This is an interesting thought: Meta's wearables seem geared toward consumers. Recording all your actions and sending the data to a robot lab to train their robots could raise privacy concerns, since data from many people would be captured. If you use them at home, labs might prefer VR instead, as it already includes data from handhelds and controllers.
Under the hood, how comparable are “physical AI” systems to modern multi-modal LLMs? While I appreciate the comparison as a useful side-by-side for the sake of this discussion on the scale of training data, I’m genuinely curious how different the model architectures are when processing and reasoning through spatial and sensory data compared to tokenized text, audio, or video input. Does the comparison validly extend that far or does the framework for physical AI stand completely apart from the scaffolding that has led to chatbots and code/text-based agentic workflows?
How likely is it that Meta makes a comeback in the AI race with wearables? That seems like the best way to capture this kind of data.
This is an interesting thought: Meta's wearables seem geared toward consumers. Recording all your actions and sending the data to a robot lab to train their robots could raise privacy concerns, since data from many people would be captured. If you use them at home, labs might prefer VR instead, as it already includes data from handhelds and controllers.
Under the hood, how comparable are “physical AI” systems to modern multi-modal LLMs? While I appreciate the comparison as a useful side-by-side for the sake of this discussion on the scale of training data, I’m genuinely curious how different the model architectures are when processing and reasoning through spatial and sensory data compared to tokenized text, audio, or video input. Does the comparison validly extend that far or does the framework for physical AI stand completely apart from the scaffolding that has led to chatbots and code/text-based agentic workflows?