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

[UnknownSoldier]

I want to say this about Keller: He is a hardware design guy, not a software design guy. So I don't put really that much stock into his assertion that FSD is closer than we think over a software designer.

Every time I have to kick the brake pedel to turn Navigate on Autopilot off because it's trying to cause an accident at an off-ramp, I keep in mind that FSD is probably further away than we think in the short term.

In the long term, I look forward to the day when you pretty much only engage manual driving for fun.

 

[dc_h]

Friedmen did not get owned. The interview was great. FSD is not easy. If it was it would be done already and it’s nowhere near done.

really liked much of this guys insights, but like most people, he’s not 100% right on everything.

FSD being easy is a relative concept for Keller. There are harder problems out there. I think FSD was one of those 10x step changes in computing and maybe it is a 100x challenge, which is about where the FSD 3 chip is vs the first AP chip. My optimism has been poorly rewarded so far, but I expect 2020 is going to be a good year for AI driving systems-at least for Tesla.

 

[jerry33]

You don't need to go that far North.

But if you don't, you won't be able to be Hyderized.

 

[ZeApelido]

Here are the reasons I believe we’ll see FSD sooner rather than later:

1) There should be enough compute power available today to not only train networks, but also to evolve novel and suitable networks. This assumes Tesla has the data and I believe they do, though, yes, labeling is likely a serious bottleneck. Mammalian vision is an evolved system; it should be possible to evolve a suitable system with enough compute power.

2) There are decent models of a lot of the functions of the human brain, including vision, memory, sensorimotor control, and reinforcement learning. These models could be used to guide or seed the evolution of the networks. That many people are unaware of these models is no reason to believe that Musk and Tesla are unaware of them — even if they don’t speak about them. I’m partial to the work of Stephen Grossberg (Stephen Grossberg - Wikipedia) and his colleagues at what was once the Cognitive and Neural Systems Department at Boston University (CNS Classes | Department of Cognitive and Neural Systems).

3) Thinking like a futurist, we are overdue for this development as I estimate human technological progress. Read Kurzweil ( Ray Kurzweil - Wikipedia ) as a starting place on estimating technological progress. Remember, experts tend to overestimate what can happen in the short run and underestimate what can happen in the long run. I’m afraid I was a couple decades early on neural networks, so for me this is the long run. I was a few years late on coming to mobile and iOS. I like to think I’m getting more accurate at forecasting and that I’m bang on with Tesla.

Since Musk appears satisfied with their sensor suite and now aims for people to be able at least to commute to work without a single intervention, I take this to mean FSD is working quite well though still needing (relatively few ) corrections. Thus, I think Tesla is actually far along a viable path, but not quite there.

They are now working, in my guesstimation, to slay a couple more 9’s in their FSD March of Nines. Perhaps it’s still just a way of saying "Elon said so," but if anyone can fight exponentials with exponentials it’s Tesla (referring for example to the three orders of magnitude improvement in labeling mentioned by Elon in the most recent earnings call).

Very interesting stuff, but none of this really supports full FSD (Robotaxi level) being able to operate on FSD.

Autopilot on City street will work on HW3, but full FSD might require a few more iterations of hardware.

 

[jerry33]

Very interesting stuff, but none of this really supports full FSD (Robotaxi level) being able to operate on FSD.

Autopilot on City street will work on HW3, but full FSD might require a few more iterations of hardware.

But no one will know until we see the actual code running. AP != FSD

 

[Artful Dodger]

That’s me, and the “Northern Lights” are on my bucket list.

P.S. Elon, if you’re listening, I would love superchargers along the Alaskan Highway.

Elon announced during the Q&A session at the June 2019 AGM that Tesla will expand the Supercharger Network into Alaska. There's video on Youtube if you want to watch it.

Cheers!

 

[ggr]

Moderator extended normative note:
1. I just took half an hour to go back 13 pages and move the discussion about programming languages to: Programming Language Wars: Python/C/C++ et al. .
This is not great fun to do. But it's clearly off-topic and we have already seen this subject explode before, leaving 2/3rds of the audience grumbling. No more posts in this thread on that subject will be tolerated. See Figure 1.

2. Comparisons with Taycan have been declared off-topic for ages. It was reasonable to post the existence of a new comparison article but no followups here. See Figure 1.

