Oh, I am sure there will be a healthy demand for “personalized” varieties
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Tesla, TSLA & the Investment World: the Perpetual Investors' Roundtable
- Thread starter AudubonB
- Start date
FSDtester#1
"Circle Jerk Old Guard" per cyborgLIS
Unfortunately it was not sarcasm ! Many people have no clue, as stated earlier, that there is no comparison to be made between Musk and Bezos. One is in it for $$, one for a better world. We all know who is who... at least in this forum.
Yup. This is how my Optimus army will walk.Oh, I am sure there will be a healthy demand for “personalized” varieties
ElectricIAC
Good-Natured Rascal
And he won’t look a day older.
Uglier? You planning to marry one of them?
I saw another reply on similar lines. First I thought I will ignore.
I am fully clued in on the AI part of the humanoid robot, which is path breaking, ground breaking, earth shattering and that is what will make this super useful in many situations and no one else is even doing any work in this area . Robots doing niche singular work on a controlled environment to general purpose intelligent tasks on a chaotic world is a huge leap.
I get that. I don’t need to be schooled in on that.
If you go back and read my note, I specifically only made an observation on one aspect - locomotion aka walking. And that seems to be a solved problem given where BD and many others are.
StealthP3D
Well-Known Member
Even cars without FSD can be used to collect edge case data so I don't think the decision to do wide release would be driven by the need for more data.
The "march of nines" simply refers to the perfecting FSD so it only screws up once in a great while. And, yes, that is generally thought of as solving edge cases. My concern is that these edge cases are so varied that the current system doesn't have the resources to accommodate them all. Because there are a lot of them. If I'm right, and I hope I'm not, FSD (with current hardware and techniques) will continue to improve towards a plateau that is just below what is necessary for it to reliably drive without supervision.
Previously I thought as long as the system was statistically safer than the average human driver, say twice as safe, then it would be criminal to not approve and implement it. But I hadn't really considered there was a third option, allowing it be used only with human supervision (as is currently the case for those with FSD). If FSD alone is twice as safe as a human driver alone, then FSD with human oversight might be 3X-4X safer than a human alone (and 1.5X-2X safer than FSD alone). That is a powerful incentive to not approve it for fully autonomous operation (because we now have an even safer option). I had failed to consider this conundrum previously.
To explain this more fully, I've long held that FSD will still have accidents, but they will be fewer, and on-point here, they will be different types of accidents than humans have (most of the time). That is to say, FSD will prevent most accidents human have because those accidents are concentrated around not paying attention at key moments. But FSD will fail in odd situations, maybe due to a "glitch" in the system (for lack of a better word) that will cause an accident in a situation that would be unlikely with a human. Thus, human oversight can make the system safer yet. And I think auto accidents are common enough and the results of accidents are serious enough, that it makes sense to keep the system as safe as possible before the requirement of human oversight is lifted. In other words, I don't think the requirement of human oversight can be lifted until human oversight cannot prevent a significant number of accidents from happening. This is a big change compared to how I previously viewed the ethics involved. Which was, as long as it was definitely safer than a human alone, it would be criminal to not approve it.
Tesla's own FSD system with human oversight is competing directly for safety with Tesla's FSD system without human oversight. And I think that might explain why Tesla raised the price of FSD so high before it has that high of a value. This positions FSD as something that will not be used by a large percent of automobiles unless it can be used fully autonomously which bolsters the argument of approving it sooner with no human oversight because it's safer than the current status of mostly 100% human only drivers.
On the other hand, if unsupervised FSD can be 10X as safe as a human driver alone, then these arguments largely dissolve (because the accident rate becomes so low that it's no longer logical to require supervision.).
The significance of this conundrum is lessened if the "march of nines" happens quickly via exponential improvements in FSD, and this significance increases if improvements flatten out. However, my impression is that experts in the field like Musk and Karpathy have always expected improvements to flatten as the system approaches autonomy, hence the "march of nines". Regardless of how it plays out, it's exciting to watch, and I do not lack lottery tickets in case things happen sooner than I expect!
