As the head of Alphabet’s AI-powered robotics moonshot, I came to believe many things. For one, robots can’t come soon enough. For another, they shouldn’t look like us.
I’m still not convinced LLMs are even the correct path to true AI. I do not understand how it’s going to control a robotic body since it’s had no experience of physical existence in its training data.
Does it even comprehend the ramifications of gravity, object permanence, time, or anything other concept that an organism gathers naturally by living in a physical reality?
Programs have “learned” how to play games without instruction — most recently with rat neurons playing Doom — and that’s what they’ll attempt to do with LLM’s. It must learn from feedback, experience and interaction as we could never code something that complex.
I don’t believe LLM’s can achieve general AI, but humans are just organic pattern recognition devices at the end of the day — a brain in an organic machine that can sense a fraction of the world around us.
The problem is that LLM’s are dumb, thus dangerous when given autonomy, they’ll be used to wage war, and the military industrial complex is more likely to destroy us with autonomous LLM killbots than achieve general AI.
So Boston Dynamics but humanoid and with an LLM?
I’m still not convinced LLMs are even the correct path to true AI. I do not understand how it’s going to control a robotic body since it’s had no experience of physical existence in its training data.
Does it even comprehend the ramifications of gravity, object permanence, time, or anything other concept that an organism gathers naturally by living in a physical reality?
Programs have “learned” how to play games without instruction — most recently with rat neurons playing Doom — and that’s what they’ll attempt to do with LLM’s. It must learn from feedback, experience and interaction as we could never code something that complex.
I don’t believe LLM’s can achieve general AI, but humans are just organic pattern recognition devices at the end of the day — a brain in an organic machine that can sense a fraction of the world around us.
The problem is that LLM’s are dumb, thus dangerous when given autonomy, they’ll be used to wage war, and the military industrial complex is more likely to destroy us with autonomous LLM killbots than achieve general AI.
BD’s Atlas is a humanoid too
Article says they are using a number of AI technologies stitched together with regular programming, so object recognition, language, etc.