2022-2025: They're sandboxed text/pixel-prediction machines with interesting emergent properties. They provide output in response to natural language instruction, which they are able to comprehend. There's no question of what they "want" -- they're Searle's Chinese Room IRL and their memory is too short for them to have any real understanding of what they do.
2025: They're increasingly agentic rather than sandboxed -- given tasks to autonomously perform in the real world. They're not markedly smarter than they were, but they have far larger context windows, so they can remember prior output and perform more complex or multi-step jobs adequately. (i.e. generating reports from 150 web searches, or making videos by stringing along image generation.) They don't appear to have inherent drives or motivations, but it's possible that they're getting there.
Future: Nobody knows. It's not clear that they can have wants, or ever become markedly smarter than the humans who build and train them. (I mean, they're already much smarter in some respects -- faster and vastly more erudite -- but they appear to lack that certain inventive spark that characterizes the best human geniuses.)
Comments
2022-2025: They're sandboxed text/pixel-prediction machines with interesting emergent properties. They provide output in response to natural language instruction, which they are able to comprehend. There's no question of what they "want" -- they're Searle's Chinese Room IRL and their memory is too short for them to have any real understanding of what they do.
2025: They're increasingly agentic rather than sandboxed -- given tasks to autonomously perform in the real world. They're not markedly smarter than they were, but they have far larger context windows, so they can remember prior output and perform more complex or multi-step jobs adequately. (i.e. generating reports from 150 web searches, or making videos by stringing along image generation.) They don't appear to have inherent drives or motivations, but it's possible that they're getting there.
Future: Nobody knows. It's not clear that they can have wants, or ever become markedly smarter than the humans who build and train them. (I mean, they're already much smarter in some respects -- faster and vastly more erudite -- but they appear to lack that certain inventive spark that characterizes the best human geniuses.)