Last edited:
TL-DR; Probably better in the FSD thread, but hey, it is the weekend! It is happening faster than I think most realize, but I think you have a good grasp on the challenges at hand. And I realize Tesla has hit some local maxima in the past and had to re-architect, but this time, it seems apparent that their approach has no fundamental flaws and they've exposed nearly the entire process. And the more I typed below, the more questions might come up, so happy to respond if ping'd in the FSD thread...
To address your points:
I think a practical ramification would be that it demonstrates safety is higher with FSD being enabled and thus those with it are inherently safer on the road with it on vs off. And being able to recognize revenue as well.
The method of releasing beta first to a small subset of objectionably measured extraordinarily safe drivers I thought was brilliant. Demonstrate >10x safety statistically, then expand and repeat. At wide release, all the eyes of the world will be descending on the data, even more so than it is now. This is great and good as the data will demonstrate >10x safety. Possibly even much higher like >100x based on what I'm calculating.
For wide release, I would think that they would have objective and subjective key metrics by which they have determined would be a minimum level of practical usefulness and safety in order to release it wide. I'd imagine that one would be what percentage of miles driven is with FSD on vs off. I use it basically all the time, but until it gets smoother with near zero unnatural slowdowns, my wife won't use it. This is why I say that wide release is imminent as they are targeting and have largely solved these top objective and subjective issues with the current 10.69.3
With 10.69.2.4 there are still legitimate safety issues. I have two that are recurring on my normal routes (a UPL without a creep wall and several multi-lane roundabouts) but that is entirety of where it fails for critical safety.
Do they need to solve these prior to wide release? No, as FSD will still have "beta" in its name.
Does it need to demonstrate objective >10x safety? I think they are targeting this.
Do they need the majority of people to use for a majority of their miles? Totally, because that is the key to getting more useful data.
Do they have a way to consume the vast amounts of data that will be produced? That is what the 'human-out-of-the-loop-auto-labeller' and Dojo is for. Obviously, they can do that with their GPU clusters in the meantime, just in slower iterative steps.
Is it possible they could hit a "local max" (aka technological wall as you put it)? Sure, but that is highly unlikely at this point and it seems 10.69.3 is a test of that. This build seems like the first real step towards a wide release candidate. The items in the release notes suggest they are polishing out as much as they can to output a very stable release.
The other thing that I want to address for folks is this idea that the build is going to get exponentially better. For me, it has, however, this is most likely vastly different as a subjective discernment for everybody. Normal human opinion is great, and valued, as well as the objective measures. Objectively we can measure it simply as (FYI, these are my actual numbers):
1. What % of the time is AP used on highway? >90% (meets criteria for wide release)
2. What % of the time is FSD used for straight surface streets? >70% (needs to be >90%)
3. What are the top issues holding back the biggest gains for surface streets?
- 40% of all interventions are due to unnecessary slowdowns
- 30% are due to short duration lane change or missed turn due to wrong lane
- 20% are due to the high jerk rate in steering wheel and throttle
- 10% are due to construction/school zones and roundabouts
- 10% other
So, practically, you'd start looking at ways to address these and I think they have done that with this build.
Assuming they achieve wide release, then what is next? This is then the true march of 9's.
My worries are that when their current implementation of the lane connectivity graph is applied, that is it not able to achieve a 99.99999% success rate overtime (where overtime could be a year) and the neural planner on top of that is also not able to achieve a 99.99999% success rate. These stacked tolerances are key to fully human unsupervised driving. The reason I see it succeeding is that they are not only optimizing the NNs, but are adding new ones to essentially supervise other NNs. While this might be considered a 'crutch', this is a faster way to get to a full solution as one NN might be able to achieve some aspects of driving to an exceptional high degree, while others it might not be as good at. Now take that output and apply it as input to another NN that *CAN* achieve high levels of success. This is why they have a 3d occupancy NN and then it's output is an input to the occupancy flow NN and then outputting to an object detection NN which has outputs for vectors of those objects, which is then used for the neural planner and then for control. All of these have tolerances or limits on their accuracy, precision and recall. Stack them all together and it becomes much less. But if you can have a layer supervise another, correct those issues, before they are pushed to another (aka two layer model) then you start to reap benefits of one layer without consuming the entirety of its weaknesses.
