Tesla, TSLA & the Investment World: the Perpetual Investors' Roundtable

[neroden]

Lets assume the probability distribution of edge cases causing crashes in geometric.

I believe it's hypergeometric

 

[Artful Dodger]

Unfortunately, in the near term I think this means that Tesla continues downward in the wedge pattern it has been stuck in...

If there is nothing to break it out of it, you could see $220 by June-July.

Well, support held in the 'downward wedge' pattern at about $232. So now if we close over $265 on Mon, May 6 it looks like we finally break out of that falling wedge pattern.

Wedge cuts both ways, wot?

Cheers!

 

[lklundin]

Thanks for this clear analysis. It's pretty clear to me that as you go out the long tail of the probability distribution you have more individual corner cases each of which is less likely, so progress slows down as you get further out.

Agreed, on the fair assumption that Tesla needs a fixed, minimum amount of time for each new solution to a corner case. (This time would include: Identifying the case as relevant/gathering data for it, training, validation w. shadow mode but also incl. regression testing against all the previously trained cases(*) and actual deployment).

With Tesla having solved the object and perception detection with their current sensor suite + FSD computer, I feel convinced that their described method for training the FSD will actually bring about a system that over time can become arbitrarily better than a human driver.

I foresee the major hurdles to be regulatory approval combined with a literal explosion of sensational media attention when a self-driving Tesla gets involved in an accident fatal for someone not in the Tesla.

(*) During the Investor Autonomy Day Karpathy talked about creating a 'unit test' for each new case that the NN is trained for. These are the tests that each newly trained version of the NN would need to be able to still pass.

 

[neroden]

Agreed, on the fair assumption that Tesla needs a fixed, minimum amount of time for each new solution to a corner case. (This time would include: Identifying the case as relevant/gathering data for it, training, validation w. shadow mode but also incl. regression testing against all the previously trained cases(*) and actual deployment).

With Tesla having solved the object and perception detection with their current sensor suite + FSD computer, I feel convinced that their described method for training the FSD will actually bring about a system that over time can become arbitrarily better than a human driver.

I foresee the major hurdles to be regulatory approval combined with a literal explosion of sensational media attention when a self-driving Tesla gets involved in an accident fatal for someone not in the Tesla.

(*) During the Investor Autonomy Day Karpathy talked about creating a 'unit test' for each new case that the NN is trained for. These are the tests that each newly trained version of the NN would need to be able to still pass.

I agree with you on all of this, in fact... I simply conclude that "over time" is a really long time. Robotaxis in 20 years? Sure. In 10 years? Maybe. In 2 or 3 years? No way in hell.

 

[Tslynk67]

I think Semi will be bigger than Model 3. I don’t know why this is rarely discussed

Of course it will be bigger, it has to haul huge trailers and sh1t

 

[Zhelko Dimic]

From FCA call:
----------------------------
And as mentioned by Mike, we're talking about or compliance strategy. We did enter into various agreements in the quarter to ensure that we have access to regulatory credits to complement our vehicle launch strategy towards meeting emissions compliance in EMEA and NAFTA going forward. So the total commitment under those contracts is about EUR 1.8 billion, which will be spent over the next 3 years. Last year, we had cash outlays between credits and compliance payments of about EUR 600 million included in our cash flow. We expect 2019 number to be moderately up from that. And we think it's important that we have managed to secure these credits, which we believe to be a very economic way of complementing our compliance strategy through the launch of the electric vehicles that Mike mentioned.

 

[Tslynk67]

yes but the article i was directed to by FT was about fca earnings and didn’t quote anything about them paying tesla, or what amount.

what am i missing? where is the source of this info?

thanks in advance

 

[oneday]

Roughly 3 million shares at $243 on or about May 7 will have been sold. Anybody have any data or guesses as to how this will effect the SP beginning of next week?

 

[lklundin]

Easy as in more well known with definite timelines.

BTW, I've to call out Tesla on the whole power usage stuff. There is nothing sacrosanct about either 100 watts or why 500 is too much. Tesla had a design limitation of 100 watts because they want to be backward compatible. 500 watts isn't a big deal in a car that may use 10 kWh or more per hour even in urban areas.

So, if we ignore that power usage, I'm not sure Tesla has a 4 year lead over NVDA. Ofcourse, for an OEM o build a car with that NVDA chip and needed sensors is a different story.

No, it was not just for backwards compatibility that Tesla imposed the 100 W power design limitation, there is the overhead of removing that heat and everything Tesla does is with a view to efficiency.

Maybe it's something for the next FSD computer ("dojo'), but it should be possible for the power consumption of the FSD computer to be dependent on especially on the car's speed, and secondarily on the (possible) differential speed of other actors. If everything is happening at a low speed, then there is less information to process, at a lower power.

