What you described (incremental learning) is the holy grail of AI: AGI. Some people believe we are at the cusp of achieving it. I don’t know enough to form an opinion yet.
But the recent progress has shown you don’t need AGI to solve many interesting problems that seemed to be unsolvable just a few years ago. We have reliable self driving available today. Commercially. You can literally just tap on your cell phone in SF and a driverless taxi will come to your door.
Only Tesla fanboys think self driving is years away. Yes for Tesla cars it is. But not Waymo and cruise. They are available now.
You can buy this self driving kit from Comma AI and put it in your car.
Again only Tesla fanboys think self driving is waiting for Elon’s magic touch. L2/3 is pretty widespread already. L4 is getting rolled out commercially by both Waymo and Cruise into more and more cities.
Btw they don’t hire remote. On site in San Diego only.
Your construction of the statement implies the opposite of what I have said.
Subset i.e. limited scope, always possible. Too many situations, can’t do. So is not ubiquitous yet. I won’t use the word, “solved” without specifying the applicable scope and limitations. (Context ). I would use good enough rather than “solved”.
Everybody think whatever they believe in is the best ever. May be is true. At this juncture, TSLA seems to be behind but EM could release a FSD that leapfrog everything in the market.
What does real understanding mean? How do you measure that?
If you don’t know whether an answer came from chatGPT or a human, can you really tell who has “real understanding” of a piece of text?
And does it matter whether that human has advanced degree or some random Joe who didn’t finish high school?
What I am getting at is that we shouldn’t get too hung up on implementation details. Whether a piece of software runs on a piece of flesh aka human brain or 800 Nvidia GPU’s, does it really matter?
Something requires Skill X and therefore it’s all Skill X and nothing else?
More importantly why does it matter how reading comprehension is implemented in AI? Using techniques that involve pattern recognition means it’s not “real understanding”?
Have you figured out what “real understanding” means and how you measure it?
Getting too academic for me. It started off from a comment that many ways of viewing … so far, no views are sufficiently rigorous… pure idle chit chat. Surprisingly, some people love to go in circles.
It’s not circular for me. It’s also very meaningful. For me anyway.
Many people think just because machines use a different way to achieve a result means they don’t actually achieve it. Hence your “real understanding” comment. Even though if you score it objectively machines’ skill levels on some tasks already higher than humans.
Many Tesla fans also think that way. Because humans can drive with vision only input therefore Waymo and Cruise cars’ self driving doesn’t count. They use LiDAR and high definition maps. Real driving AI shouldn’t need those.
IMO they lack imagination. Who says intelligence has to come in one specific form: human? Planes don’t fly by mimicking birds flapping wings. Cars don’t have legs. Washing machines don’t have little hands inside rubbing clothes. Machines work differently than humans.
You love to idle chit chat. Merely pointing out your opinion wasn’t the only way to view it. Not at all trying to prove anything other than that. Then you go in circles and start talking cock. Too lazy to comment on your latest talk cock… some I agree, some my toes are laughing…
Dell can deliver servers with A100 in 5 days and H100 in 15 days now. Either nVidia MASSIVELY scaled manufacturing or demand is already dropping. It wasn’t hat long ago lead times were several months. SuperMicro has similar lead times.
). I would use good enough rather than “solved”.
