Anybody think the Gigafactory announcements, coast to coast drive will effect the stock price much? I think the Gigafactory thing could go either way depending on the terms and partnerships, particularly in China, and I'm not sure if the coast thing would be a big deal or not. I guess if they do the coast to coast thing AND download it immediately to all cars that might be big but if they just do the demo then people don't get it for a while that might be disappointing.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
- Status
Waiting4M3
Active Member
IIRC the field strength around a conductor scales with the current, so increasing voltage but keeping current constant should not increase the magnetic field. Does that also work in AC case?
Yah, the difference being with AC the field goes to 0 120 times a second, so the wires can vibrate, and that changing field interacts with the environment. At high voltages like transmission lines, the voltage field does interesting things too. Watch videos of the helicopter based line inspection crews for examples.
Waiting4M3
Active Member
Common metal or plastic parts that are machined should not take longer than a few weeks, assuming one is willing to pay to jump ahead in the queue with the suppliers. If the machined parts is large/heavy or require high precision machining, then the availability of capable supplier dwindles, and it becomes more difficult to push the schedule. I've seen equipment with hundreds of lb of large and high precision metal pieces taking ~3 months to build.
Things like circuit components such as microchips, vacuum relays, high voltage/power/RF connectors could also take long lead time. Many of them you can find off the shelf. But if you can't find them in stock, then those have to be built-to-order, again the tools required to build those are typically specialized with few suppliers, and 1-3 months lead time is common in this case.
Anstandswauwau
Member
Not an DL engineer myself – sorry to disappoint you, neroden! – but I've heard that temporal problems are often tackled with CNN LSTM (Convoluted Neural Networks + Long Short-Term Memory Network).
In low complexity scenarios I've seen people treating time naïvely like an extra dimension of the input layer.
MitchJi
Trying to learn kindness, patience & forgiveness
I think that they need to wait for battery prices to drop. Those prices will drop by at least 20-30% when the Gigafactory scales to 100% production. The production of packs is the main obstacle to production of the semi’s. So building out the Gigafactory will trigger the necessary cost reductions and will resolve the largest production challenges. The semi reveal was clearly trumpets blowing announcing the forthcoming arrival of Tesla 3.0, which I define as the point at which EV’s cost Tesla less to produce than ICE’s.
Basically they are doing exactly what the other oem’s are doing (maybe less competently). They are waiting for battery prices to drop, as opposed to Tesla-Elon who are driving down the costs. We could help them out by sending them the relevant quotations by Elon and JB from the ER’s and shareholder meetings.
Its pretty simple (at least conceptually) actually:
1. Find and develop the best chemistry possible.
Nobody else is doing all of the following. Possibly working on 2, but not to the extent that Tesla is doing.
2. Automate the heck out of the pack building procedures.
3. Design and build custom large scale cell manufacturing equipment.
4. Collocate all of the manufacturing processes.
They could either buy batteries produced using those principles or do it themselves but if they don’t do that their alternatives is to fall further behind. Daimler for example is planning to assemble cells into packs at a dedicated facility. That can’t be competitive with the Gigafactory, no collocation. Probably not buying cells produced at a very large scale either.
For those interested in Neural Nets and discussions, there are several threads in the autonomous vehicle sections which I find interesting:
Neural Networks
HW2.5 capabilities
Neural Networks
HW2.5 capabilities
Could you elaborate on this? I'm especially intrigued by naïvely.
Warning: where I get off track with questions like this because of epistemology. Time exists for us, for example, only because we have a digestive track, Prigogine might say. But then how could you have energy without motion and that is the time change with distance? Years ago I had a conversation with Kip Thorne after he gave a speech on campus. My puzzle was trying to confirm that time was really dimensionally equivalent to the other three of distance. If I recall correctly, he said you could express the universe in only three, but the math gets too hairy. That shut me up!
