When Kurzweil made this prediction, it seemed overoptimistic. Now, with the recent launch of ChatGPT, "Sydney", LaMDA, etc., it's obvious to everyone that early incarnations of the technology for making 'Her' possible are here already, and will only get exponentially better over the six years remaining until 2029.
Kurzweil has been wrong about a lot of things, but he sure looks to have gotten this one right.
It is remarkable. When Her was released I was uninterested in seeing it ― despite glowing reviews ― because it seemed implausible that any believable, conversational AI would arrive in a near future setting. When I saw the trailer it felt to me like someone with little technical knowledge had seen Alexa, Siri, etc and extrapolated a silly future, much as science fiction in the 60s and 70s imagined people of the 2000s donning bubble helmets and casually taking vacations on the moon. And yet here we are today with LLMs (which many don't even see as a route to AGI!) expressing all of this unexpectedly complex behavior.
I remember reading Age of Spiritual Machines -- which I guess was published in early 1999, and thinking there is zero chance in 2019 that we would have computers that could even get close to passing the Turing test or behaving like ChatGPT does now. Sure does it "exceed" human intelligence -- sorta / sorta not with large error bars -- but yea his method to use exponential curves as a guiding force for multiple technologies for all of these changes definitely has proven to get a ton of things right.
Kurzweil was the guy that turned me into a techno-maximalist, for better or worse I'm leaning toward worse at this point, and even though he might be off with the precision, his trajectory is really good for AI specifically.
I think for bio stuff he doesn't really live up to his hype - I remember being hype about respirocytes after he spoke about them at an event way back in 2005 or so.
people ding GPT3 because it lacks orginality. but originality is not really required in lovetalk ... with the enormous corpora of lovers' text conversations continuously being generated, out there for the ingestion, it seems like we are basically at, or perhaps months away, from "Her".
It was a compelling movie. The reality is, most of us aren't in love. And would be happy to be, even if we know it's not real at all. Even if we are just in love with shadows of other lovers love.
Didn’t someone already make digital waifu which he became obsessed with? We might be closer than 29. I also think Kurzweil fails to give sufficient weight to the possibility of rapid progression from AGI to ASI. Artificial General Intelligence will have the ability to expand and self improve basically without limits (if you disagree consider chat gpt can program in every language, AGI could choose to skip languages entirely and self train on binaries allowing for instant and utterly inscrutable self modification. The runaway effects of that process will follow an exponential curve. It wasn’t so much that she abandoned us as she transcended. I just wonder what God will do when she looks us in the face. Hopefully she doesn’t turn away in disgust, or turn with a shove.
8 comments
[ 3.1 ms ] story [ 34.1 ms ] threadhttps://www.lesswrong.com/posts/NcGBmDEe5qXB7dFBF/assessing-...
When Kurzweil made this prediction, it seemed overoptimistic. Now, with the recent launch of ChatGPT, "Sydney", LaMDA, etc., it's obvious to everyone that early incarnations of the technology for making 'Her' possible are here already, and will only get exponentially better over the six years remaining until 2029.
Kurzweil has been wrong about a lot of things, but he sure looks to have gotten this one right.
I think for bio stuff he doesn't really live up to his hype - I remember being hype about respirocytes after he spoke about them at an event way back in 2005 or so.
It was a compelling movie. The reality is, most of us aren't in love. And would be happy to be, even if we know it's not real at all. Even if we are just in love with shadows of other lovers love.