And to close out, this is why I'm most convinced they have the right architecture which is the the lane connectivity graph. When this outputs *about* where the destination is located (which is beyond the current perspective of the real-time camera system), the neural planner can essentially *always* start you about in the right direction. This might sound simple and obvious, but to create this lane connectivity graph was seemingly profoundly difficult and complex. Have you noticed how on AI day they used satellite photos of intersections to demonstrate how they build the topology? Obviously, they can't use satellite images as they become stale immediately and thus could not be used. They must be harvesting data from the fleet they know has a certain age (freshness) and then building out their understanding of the intersection from those images (not necessarily NeRFs, but enough to build the topology), guiding the car with that learning and at some point this information goes stale and has to be refreshed via the same or similar process. It is this process which will enable the foundation of self driving, being able to graph the possible destination options and then predicting which one is best and being right 99.99999% of the time.
StealthP3D
Well-Known Member
I'm of the mind that Tesla will use AI to teach the robot how to walk naturally and efficiently. The current state of walking is just a simple machine walking to jumpstart the training.
Knightshade
Well-Known Member
Generally speaking- yes- they're not going to have enough seats to override a veto in either chamber let alone both.
Though they could potentially attach language that'd repeal or cripple some or all of it to "must pass" type bills and see who blinks first if they care that much (or even hope nobody notices if it's a subtle change snuck into an amendment to some unrelated 1000 page bill)
ElectricIAC
Good-Natured Rascal
So itll probably just be a lame duck session for the next two years.
No chance of removal at all. Any removal must PASS ALL three chambers. House, Senate, and President must agree. Overriding a Veto is virtually impossible in closely divided Senate and House.
insaneoctane
Well-Known Member
IIRC, we saw women from Optimus team present at AI day #2? Statistically, for every 7 men you should see 1 woman. I agree that having zero in their recruiting videos might give the wrong impression and is potentially a missed opportunity to appeal to about 15% of the potential workforce. Hopefully the top talent in the field, regardless of gender, is attracted to Tesla because of the appeal of the mission...
Correct. Up until AI day, no bot had ever walked via AI NN. BD (owned by Hyundai) does it via direct heuristic coding, and I'm sure a few other tricks, but they've recently started hiring AI folks...in August!
íëìë차그룹
íëìë차그룹ì âìì íê³ ìì ë¡ì´ ì´ëê³¼ ííë¡ì´ ì¶âì´ë¼ë ì¸ë¥ì ê¿ì ì¤íí´ ê°ëë¤. íëìë차그룹ì íëê³¼ ìµê·¼ ììì ë§ëë³´ì¸ì.
FSDtester#1
"Circle Jerk Old Guard" per cyborgLIS
If memory serves, that statistically means the markets should do better in a divided congress due to gridlock and large corporations being able to plan long term without the possibility of a new law that might hinder a business decision.
Edit: i.e Market returns for Trump's first 2 years vs his second 2 years. Wallstreet was loving on Democrats taking the House.
insaneoctane
Well-Known Member
Excellent and very helpful. So, based off your comment, does anyone else feel like Optimus's walk is the robotic equivalent to "the uncanny valley" and does it bother enough people that Tesla would go though the added expense to eliminate it?The human knee has a screw home mecanism where the tibia goes in external rotation compared to the femoral condyles. The geometry of the tibial plateau makes it easier and effortless to stay standing with a locked knees in full extension.
When rubbing you never go into full extension because you need your knee to using biarticular muscles like the hamstrings and the gastrocnemius to bounce and recoil 17% if the energy stored in the buarticular tendon collagen for more effective bouncing while running.
Optimus will never had a normal gait if they don’t force it to walk like a human being. If they use actuation motors in the joints there is no reason to reproduce a screw home mechanism as the motors just lock the joint in the desired position without having a muscle or tendon to fatigue to keep a joint in a desired position from a prolonged period of time.
This was also a publication I have made and presented on during my masters degree in training so I had to share that useless knowledge I have not used for over a decade.
Thank you for listening to me
Artful Dodger
"Neko no me"
OT
from Marco Marco NYFW 2018: Very first New York Fashion Week show of its kind using all transgender models...
insaneoctane
Well-Known Member
Similar threads
- Locked
- Locked
- Locked