Edit: I think I need to cruise the appropriate threads posted @32780
Last edited:
Speaking of time. If you want to save some, look through the HW2.5 thread and look for @verygreen and @jimmy_d posts to start. Obviously others are very instructive as well, but lots of great data/observations from the two of them on that thread.
The Neural Networks thread is worth reading from the beginning. Good intro videos etc...IMHO.
The Neural Networks thread is worth reading from the beginning. Good intro videos etc...IMHO.
MitchJi
Trying to learn kindness, patience & forgiveness
The other thing that can be inferred from the semi reveal is that the cost of batteries and solar production are going to drop enough for Tesla to sell,electricity for $0.07 per kWh profitably. The implications for TSLA are huge!
Tesla has been successful in spite of the facts that due to the high costs of batteries EV.’s and TE cost more than ICE’s and coal power. Those things have been exacerbated by Tesla’s Gigafactory production problems, which are troublesome now, but we all share know that Tesla will solve them. Might take a longer than we hope. It sounded to me like they have designed and are building (probably grohmann?) custom robots to solve their pack production problems..
By 2019 or 2020 when those things switch Tesla will be the only company with those advantages. At that point they will be able to not only sell the semi’s at a good profit but the electricity to power those semi’s. They will be like a combination of Daimler (selling semi’s) and exon selling the fuel.
Instead of celebrating, we should be taking a victory lap, we have the “the sky is falling“ posts and people who don’t even understand the basic technology stating that Elon should modify his pricing. Those things are ludicrous!
Last edited:
Thumper
Active Member
Let's do a thought experiment. Assume we have FSD and it is X times safer than humans. Let's pick 3 times. We turn it loose and because of the law of large numbers, we get 3 fatalities early in the data like we did with the battery f*res. It will be difficult to surmount the screaming and naysaying that will result. We have to be so safe that we can get a significant number of miles in the database BEFORE the first fatality.
Anstandswauwau
Member
Yes, I'll try to do my best
Think of handwriting or optical character recognition (OCR) as a three-dimensional problem that is made out of a two-dimensional drawing pad with a height and width of 32 pixel. For the sake of it, let's say very pixel holds information of three bits:
- 000: Pixel is completely black (not written on)
- 001: Pixel is slightly grey-ish (not directly written on, minimal "bleeding" from neighbor pixel)
- 010: Pixel is a bit more grey-ish (not directly written on, a bit more "bleeding" from neighbor pixel)
- etc.
- 111: Pixel is completely white (directly written on)
But if we now want to factor in / t, we could just use a three-dimensional array for every frame (32 * 32 * 6 * t), with the result of pumping the newly created array into the fourth dimension. This four-dimensional array can then be fed into the CNN just like any other, one-, two- or three-dimensional array, hence calling this approach "naïve"
This is pretty common technique for 2D data + time (eg. stock charts) but you can theoretically add extra dimensions to datasets of any given dimension as described above.
If Kip says the math gets too hairy, you better believe him, haha.
Referring to my example above, I'd like you think of dimensions in the mathematical sense: Time is just a bit of (extra) information that isn't special in any way
EDIT: This is my last post on this topic. At least in this thread. Sorry for de-railing the discussion.
Globally Semis consume about $100B worth of diesel each year. If Tesla can supply the energy with solar panel and do it profitably, it's indeed a big market, worth about $35B per year. If the all in cost is 5 cents, selling for 7 cents, there is $10B profit to be made each year. Tesla will achieve it in the long term, solar cost goes down every year while efficiency goes up. Gigafactory 2 is in place. Low cost shipping with their own Semi is in place.
On top of that, sooner or later Tesla network will work. Tesla is aiming for something huge, shorts are arguing about mouse nuts.
adiggs
Well-Known Member
I agree with the direction. One quibble, based solely on my guess of how the market will actually evolve, is that it looks like you're thinking about charging in the traditional fueling station paradigm. That is to say, all of the current fueling station business shifts over to a new charging business (such as Tesla).
