You're right. Nevermind. I'm perfectly happy in the way EAP and FSD are being marketed and sold. Keep it up team.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
You're right. Nevermind. I'm perfectly happy in the way EAP and FSD are being marketed and sold. Keep it up team.
@TaoJones is no Tesla hater, far from it. But I think all of us, (and as you all know, I am a proud fanboy), feel a bit frustrated by the unforced error of expectations that the video created... My hope is that frustration doesn't end up festering into rage... I'm still optimistic but the longer we go without a cross country drive, the louder it will get...
That being said, I'm glad they haven't resorted to panicking and doing a cross country drive by cheating, by highly managing the route and escorting the cross country football with the help of undercover ICE Offensive Linemen cars that limit the potential road hazards and ambitious lane lines along the route. Let's get some progress... try to hurry along Karpathy!
I am happy to see easy entry on the latest firmware, which means Elon is tweeting with some sort of connection to his team.... hopefully we will see the dash cam feature he mentioned sometime in the next year in some capacity.
That being said, I'm glad they haven't resorted to panicking and doing a cross country drive by cheating, by highly managing the route and escorting the cross country football with the help of undercover ICE Offensive Linemen cars that limit the potential road hazards and ambitious lane lines along the route. Let's get some progress... try to hurry along Karpathy!
I am happy to see easy entry on the latest firmware, which means Elon is tweeting with some sort of connection to his team.... hopefully we will see the dash cam feature he mentioned sometime in the next year in some capacity.
strangecosmos
Non-Member
For better of for worse, Internet is mostly anonymous, so need to take appropriate amount of salt with everything.
stopcrazypp
Well-Known Member
I've seen this mentioned as "evidence" before of zero progress, but the CA disengagement reports are released annually, the next report after the previous one is not due until January of next year. Not having one can't be used as a metric for progress. Also, many of the self driving companies have moved their testing to other states that do not have such reporting requirements (Tesla can do the same).
Testing of Autonomous Vehicles
How do you tell a Tesla with FSD software active vs one that doesn't have it active? The fact that you can't is the whole point (hardware is the same as production cars). This is completely different from Waymo and Uber's approach where it's obvious that it's a FSD car.
The first video had eyewitnesses, but it was a big production (had drones filming). The second video had absolutely no eyewitnesses.
Last edited:
strangecosmos
Non-Member
I think you have a point here, but I don't believe the attacks are coordinated. That sounds like a conspiracy theory itself. I think the attacks are spontaneous and emergent. They come from cognitive biases, bad epistemology, cynicism, pessimism, misleading reporting and analysis, and other uncoordinated phenomena.
What is the question they ask?
strangecosmos
Non-Member
stopcrazypp
Well-Known Member
I was completely blown away by a post over at reddit yesterday.
A guy called PM_YOUR_NIPS_PAPER - who says he is / was training Tesla's computer vision network - claimed:
I don't know if this guy's for real or not. Nevertheless, I find what he's saying a bit scary. Could it be true? What pieces of evidence or indications do we have either way?
A quick Google search tells me there are several software tools out there that could be used for object tracking / annotation. Even YouTube seems to have a tool for this.
YouTube-BB Dataset | Google Research
ViTBAT - Video annotation tool
Random video:
I really hope this isn't the case. Unfortunately that dude's post got me thinking ...
Tesla Self-Driving Demonstration
I think it helps to break the statement down into parts.
First thing to keep in mind is that there were two videos. The first one has already been documented to have been done with many runs and disengagements, while second video had no disengagements. I exhaustively documented all the details of the timeline here:
FSD may require a hardware upgrade...
First of all, doing this does not conflict with the self driving definition. The SAE definition excludes strategic effort from the dynamic driving task (page 29): "Strategic effort involves trip planning, such as deciding whether, when and where to go, how to travel, best routes to take, etc. ... The definition of DDT provided above (3.4) includes tactical and operational effort but excludes strategic effort."
Also, pre-knowledge of traffic light locations is quite common in self driving cars (you should know this already from all the previous discussions we have had).
https://wiki.unece.org/download/attachments/40009763/(ITS_AD-10-08) SAE_J3016_Taxonomy and Definitions for Terms Related to Driving Automation Systems.pdf?api=v2
How to parse this depends on which video this refers to. The first video we already know that they did multiple runs (most of the days had rain) and picked the one that didn't have any disengagements. There was even speculation that the video was pieced together from multiple runs (but no evidence).
