That's probably based on the improvement he's personally seen since he's been in the Beta program, which dates back to Beta 8.x. So most likely "great" in respect to what he's seen as far as progress goes since then.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Rocco on 10.6.1 with two of the nearest misses I've seen in a video from the bigger Beta testers
Rocco uses one of them as what he thinks is the perfect example of the B-pillar camera limitations
Rocco uses one of them as what he thinks is the perfect example of the B-pillar camera limitations
This is a well done video with a mixed bag of good and bad results. The video is very well done technically with some unique camera angles (including the one in the thumbnail below).
EVNow
Well-Known Member
The first error comes after ~10 minutes. Wonder if it would have gone away if he had done a reset.
What was Elon smoking when he said in January that Tesla would have Level 5 by the end of the year?
And don't say weed, because weed does not impair your cognitive functioning that much
EVNow
Well-Known Member
This is the experience of the industry with NN
- Takes a long time for neural network to do basic things humans can do easily
- But once that happens, very quickly NN comes to human level and then surpasses the best human
Well, Elon thinks this will be the case for self-driving too. IIRC, he has explicitly talked about this.
Did anyone else in the autonomous vehicle space believe that we could ever possibly have Level 5 by the end of 2021?
What has changed so fundamentally in not even 12 months?
Why is Elon still making unachievable promises with 100% certainty?
Not on topic, but recent videos seem to be illustrating how staggeringly far we are from anything resembling Level 5. Lets say we cut out the rest of the world and geofence FSD strictly to the United States, which I think already makes it Level 4 at best. When do we think we could see a consumer production vehicle with these capabilities?
Do the vehicles even have cameras in the spots necessary to achieve this? Are other sensors truly not required? There are a lot of questions around whether this is even possible with the current setups much less achievable in a calendar year. I have a really hard time imagining being so far off on this, especiallyon the back of six years of similar unmet timelines.
Last edited by a moderator:
EVNow
Well-Known Member
EVNow
Well-Known Member
AP includes "sensing" - so not sure what you are saying.
Are you saying since only part of FSD is NN and rest not NN, he can't apply NN learnings to FSD ?
There are actually a lot of reasons NN lessons don't seem to apply to autonomous vehicles. Thats a different thing - I'm just stating why Elon seems to be so bullish on FSD unlike most of us.
Haven't watched this yet but FSD is pretty solid when it comes to just navigating roads and lane lines and curves etc, complexity and especially interactions with other people are where it really struggles
Came across this last night, two guys who recently got their hands on FSD. One of them says he has watched hours of footage from Beta testers and thought he would be prepared for the experience. Starts out really strong but eventually ends in the one dude looking like he's about to get sick
Karpathy tweeted this yesterday in response to the above Tail of the Dragon video
The new lead is still calling it a baby! There was zero chance of this happening, and we have no idea when Level 5 will happen. Elon needs to stop overpromising, stop setting people up for disappointment when yet another promise fails to materialize.
Last edited by a moderator:
All of a sudden, FSDbeta doing well is no big deal.
I doubt this has been repeated by other ADAS systems and Tesla is the last in line to do this. No instances of FSDbeta spontaneously going into the opposite lane after all of those turns. He went both ways, as well.
I doubt this has been repeated by other ADAS systems and Tesla is the last in line to do this. No instances of FSDbeta spontaneously going into the opposite lane after all of those turns. He went both ways, as well.
EVNow
Well-Known Member
I guess you are also talking about FSD Beta when you say AP ? I guess the idea is to move more and more stuff to NN from AP. If once they move a large portion of AP to NN - and it works flawless ... you have close to L3/L4, if not L5.
EVNow
Well-Known Member
New lead ... ? Karpathy has been there for quite some time now. Looks like he shares my view FSD beta is like a 2 year old baby
Though, I think he is probably using the term differently (xyz project is my baby) - not in terms of abilities.With Elon, you take the bad with the good. You learn that over the years if you are an investor - its maddening to see him tweet something moronic and the stock tanks.
FSD Beta has done quite well in these types of situations for a while now. Even city roads at night when an update first hits, many of the Beta testers post up zero disengagement initial drives. Open roads, good lines, no traffic, decent weather, that's a breeze for FSD.
There's nothing particularly challenging about the road shown in this video, it's dangerous because people race it
That said, the guy complains about the car hugging the double yellow line and describes some of the turns as "dicey" for that reason
My bad, I thought Karpathy took CJ's job as the Director of AutopilotNew lead ... ? Karpathy has been there for quite some time now. Looks like he shares my view FSD beta is like a 2 year old baby
With Elon, you take the bad with the good. You learn that over the years if you are an investor - its maddening to see him tweet something moronic and the stock tanks.
Similar threads