I expect that we'll see plenty of businesses that build some of their own charging infrastructure that they either use exclusively (all trucks return at night to a central parking lot / charging area), or a mix of own charging plus on-the-road charging. I don't have any kind of guess at the proportion - only that own-charging (comparabe to today's garage charging) will be >0% and <100% (not exactly going out on a limb there
).Thinking about the transition, I'm hoping the own-charging dominates the early deliveries, and it looks from a Tesla perspective like they sell a fleet of trucks along with the hardware and installation services to install an own-charging system at the customer's facility(s). Which the customer pays for relatively quickly (such as 1/3rd up front for the charging system, and balance due on system completion) -- because that'll help cash flow, and help Tesla scale the truck AND energy business faster from nothing to something.
MartinAustin
Active Member
Somewhat OT, but back in the days when we all thought the Model 3 was going to have a HUD, this is basically what I thought Tesla were going to project onto the screen... especially when a nav route is plotted. (Based on some of the hires)
Alibaba-backed augmented reality start-up makes driving look like a video game
Alibaba-backed augmented reality start-up makes driving look like a video game
Yes, I'll try to do my best
Think of handwriting or optical character recognition (OCR) as a three-dimensional problem that is made out of a two-dimensional drawing pad with a height and width of 32 pixel. For the sake of it, let's say very pixel holds information of three bits:
That makes a 32 * 32 * 6 bits three-dimensional array we could now feed into a CNN.
- 000: Pixel is completely black (not written on)
- 001: Pixel is slightly grey-ish (not directly written on, minimal "bleeding" from neighbor pixel)
- 010: Pixel is a bit more grey-ish (not directly written on, a bit more "bleeding" from neighbor pixel)
- etc.
- 111: Pixel is completely white (directly written on)
But if we now want to factor in / t, we could just use a three-dimensional array for every frame (32 * 32 * 6 * t), with the result of pumping the newly created array into the fourth dimension. This four-dimensional array can then be fed into the CNN just like any other, one-, two- or three-dimensional array, hence calling this approach "naïve"
This is pretty common technique for 2D data + time (eg. stock charts) but you can theoretically add extra dimensions to datasets of any given dimension as described above.
If Kip says the math gets too hairy, you better believe him, haha.
Referring to my example above, I'd like you think of dimensions in the mathematical sense: Time is just a bit of (extra) information that isn't special in any way
EDIT: This is my last post on this topic. At least in this thread. Sorry for de-railing the discussion.
Anstandswauwau,
Thanks for the post! This is a general discussion thread, image recognition is related to autonomous driving, I guess it's not totally irrelevant. As least I learned something from your post.
For long term Tesla investment, the progress of autonomous driving is the most important piece of puzzle, alien dreadnought the second.
@Anstandswauwau
Not irrelevant at all. We need to understand the progress and implementation (as well of the theories) of NNs for autonomous driving. Figured those of us (including me) who have scant knowledge/understanding could learn something from the other threads to understand some of what you are saying. Would love to see your insight on those threads as well.
Not irrelevant at all. We need to understand the progress and implementation (as well of the theories) of NNs for autonomous driving. Figured those of us (including me) who have scant knowledge/understanding could learn something from the other threads to understand some of what you are saying. Would love to see your insight on those threads as well.
That's also a reason why we won't have FSD all at once. The progression will be so incremental that you won't really notice it.
It will be step by step, like it's already happening. There will certainly be some " big " leaps, like when the Autopilot will start to recognize and act on vertical signs for example. Or when Autopilot will be able to work well in downtown cities ...
But, even when the Tesla will be able to fully "FSD everywhere", I'm pretty sure Tesla will still require people to be " alert " and put your hands on the steering wheel every once in a while like it's doing now.
Then the time between each time where you have to show the car you're alert will be bigger and bigger (every 2 minutes, then 5 minutes, then 10 minutes, then 20 minutes, then 30 minutes ...) , until you don't anymore. And that's when FSD will be achieved. But because it will have been so incremental you won't really have noticed it.
- Status
Similar threads
- Sticky