If this refers to the second video, we know there were no disengagements, so doing multiple takes is likely just the standard convention in commercial video to have multiple takes (so you have choices to pick even if all of them were "good" takes). That one did have some interesting parts like it slowing down for some joggers.
I don't find this surprising. If you zoomed in on the video, you will notice that in the car there was none of that visualization, but rather the same UI as in regular AP. The AI doesn't need that visualization to operate, so that they spent no effort on the FSD UI for human visualization doesn't really tell you much.
Rather, the core question is if the car was doing any of the object detection/classification, lane line mapping, and road surface pixel labeling under the hood. If not, then that would be worrisome. However, the person never addressed this point directly.
One easy answer to this question is the two have different codebases or FSD is in a branch that has yet to be released. From what @verygreen released so far, AP2 code so far appears to be mainly emulating the functionality of the old mobileeye chip, which doesn't make sense as an architecture for FSD. The emulation is probably so they can more easily maintain both AP1 and AP2 (don't forget they still have to support AP1 owners even though Tesla broke with Mobileye).
Last edited:
Which is why everyone's blood pressure goes up if they spend too much time online.
I wonder if you missed this irony.
We just discussed how Waymo doing mapping first is somehow different from how Tesla is doing FSD... Now all of a sudden we are discussing Tesla also using a mapping-first approach to get started. The difference a few days make.This is true. EAP is only representative of failures to meet EAP promises. It is not likely representative of FSD status, though Elon's comments about the NN in the most recent conference call suggests the same NN developed for EAP is also at the core of their FSD efforts, so the delay in getting that done may well be shared by both branches.
stopcrazypp
Well-Known Member
Using waypoints for a route doesn't suggest a "mapping-first approach" only that the system may not be linked with the onboard navigation (rather they upload coordinates to the car).I wonder if you missed this irony.
Neither does using a map of traffic lights, which is still only supplementary (a flip of the lidar based systems depending on camera to check on light status). The difference is the visual approach doesn't depend on syncing to that centimeter accurate map of the entire environment (including lanes, curbs, walls, buildings, etc). It's doing that detection based on visual processing.
But the suggestion here, if the rumor is to be trusted, is that indeed the FSD video was counting on mapping and geo-fencing. Just like Waymo is. Do you think the Tesla could have ventured outside of its plotted route in the video?
My point is the same as it was in the Waymo thread: unless people artificially want to see Tesla in a better light than it is in, it has to be admitted we have no idea how they will release their approach to the public and where Waymo is at that time. It is well possible Tesla has to rely on mapping to allow FSD, just as it is possible Waymo will no longer rely solely on mapping by the time Tesla's FSD is out. All we know is Waymo uses mapping to get started, and there is rumor of Tesla doing the same.
This does not tell us much about the future of their approach, it just tells us that starting off with mapping is easier than going free roaming. Likely everybody starts off with limitations that may not be there in due course. Starting with a geo-fenced solution seems like a likely approach for many Level 3/Level 4 players IMO, but that doesn't mean their approach - whatever it is - fundamentally has to stay at that level.
NerdUno
Member
Just occurred to me that Elon may have left himself an out on the cross-country trip. He just said it would be with no hands. But did he specifically rule out towing the car behind a U-Haul truck? Still would be with no hands, and it would really save on all those pesky pit stops for electricity. And they could hide the truck at night so we could have some juicy photo ops and a few more videos.
Well... me for one. I never once imagined that this was the state of their production software. Tech demos are routinely fake. It's well known in the tech industry.
The intent was to demonstrate what the product could do with the hardware installed. It did that. And it served as a commercial for waht would be available in the future.
Now... there is a legit but different question as to whether Tesla's selling of FSD is ethical. They clearly haven't met their timelines. And I think they're a lot further away than most people expect, even still. I think real FSD is still a decade or more away. And I don't think the current hardware suite is remotely capable. But what do I know?
3Victoria
Active Member
Buy now at a great special price, or wait and pay full price -- that is the offer.
In addition, Tesla has made it quite clear that early adopters are paying a premium to pay for new tech.
In addition, Tesla has made it quite clear that early adopters are paying a premium to pay for new tech.
Those two sentences are literally in contradiction. Read them again. Low price now and premium for early adoption?
Similar threads
