Silicon Valley is racing to develop huge robotic brains with GPT-style AI. But a stunning new paper says this entire strategy is fundamentally flawed for developing really intelligent machines.
Allow’s be real. The hype around AI in robotics goes to its top. Every various other week, we see a slick trial of a robot doing something that seems like magic. Sorting laundry, making coffee, complying with complicated voice commands … it’s the future we were all assured in sci-fi movies.
At the heart of this revolution are Generative Pre-Trained Transformers (GPTs), the very same brainy design behind ChatGPT. The idea is simple: educate a substantial design on a statistics lots of information, and it will certainly find out to do … well, pretty much anything.
It seems amazing. It looks impressive.
But what happens if this whole method is a huge, expensive, and essentially flawed dead end?
A recent perspective paper released in npj|robotics called Are transformers absolutely fundamental for robotics? just gone down, and it’s generally throwing a huge container of cold water on the whole hype train. The authors, James A. R. Marshall and Andrew B. Barron, are asking an inquiry that’s both basic and …