How likely on a scale from 1-10 do you think it is that if use the same fundamental tech for one non-safety-critical and non-time-critical application and “fail” and apply the same thing to a safety-critical and time-critical problem and have it succeed?
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Elon: FSD Beta tweets
- Thread starter Terminator857
- Start date
AlanSubie4Life
Efficiency Obsessed Member
Wow so that confirmation repost by Elon brings in the timeline for cars driving themselves with no one into the end of 2024, doesn’t it?
This counts as Elon confirming this timeline, for sure. Since Twitter aspires to be the best source of truth on the Internet, this sort of post would quickly be hidden if it were not going to happen.
Big, if true.
Elon's timelines should generally be taken as aspirational, at best. I always shift the duration units to the next level. Week becomes month. Month becomes year. And year becomes decade.
I find this makes his (and any other CEO's) predictions much more accurate.
sleepydoc
Well-Known Member
At this point, anyone who believes Elon's time predictions is either naive, a troll or just plain stupid.
Daniel in SD
(supervised)
I think there’s zero chance end to end with existing NN architectures will achieve human performance (per Tesla, 1 collision per million miles).
Will the humanoid robots take over after or before that happens?
AlanSubie4Life
Efficiency Obsessed Member
Guilty on all counts. (Still have doubts about Elon’s timeline.)
Daniel in SD
(supervised)
Not sure about taking over but humanoid robots are easier because they don’t weigh two tons and move 60mph.Will the humanoid robots take over after or before that happens?
I disagree. Cars are operating in a highly structured environment. Robots (in a factory, or domestic) need to navigate a completely novel environment that it has never seen in pre-training. There are no traffic lights, stop signs or lane markings. Also it will kill your toddler or pet if it falls over even at 50-70kg and the noise level is probably 60-70dB at present from the motors and fans. Not happening before 2040 is my best guess. Probably later for domestic and outdoors (autonomous) operations. There are no ML architectures really capable of task breakdown and high level planning right now. There are no ML architecture that learn/adapt while deployed outside trivial sandboxes.
Last edited:
Daniel in SD
(supervised)
Lots of difficult to predict humans of all shapes and sizes in that structured environment though.
Good point about humanoid robot around kids though. It seems lien you could have humanoid robots in factories but I don’t see the advantage over other robot designs.
clydeiii
Member
ChatGPT very much lives up to its hype.
clydeiii
Member
clydeiii
Member
It's actually really good at creating cooking recipes. Seriously. Best use for it.
It’s really good at transcoding and explaining the transcoding line by line.
It can also write convincing legal briefs, but will completely make up citations.
I've had it write really good looking code that uses non-existent API calls. When I point them out, it will happily rewrite the code to use new non-existent API calls.
The recipes are more reliable.
AlanSubie4Life
Efficiency Obsessed Member
Been posted elsewhere but belongs here.
Tough to find relevant Tweets amongst the stream of diarrheal drivel these days.
Chief Cheerleader (mild term): No. Hardware 3 is a little smoother
Elon: Hardware 4 will ultimately be better, but all training is for Hardware 3, with HW4 running in emulation mode
Tough to find relevant Tweets amongst the stream of diarrheal drivel these days.
Chief Cheerleader (mild term): No. Hardware 3 is a little smoother
Elon: Hardware 4 will ultimately be better, but all training is for Hardware 3, with HW4 running in emulation mode
Similar threads
