“Data will solve robotics and automation: True or false?”: A debate. Held @ieee_ras_icra #ICRA2025 and featuring BAIR faculty @Ken_Goldberg and BAIR alumnus @animesh_garg
.
RESEARCH
-
Data’s role in solving robotics and automation challenges
By
–
-
Engineering closes 100,000-year data gap in robotics
By
–
New @SciRobotics editorial by BAIR Faculty @Ken_Goldberg
: Good old-fashioned engineering can close the 100,000-year “data gap” in robotics -
LLMs Beyond Matrix Multiplication: Understanding Model Capabilities
By
–
Yes. yes LLMs are not just matrix multiplication but adding that there are non-linear functions as well doesn't really do anything to resolve the central mystery of why these models can do what they do. And here is the source of Wolfram's paragraph:
-
AI Vision Models: Weaknesses in Counting and Image Generation
By
–
Clear weak spots remain counting, generating alternate images when the training data is thick (full glasses of wine, clocks with oddly set hands), etc. It isn't hard to make them fail. But there is a lot they do very well, and the gains have been pretty quick so far.
-

Why LLMs Should Weird You Out: Understanding Their Capabilities
By
–
I don't know anyone who uses LLMs who is not occasionally weirded out by what they can do. If you are not, you should be. They are weird. Wolfram had a rather startling (at least at the time) theory after using ChatGPT. Understanding whether he is right is important.
-
The Deep Mystery: How LLMs Simulate Human Thought
By
–
We really have not made a lot of progress on explaining the deep mystery of LLMs: How does a model using matrix multiplication to predict the next word manage to simulate human thought well enough to do all the very human-like things it does? And what does that mean about us?
-
How Do LLMs Simulate Human Thought Despite Small File Size
By
–
I think it is actually makes LLMs even weirder! The next question is "how does a file the size of a moderately sized video game simulate human thought" and I don't think we have good answers.
-
MiniCPM-V 4.5 Chat App Development with Impressive Benchmark Performance
By
–
vibe coding a MiniCPM-V 4.5 @OpenBMB chat app in anycoder
— AK (@_akhaliq) 28 août 2025
MiniCPM-V 4.5 achieves an average score of 77.0 on OpenCompass, a comprehensive evaluation of 8 popular benchmarks. With only 8B parameters, it surpasses widely used proprietary models like GPT-4o-latest, Gemini-2.0 Pro,… pic.twitter.com/r1i5b6JfFpvibe coding a MiniCPM-V 4.5 @OpenBMB chat app in anycoder MiniCPM-V 4.5 achieves an average score of 77.0 on OpenCompass, a comprehensive evaluation of 8 popular benchmarks. With only 8B parameters, it surpasses widely used proprietary models like GPT-4o-latest, Gemini-2.0 Pro,
-
AI System Achieves Perfect Performance in Single Attempt
By
–
I agree. I was super impressed with how it got everything right in just one shot. Haven't seen anything like that yet.
-

Self-Rewarding Vision-Language Model Through Reasoning Decomposition
By
–
Self-Rewarding Vision-Language Model via Reasoning Decomposition