomous driving research.
Nvidia just dropped 1700 hours of public driving data on HuggingFace from over 2500 cities:

huggingface.co/datasets/nvi...

interesting, didn't know that. is there any documentation / paper about how they are using rt-cores for physics acceleration?

not sure if still up to date, but there used to be an exception for universities from the no geforce in clusters rule. we have an old 2080ti cluster.
(wouldn't recommend though they break often. rtx should be the better choice nowadays).

uses raytracing for rendering which requires raytracing cores which these gpus don't have.

Still bad engineering should have the option to run without rendering on A100.

Some chinese folks I talked to mentioned that the local governments are in a fierce competition about who can invest the most money in AI.

Takeru Miyato - AI Research Scientist PhD student at University of Tübingen working on artificial intelligence, machine learning, and deep learning research.

Congratulations to our PhD student @takerumiyato.bsky.social for winning the Google PhD Fellowship in the category "Machine Learning and ML Foundations". Takeru is pioneering new neural architectures that improve generalization and efficiency. Check out his research: takerum.github.io

h-index is definetly broken but this is making it worse not better.

I now understand the position of the YOLO author a lot better than I did before