"With GEN-1, though, Generalist says its physical models have reached a GPT-3-style inflection point, where some tasks are starting to “cross the level of performance needed to be deployed in economically useful settings.”
I think humanoid robots are one of the sleeper tech trends most people are underestimating. They don’t need AGI, or even ‘perfect’ AI, to do most unskilled & semi-skilled work. With enough development & training, today’s AI models will probably be fine. Here’s another sign that this hypothesis might be true.
How soon will they get there? At current rates of development, 2030 seems a reasonable estimate for general-purpose humanoids easily trainable for most unskilled/semi-skilled work. Just when most driving jobs will be disappearing to robo-taxis. No one seems prepared for this future rapidly bearing down upon us.
Neural networks are universal approximators. Anything with complex inputs and known correct outputs will gradually become tasks we expect to Just Work. Especially when failure is perfectly safe. A self-driving car has to be better than most humans, all of the time, before we’ll tolerate the cases where it’s already fine. An assembly line of arms folding cardboard does not have this problem. Nor does a Roomba that can sift a litterbox. Things can go wrong… but in a way that’s quietly hilarious, to everyone but you.