3. Suggestion: When you want to comment on something that you know will be off-topic, check to see if a thread has already been created for it. (You can open TMC in another tab at the same time, with minimal interference in your browsing.) If not, you can add the post to your quote queue, create a new thread with a reasonable title, and add in the quote referring to the original. You get fame and fortune as the creator of the new thread! If you don't do this, see Figure 1.

Reminders:
1. It is very difficult for moderators to handle multiple off-topic subjects at once.
2. It is relatively easy for moderators to delete bulk posts, even on different subjects.

Figure 1:
Moderator actions for off-topic posts:
First post: nuke the post.
Second post by same author: nuke the repeat offender.

THIS is the Line in the Sand!
--ggr

 

[johnm6875]

Lol, luv it! Factory could be very fast to build and get into production:

• no huge Stamping press with long-lead/expensive dies
  • just waterjet the pattern on flat sheet steel
    
  • then fold 'em with an automated metal brake
  • down the line she goes like punching movie tickets
• no Body line
  • just clamp and MIG weld
  • robot vision assisted AIs do the welding
• no Paint shop
  • color SS with a varying thicknees chrome oxide layer to get:
    • blue
    • bronze
    • green
    • gold
    • purple

All that's left is final assembly, and at GF5/Houston that would be old hat for Tesla by then.

Oh, how we wants a desert camo Cybertruck... (Mordor trucker Sméagol)

Cheers!

It appears the interior structure of the door frame is stamped. Do you see these being shipped from another factory?

 

[StealthP3D]

Hi, you kept mentioning the mispriced option ,
but what is that? Call or put? What do you base one?

That was a reference to the cheap calls they were selling last year when it was more than obvious Tesla had turned the corner. This caused an outsized demand for the calls relative to the puts so they had an imbalance that needed to be corrected by buying more Tesla shares which turned into a positive feedback loop since it happened at the same time that many other investors who were slower than most of us here started buying/increasing long positions also.

 

[StealthP3D]

Moderator extended normative note:
1. I just took half an hour to go back 13 pages and move the discussion about programming languages to: Programming Language Wars: Python/C/C++ et al. .
This is not great fun to do. But it's clearly off-topic and we have already seen this subject explode before, leaving 2/3rds of the audience grumbling. No more posts in this thread on that subject will be tolerated. See Figure 1.

2. Comparisons with Taycan have been declared off-topic for ages. It was reasonable to post the existence of a new comparison article but no followups here. See Figure 1.

3. Suggestion: When you want to comment on something that you know will be off-topic, check to see if a thread has already been created for it. (You can open TMC in another tab at the same time, with minimal interference in your browsing.) If not, you can add the post to your quote queue, create a new thread with a reasonable title, and add in the quote referring to the original. You get fame and fortune as the creator of the new thread! If you don't do this, see Figure 1.

Reminders:
1. It is very difficult for moderators to handle multiple off-topic subjects at once.
2. It is relatively easy for moderators to delete bulk posts, even on different subjects.

Figure 1:
Moderator actions for off-topic posts:
First post: nuke the post.
Second post by same author: nuke the repeat offender.

THIS is the Line in the Sand!
--ggr

In all the excitement we've had in the last several weeks with the appreciating share price and the exploding number of posts, I think one thing that has gone under-appreciated is the fine job the moderators here do for the good of all.

So I just wanted to shout out a big "Thank You" for doing a job that is often thankless! We know it takes a lot of thought, finesse and time.

 

[ggr]

In all the excitement we've had in the last several weeks with the appreciating share price and the exploding number of posts, I think one thing that has gone under-appreciated is the fine job the moderators here do for the good of all.

So I just wanted to shout out a big "Thank You" for doing a job that is often thankless! We know it takes a lot of thought, finesse and time.

Ha! You get more likes for this than I did... ;-)

 

[Jackl1956]

Elon announced during the Q&A session at the June 2019 AGM that Tesla will expand the Supercharger Network into Alaska. There's video on Youtube if you want to watch it.

Cheers!

I’d like to make the drive from Seattle north on Alaskan Highways through British Columbia into Alaska. A beautiful drive, nothing short of adventure. To do it in my Tesla would be epic. If I could it in a new Cybertruck...I would die a happy man.

