3 ms·
Right now, AI need a huge number of examples to reach decent performance. Therefore, things where a huge number of training examples don't exist and can't be ge
by ben_w 3y ago
Right now, AI need a huge number of examples to reach decent performance. Therefore, things where a huge number of training examples don't exist and can't be generated by a simulation or self-play. When AI get past that problem, I expect the whole global economy to shift in a couple of months[0].
Also, things where the point is to be human. Working for rich people in any capacity may fill this role, for the same reason they by watches which cost more than apartments yet keep worse time than a Casio F91W.
[0] I'm also assuming this is more likely to happen after we get a significant number of android robot workers; if it happens before that point, then it may take a year or two for the global economy to shift.
- leashless 3y agoCame here to say this. And to add: we don't know to what degree human behaviour is regular. It could be that in practice an AI can't really generalise well to All People compared to humans, because the humans are so diverse and the humans who are doing some jobs generalise to just the small set of people they are working with. Right now it does look like prompts like "answer this as if you're one of Napoleon's generals" work fairly well, but I wonder if in practice we'll find out that specific human experiences which are hard to document will turn out to create niches which AI is very very poor at dealing with. Potential examples: * what does it feel like to fall in love? humans can't describe well. * what does it feel like to summit Everest for the first time? ditto. * what are spiritual experiences like? Probably many other areas which are clearly important to people, particularly falling in love, but language is weak, and at the end of the day... It's a large language model.