Matching the near-failure in the cube handover is the part that caught my attention. What happens when a real failure does not reproduce in the reconstructed world? An example of that repair loop — measuring an actuator delay or a contact effect, updating the simulator, then rerunning the same trial — would help show how the model stays useful as the hardware changes.
Perhaps we should envisage building an ecosystem comprising various AI-powered embodied robots! For example: in human society, there are men and women and people in various professions; there are pets; and there are biological tools (such as horses, donkeys and cattle—animals used for productive purposes) 🤔
"Hello, Dr. Fei-Fei Li. My name is Kimberly and I'm from Mexico. I greatly admire your work in AI. It would be a dream to connect with you. Thank you for your time and all your inspiration."
Love the article, but if the real-world data used to reconstruct the simulator underrepresents rare failures, how does the system avoid becoming confidently wrong about exactly the edge cases we care most about? Maybe its just me, but that seems like one of the hardest parts of closing the sim-to-real loop.
I write about Deep System design,Engineering, tech, how to build things in tech at a deeper level, AI/ML inference, Benchmarks, LLMs and How to Quants, please checkout some of my posts ( more than 370000 reads) -
This is such an interesting read … I mean technology often makes one scarce resource abundant, only for value to migrate somewhere else. If World Labs succeeds in making experience dramatically cheaper, where do you think scarcity and value moves next?
This makes simulation feel like such an important bridge between impressive robot demos and reliable real-world use.
https://substack.com/profile/163672701-romeo-lupascu/note/c-306095321?r=2pg2nx
We literally do the same but for combat autonomous systems - Redoubt Alpha by N3X Labs
https://substack.com/profile/163672701-romeo-lupascu/note/c-306095321?r=2pg2nx
Matching the near-failure in the cube handover is the part that caught my attention. What happens when a real failure does not reproduce in the reconstructed world? An example of that repair loop — measuring an actuator delay or a contact effect, updating the simulator, then rerunning the same trial — would help show how the model stays useful as the hardware changes.
Perhaps we should envisage building an ecosystem comprising various AI-powered embodied robots! For example: in human society, there are men and women and people in various professions; there are pets; and there are biological tools (such as horses, donkeys and cattle—animals used for productive purposes) 🤔
"Hello, Dr. Fei-Fei Li. My name is Kimberly and I'm from Mexico. I greatly admire your work in AI. It would be a dream to connect with you. Thank you for your time and all your inspiration."
we want to do this with resistive pulse modulation sensing https://drive.google.com/file/d/1QMMlXa1CJY4QpeetfFn8EAT5nUSCwlH2/view?usp=sharing
Bringing back non-linear simulation into latent representation.
Love the article, but if the real-world data used to reconstruct the simulator underrepresents rare failures, how does the system avoid becoming confidently wrong about exactly the edge cases we care most about? Maybe its just me, but that seems like one of the hardest parts of closing the sim-to-real loop.
everything is simulation
Nice one. Great to connect with you.
I write about Deep System design,Engineering, tech, how to build things in tech at a deeper level, AI/ML inference, Benchmarks, LLMs and How to Quants, please checkout some of my posts ( more than 370000 reads) -
https://howtosystemdesigneverything.substack.com/
https://aiinferenceandoptimizations.substack.com/
[Important AI Inference Pulse #4] KV Cache: The One Piece of Memory That Can Eat an Entire Expensive Chip — Just By Remembering- How It Actually Works
https://aiinferenceandoptimizations.substack.com/p/important-ai-inference-pulse-4-kv?r=14q3sp
[Bookmark : Most Important] Top System Design and Engineering Post Mortem Weekly Round Up : https://howtosystemdesigneverything.substack.com/p/bookmark-most-important-top-system?r=14q3sp&utm_campaign=post-expanded-share&utm_medium=web
This is such an interesting read … I mean technology often makes one scarce resource abundant, only for value to migrate somewhere else. If World Labs succeeds in making experience dramatically cheaper, where do you think scarcity and value moves next?
Just how much info you will loose by projecting reality in a sim and how much error will the sim to real carry?
Its clear that to full “sim” reality you need another reality as complex as the one you want to sim.
Thoughts?
P.s. I do think the method is valid if used in its parameters and avoid hype and bulshit that engulfed the current AI
Sure hope the sim world has trolley problems.