On that note:

Every car has a horn that the driver can use to alert other actors of a dangerous situation.

Have I understood it correctly that the FSD computer can currently not activate the horn?

If the FSD computer sees that a different vehicle is veering into its path, it would make sense to me that the FSD computer would activate the horn, increasing the chance that the other driver will take corrective action.

A human would do this, so I think the FSD computer should as well. This could be a deciding factor in a scenario that does not allow for the Tesla to move out of the way. Imagine f.ex. to be arriving at backed up traffic in a tunnel, your car stops but the vehicle behind you does not slow down. There using the horn would make sense (the car could flash its braking lights as well).

So to tie into the above, also when the Tesla is not actually moving, its FSD computer could still be active and able to react to other actors, but it should need less power to do so.

 

[mongo]

Maybe it's something for the next FSD computer ("dojo'),

My understanding is that Dojo is the system for training the NN (potentially from video), not the HW4 processor.
(and I wonder if it will also be used for training NN for robots on the manufacturing line. Perhaps by watching videos of humans performing tasks?)

 

[lklundin]

If a million cars can't see such cases in enough frequency to train the model, how likely these situations be encountered by a human driver in a lifetime?

No, clearly there have to be sufficient data for a given case to be relevant.

The question is how to perform the triage on the relevant cases.

Wanting the FSD computer to become as safe as possible implies that it should do so as quickly as possible, so the relevant cases need to be prioritized correctly.

 

[kbM3]

1&2 were always part of the plan
3 and other price reductions have also always been part of the plan to get more EVs into people’s hands
4 I believe was more ‘goodwill’ than anything
5 also part of the plan
6 goodwill - how quickly we forget that people asked for Tesla to throw them a bone for actively selling cars on Tesla’s behalf
7 goodwill

SR + worldwide deliveries + leasing were all part of the plan to get to 10-15k /week.

You don’t find it discouraging that they’ve also reduced prices and added other incentives at 6k /week?

I don’t buy the good will argument. Not with cash being so precarious. That would be irresponsible timing, wouldn’t it?

 

[ZachF]

Hopefully FCA just bought Tesla a European Gigafactory.

 

[Artful Dodger]

OT
I think I have more $TSLAQ people that block me than I have actual followers.

Interesting thing, I have never followed one $TSLAQ member and have only engaged them a few times. I do follow a lot of the “FUD fighters” and maybe that is my link to being blocked. Not exactly sure but good riddance.

Yup, the shortz maintain and subscribe to a shared 'block' list. Pick one of the other big shorts that also follow BS. See if you're blocked after the 1st visit. Jolly jokers, fighting for a dirtier, sicker planet with all their venom. No matter, they'll lose. But they'll still have to look in the mirror.

Cheers!

 

[lascavarian]

Have I understood it correctly that the FSD computer can currently not activate the horn?

Is it even necessary to click on this link?

 

[Buckminster]

anthonyj - some remember:

Tesla Semi will feature in all new Mad Max movies and go down in history as the greatest ever product - bigger than the iPhone.

The shorts may have won the past 10 battles but they will not win the war. Because of this war, we have lost much of our mojo regarding some longer term developments. We are all very focussed on the short term. To counter this, I give you:
View attachment 371475
This is the Tesla Semi. Does anyone remember it? It is gonna be huge - literally and figuratively.

 

[Artful Dodger]

(and I wonder if it will also be used for training NN for robots on the manufacturing line. Perhaps by watching videos of humans performing tasks?)

Exactly! Elon hinted that Tesla is working on this during the Q&A session on 'Autonomy Day' when he said that the automated snake charger is doable now with robotic vision.

Not a big difference between inserting a charging cable in a randomly parked car to doing repetitive assembly line work.

Cheers!

 

[Boomer19]

thanks! i kept getting redirected to another article about earnings, or the paywall

 

[lklundin]

I thought the same (vicki s). And I’m also hoping every email sent to employees about leaks was slightly different so they’ll know who leaked this one to the media.

Agreed.
Canary trap - Wikipedia

 

[lklundin]

As far as I can tell, it’s all self-inflicted. These targets have been discussed for a long time and no one started EV work when they should have (e.g., five years ago). Also, I think a lot of manufacturers thought they would get a lot of CO2 reduction from more diesels. We know how that worked out.

In 2015 Elon Musk went to Berlin and literally begged German Economics Minister Sigmar Gabriel to incite the German car makers to start making BEVs.

Clearly, the car makers cared more about their certain but short-term ICEv profits.

I think this experience has changed Elon Musk.

Back then he probably thought the traditional car makers could be made to follow him.

Now he is just going to go it alone, the incumbents be damned. And they are.

Here is a source for the 2015 meeting:

Outline - Read & annotate without distractions