 

[Words of HABIT]

@Words of HABIT any particular reason for disagreement with this?

Definitely will see what sort of numbers I can pull out of the detailed certificates that @bhtooefr has pointed out to see if I can track the efficiencies on these other cycles.

But the fact remains that the raw HWFET is a very decent (optimistic) predictor of best-case Highway range. And comparing that result to Model 3 should be safe. And it clearly (no surprise!) is a lot worse than Model 3.

@SageBrush yes I think that squared coefficient isn’t bad as a model of relative aero loads. When calculating to get predicted Wh/mi impact, though, you have to make some assumptions about drivetrain efficiency; it can’t be used directly. So you can’t necessarily use it to compare the efficiency of two different vehicles at highway speed, if the drivetrains are different. I’ve never tried to piece this all together though. I just have my own empirical physical model, which is good enough for who it is for.

AlanSubie4Life, I disagreed based on your reference to "a lot worse". MY LR is 112 MPGe Hwy, M3 LR is 123 MPGe Hwy, agree not as efficient, but still comparable (within 10%) for a larger, taller, higher stance and heavier vehicle. Sorry, but I can't yet fathom the word "worse" and mention of a Tesla vehicle in the same sentence. Does not compute.

 

[ABCTG]

Yeah, I too think that there was a slight miscommunication about FSD policy:

Spoiler: Lex Fridman's mental image about 'drivable free space'

View attachment 509318

Spoiler: Jim Keller's mental image about 'drivable free space'

View attachment 509329

LOL, and then there's Naples Italy where for some reason they install traffic lights, and then they end up just turning them off...

 

[Artful Dodger]

It appears the interior structure of the door frame is stamped. Do you see these being shipped from another factory?

Sandy Munro has some comments about that in his recent Autoline Afterhours appearance. IIRC he said they'd parts would likely be stamped and painted at the suppliers, but of course he's speculating based on his take on how Tesla is likely to build the CT. And that method changes depending on whether volume will be 60K or 500K per year.

Sandy Munro Thinks Cybertruck Will Be A Cash Cow For Tesla

At any rate, smaller stamped parts don't need the huge Schuler press that Tesla installed for the main Model body parts. A smaller press will work fine, and is easier to duplicate when scaling production.

Cheers!

 

[pnungesser]

Hell, I'd just like to be able to cross central US east-west using I-40. Easy fix with a single SC near Ft. Smith Arkansas.

 

[Artful Dodger]

That was a reference to the cheap calls they were selling last year when it was more than obvious Tesla had turned the corner. This caused an outsized demand for the calls relative to the puts so they had an imbalance that needed to be corrected by buying more Tesla shares which turned into a positive feedback loop since it happened at the same time that many other investors who were slower than most of us here started buying/increasing long positions also.

Thanks for helping us understand what happened last week, and the machinations behind the curtain.

But, what do we can this event? It wasn't really a short squeeze: MMs were shorting like demons throughout.

It's not a liquidity crunch: there were PLENTY of shares for sale, mostly sold short.

So, what do we call it when this MMs-options-delta-hedging-share-purchase-exponential-feedback-loop occurs next time?

Cheers!

 

[Maitri982]

FSD being easy is a relative concept for Keller. There are harder problems out there. I think FSD was one of those 10x step changes in computing and maybe it is a 100x challenge, which is about where the FSD 3 chip is vs the first AP chip. My optimism has been poorly rewarded so far, but I expect 2020 is going to be a good year for AI driving systems-at least for Tesla.

Would love for that to be true. But I know a thing or two about this field and piecing together several items we are hearing and I think we are still a ways off. By that I mean probably 2022ish. But this year could bring some major advances, just not FSD itself.

 

[bhtooefr]

The trick with that kind of area is that it's all low speed.

I suspect you might have to add some sort of logic (this may mean training the neural net) to recognize that you're entering such an intersection, and upon doing so, impose a lower speed limit on yourself.

Then, follow the rules for an uncontrolled intersection in Vienna Convention signatories - yield to traffic coming from your right - and avoid collisions along your desired path.

 

[Curt Renz]

Solving The Money Problem - yesterday: