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

[TN Mtn Man]

What you said here is simply additional opinion which I share. I think it's highly likely she will invalidate the second vote. It falls in the category of "rule the way she really wants".

Yep. More activism from the bench. She apparently feels that her opinion matters more than that of the shareholders. Why are so many judges so corrupt?

On a bright note-TSLA has broken the $200 barrier-lets hope the trajectory continues and we see some good delivery numbers.

 

[TN Mtn Man]

Anyone else watch Ark Invest talking about Robotaxi valuations for Tesla over the next five years?

I generally agree with their outlook in the long term, but I feel their timeline is way too optimistic. Its going to take time for the RT fleet to be built, deploy, and take hold. Not to mention FSD needs to be solved yet for RT's to even work. Ark seems to hinge a lot of their valuation on a large percentage of current Tesla owners instantly joining the RT fleet in the next year or two, and I think that is highly unlikely. I still feel a minority of Tesla owners will even put their cars into the RT fleet. I also think RT profits will be smaller and with lower margins than Ark predicts.

$2600 / share is something I could see happening someday for TSLA with both RTs and Optimus being sold, but I think that's like 10-20 years out, not 5 years from now.

I hope they are right here. But given their performance the last 3 years, I am doubtful. My ARKK is down nearly 2/3 even now.

 

[EQC_]

She will throw out the second shareholder vote just like she did the first. She listed several problems with the original 2018 comp plan. The revote fixed one. The rest remain including that the comp plan was negotiated by Elon with himself. The directors weren’t independent and didn’t act on behalf of shareholders. Also the Judge doesn’t care about rumors or boomer mommys letter writing campaign.

First, I'm all for pessimism...so I've got no problem planning/expecting the worst...and hopefully whatever happens is at least some small surprise to the upside. There's no harm in expecting or predicting a terrible outcome from the courts...it's not like our predictions here will have any real effect on the outcome anyway.

Next: Really, any predictions we can make, whether pessimistic or optimistic, or whether based on a hunch or based on legal experience, is still just a prediction...and who knows what the judge or court or lawyers might pull next, or if there will be further appeals or whatever after that.

That said: I may be wrong, but my memory is that a key "problem" from the judge wasn't exactly that the director's didn't try to negotiate down...but that supposedly Shareholders at the time of the vote didn't know how cozy Elon and the Board were, so the shareholders didn't know that the board probably didn't negotiate. And I thought this status of the shareholders/voters being uninformed was a big part of the Judge's justification for negating their votes.

Again, I may be wrong and I have zero legal anything...but my impression was that, since the re-vote included all of the judge's documents and statements as part of the package (plus all the publicity about this case), now shareholders have been informed of those issues. So, with the proper information in place, it feels like it *should* be harder to justify negating the re-vote of the now-informed shareholders.

*Yeah..."Should" be harder this time...but as others have said, the judge can and will make whatever decision she wants to make.

 

[Knightshade]

That said: I may be wrong, but my memory is that a key "problem" from the judge wasn't exactly that the director's didn't try to negotiate down...but that supposedly Shareholders at the time of the vote didn't know how cozy Elon and the Board were, so the shareholders didn't know that the board probably didn't negotiate. And I thought this status of the shareholders/voters being uninformed was a big part of the Judge's justification for negating their votes.

Correct-- it was not the only reason for the verdict, but it was a very important one. Teslas arguments regarding the revote will be that negating that aspect of the original decision means there's not enough other reasons left, legally, to justify wholesale tossing of the comp package.

Again, I may be wrong and I have zero legal anything...but my impression was that, since the re-vote included all of the judge's documents and statements as part of the package (plus all the publicity about this case), now shareholders have been informed of those issues. So, with the proper information in place, it feels like it should be harder to negate the re-vote of the now-informed shareholders.

That's exactly the idea- it's not, at all, a guarantee the judge will find that sufficient to toss out the other issues the court had with the comp package, but it's the crux of what their argument will be (and would likely figure central in any appeal should the judge NOT change things based on the vote).

I do think the ideas the lawyers should get nothing in light of the revote is misguided... based on the original verdict the court found shareholders were denied their right to an informed vote. The lawyers got them that. The fact they largely made the same decision in said vote with MORE info doesn't change the nature of the original vote. That doesn't mean it should be any massive amount beyond actual costs though.

 

[jkirkwood001]

Still, with maxpain at $190, there gonna be a ton of downward pressure today. Still green pre-market, let's see how that holds up. Better than Biden in the debate I hope (whoops, couldn't resist).

View attachment 1060502

I know this is the investment forum, yet we can be superstitious about actually talking about what the stock is doing! But,... we're currently $12 over maxpain! I just can't believe it's going to stay flat til 4 PM. I don't know which way it's going to go - we don't, admit it - but there will be a lot of activity before end of day, that's for sure. Of course, as a HODL-er, it's a bit academic to me in the short term.

 

[M|S|M MYP]

Sure hope the interns take over trading soon, isn't it lunch time soon?

 

[GOVA]

Easy for them. From now until end of trading we'll see how easy it is for them to do whatever they want.

 

[Discoducky]

I'm sorry, but Waymo's explanation is useless. It doesn't matter whether the sensors failed or not. In fact, it'd be better if they had. Instead this was a fundamental failure of their entire system design, a violation of the First Commandment: Thou shalt not run into things.

AVs will make errors and crash. That's inevitable. They'll get tricked by confusing construction zones, misinterpret what humans are doing, get bad advice from remote monitors, etc. But this particular failure simply cannot happen.

I say all this as a long-time admirer of Waymo's technology. And I've done battle for a decade against all manner of silly claims from the Muskian horde ("billions of miles", "rides on rails", "only Tesla has neural nets" and more recently "end-to-end"). But PoleGate goes all the way to the core. Waymo must go back to square one and revisit every aspect of their system design. If they really think adding virtual curbs to this alley's map and tweaking a "damage score" is all they need, they're screwed.

Here's Waymo's full PDF response if anyone wants to read it.

Here's my interpretation of what is being said, what isn't being said and reading between the lines (pun intended )

Short version: If a different kind of static object is added to the drivable area, that is NOT in HD Maps and does NOT meet the 4 criteria, the new software will hit it. As crazy as that sounds, that is what I'm reading. This is the fundamental issue with heavily relying on static HD Map data, sensor fusion and heuristic planning/controls.

Spoiler: Long Version

1. Reliance of HD Maps and the static obstacle features is higher than even, I think, many have expected. The beginning of their drivable surface seems to be whatever HD Maps says it is. As opposed to the real-time sensors providing a "fused" interpretation of the static objects and thus seem to provide no "ground truth" delta. What this tells me is that the real-time sensors may have actually not 'seen' the pole at all as the reliance on HD Maps is so fundamental to their solution that they have effectively tuned out static objects that have been pre-computed/inferred to be somehow ok to hit. Therefore, fundamentally, any change to the static driving surface is going to cause an issue for the software. So, in order to avoid the pole, the new software is *ONLY* updated for a pole or pole-like static object that meets the 4 criteria outlined here (my comments italicized):
   • 1) the object was within the boundaries of the road and the map did not include a hard road edge between the object and the drivable surface; (IOWs the object is 'hittable' as it is inside the HD Maps drivable area)
   • 2) the Waymo ADS’s perception system assigned a low damage score to the object; (IOWs there are other things it WILL hit even if it "sees" it, but infers the damage will be lower than some threshold.)
   • 3) the object was located within the Waymo ADS’s intended path (e.g. when executing a pullover near the object); (obviously)
   • 4) there were no other objects near the pole that the ADS would react to and avoid. (this one makes little sense to call out)
2. Obvious issues with this solution
   1. Any object that is in the driveable area that is NOT in HD Maps
      1. Debris in the road
         1. Large rock (happened to me yesterday)
         2. A pile of bricks (Nov 2023)
         3. Mattress (several times)
         4. Tire/wheel (also several times)
         5. Any other item that, if struck, would be hazardous
   2. Non-reliance on road paint for non-drivable surfaces
      1. The pole that was hit was clearly marked as a non-drivable surface, but since this was NOT in the HD Map, it was NOT deemed a non-drivable area
         1. Therefore, new road paint will be discounted it would seem
   3. If there is another non-pole object added to the drivable area like
      1. A support cable that is clearly outlined like the pole was clearly outlined by road paint
      2. A mailbox clearly outlined with road paint

 

[SOULPEDL]

I know this is the investment forum, yet we can be superstitious about actually talking about what the stock is doing! But,... we're currently $12 over maxpain! I just can't believe it's going to stay flat til 4 PM. I don't know which way it's going to go - we don't, admit it - but there will be a lot of activity before end of day, that's for sure. Of course, as a HODL-er, it's a bit academic to me in the short term.

View attachment 1060555

View attachment 1060558

Ya, I'm a bit less academic over here, and more instinctive with a bit of smell and feel mixed in. So I sold my usual 25 above 200 for breakfast, that was pretty hard to do (Edit: because I'm not as certain I will get them back this time).

Today will be a battle day for sure, but I think it should percolate some more in "the base" and then release the spring. It's gonna seem like eternity to get through Q2 results followed by 8/8. Recent P&D acceleration has me pumped but it still could read as a miss against 1 yr ago, very close. So more volatility I see short-term.

Recall discussions with @Artful Dodger (along with my survey for 2024 total P&D). Sentiment then was a barely 1M while some took a more aggressive posture toward 1.2M. There are plenty of reasons that capacity could rise from here even, but boils down to battery supply and not demand ("Show me the incentive, and..."). Margin is another story but all good still. The revenue should come from Packs and FSD. Also, this is the quarter where we should see my prediction of >50% uptake on FSD subscriptions for at least one of the Q2 months (because it is summer and travel is looking strong, free FSD on Apr 1st was taken away from customers - 1.8M of them).

Elon will likely go the highest possible volume route, so I'm still on for 1.2M. But it's gonna have to keep growing steadily and no upset in the macro or gov't. It's a big ask, but it's same bet, no change.

See what I mean by Feels-like Economics? Throw in TMC activity as one of my stronger signals that correlates quite high with Jim Crammer, I'm gonna be an Analyst someday! And self-certified like the rest.

Oh look...

 

[GatorMeat]

In 2020, ARK’s price target for Tesla in 2024 was $1,500 bear case and $7,000 base case, presplit. $100-$466 split adjusted...which is almost very close to the range TSLA has traded from It's $400+ ATH set Nov 2021 (over 2 yesrs early) to it's more recent ~$140 low set Jan 2023. This was all without robotaxis, which I believe would have constituted the bull case. I would have to go back to reread the report to see for certain.

ARK just got lucky on the share price and the analysis to get the price was a dumpster fire and way off. They had huge robotaxi profits and revenue and like 10M car sales by now. They are no better than a college kid project.

 

[GatorMeat]

View attachment 1060537
Guess this is on you then?


View attachment 1060538

But the Aussie played some key minutes in that championship series.

 

[Usain]

Here's Waymo's full PDF response if anyone wants to read it.

Here's my interpretation of what is being said, what isn't being said and reading between the lines (pun intended )

Short version: If a different kind of static object is added to the drivable area, that is NOT in HD Maps and does NOT meet the 4 criteria, the new software will hit it. As crazy as that sounds, that is what I'm reading. This is the fundamental issue with heavily relying on static HD Map data, sensor fusion and heuristic planning/controls.

Spoiler: Long Version

1. Reliance of HD Maps and the static obstacle features is higher than even, I think, many have expected. The beginning of their drivable surface seems to be whatever HD Maps says it is. As opposed to the real-time sensors providing a "fused" interpretation of the static objects and thus seem to provide no "ground truth" delta. What this tells me is that the real-time sensors may have actually not 'seen' the pole at all as the reliance on HD Maps is so fundamental to their solution that they have effectively tuned out static objects that have been pre-computed/inferred to be somehow ok to hit. Therefore, fundamentally, any change to the static driving surface is going to cause an issue for the software. So, in order to avoid the pole, the new software is *ONLY* updated for a pole or pole-like static object that meets the 4 criteria outlined here (my comments italicized):
   • 1) the object was within the boundaries of the road and the map did not include a hard road edge between the object and the drivable surface; (IOWs the object is 'hittable' as it is inside the HD Maps drivable area)
   • 2) the Waymo ADS’s perception system assigned a low damage score to the object; (IOWs there are other things it WILL hit even if it "sees" it, but infers the damage will be lower than some threshold.)
   • 3) the object was located within the Waymo ADS’s intended path (e.g. when executing a pullover near the object); (obviously)
   • 4) there were no other objects near the pole that the ADS would react to and avoid. (this one makes little sense to call out)
2. Obvious issues with this solution
   1. Any object that is in the driveable area that is NOT in HD Maps
      1. Debris in the road
         1. Large rock (happened to me yesterday)
         2. A pile of bricks (Nov 2023)
         3. Mattress (several times)
         4. Tire/wheel (also several times)
         5. Any other item that, if struck, would be hazardous
   2. Non-reliance on road paint for non-drivable surfaces
      1. The pole that was hit was clearly marked as a non-drivable surface, but since this was NOT in the HD Map, it was NOT deemed a non-drivable area
         1. Therefore, new road paint will be discounted it would seem
   3. If there is another non-pole object added to the drivable area like
      1. A support cable that is clearly outlined like the pole was clearly outlined by road paint
      2. A mailbox clearly outlined with road paint

This does lay Waymo bare, doesn't it?

As @Discoducky pointed out, only the first two criteria are really relevant:

• 1) the object was within the boundaries of the road and the map did not include a hard road edge between the object and the drivable surface; (IOWs the object is 'hittable' as it is inside the HD Maps drivable area)
• 2) the Waymo ADS’s perception system assigned a low damage score to the object; (IOWs there are other things it WILL hit even if it "sees" it, but infers the damage will be lower than some threshold.)

It looks like there are two ways for Waymo to fix this problem:
1. Constant remapping.
2. Beef up the perception system so that when it sees an object in the driveable space it generates the correct damage score.

Number 1 is obviously not scalable.

So Waymo is left with figuring out how to build a perception system that can assign a proper damage score to almost anything it might encounter.

Good luck with that.

 

[Doggydogworld]

I would say it's too early to call most of their price targets.

I agree about price targets. But it's not too early to call the operating targets used to build those price targets.

Thanks for summarizing ARK's forecasts over time and including both pre- and post-split numbers. Very helpful. I've bookmarked it for easy reference. I'm going to focus on their 3/19/21 forecast.

$500 Ark Tesla bear case 2025 (March 19, 2021) $1,500 price target (pre-split, would be $500 post split)
Ark Tesla case 2025 (March 19, 2021) $3,000 price target (pre-split, would be $1,000 post split)

Also $1333 ARK Tesla bull case 2025 ($4000 pre-split, $1333 post-split)

Their operating metric estimates were:
Bear - 5m cars / 299b revenue / 43% gross margin
Expected - ?? cars / 507b revenue / ???
Bull - 10m cars / 700b revenue / 50% gross margin

Even the Bear Case numbers are loony tunes. Here's what we're likely to see for 2025:
Realistic - 2m cars / 120b revenue / 18% gross margin

Note that Realistic includes Energy, Supercharging, etc. while ARK only includes EV sales, Insurance and Ride-hail.

 

[cliff harris]

Waymos approach is so clearly fundamentally broken. In theory its incredibly hard to construct a neural net that can drive anywhere, but as luck would have it, we have five million Tesla owners, and all of them have had at least 18 years to train their neural nets, and all of this training data is available for free to Tesla.

Its actually way, way easier for Tesla to achieve FSD than getting optimus useful, because the difference in the quantity of training data is many, many orders of magnitude. As far as I know, FSD beta has not driven into a post, or a bollard or any similar object yet. Given this, it seems there is now a hard clear line between FSDs capabilities and waymo.

The sensible route for waymo is to license FSD from Tesla. That may seem like heresy, but actually its the only way they preserve anything from their business. They already have an app, and brand recognition, and they have a bunch of remotely connected vehicles. They are used to having multiple cameras in those vehicles and collecting the data.
In this regard, waymo are way better placed than VW, Ford, GM or Toyota in terms of FSD. They can license FSD right now and improve their safety.
Of course, this being sensible will not matter one bit because of corporate egos, but in a few years time they are going to wish they had done so when they could still have have salvaged something.

 

[SOULPEDL]

Any object that is in the driveable area that is NOT in HD Maps

1. Debris in the road
   1. Large rock (happened to me yesterday)
   2. A pile of bricks (Nov 2023)
   3. Mattress (several times)
   4. Tire/wheel (also several times)
   5. Any other item that, if struck, would be hazardous

Ha.... or like "a 5M Steel Ball in the road"?

Why would Waymo take this approach - easier on the compute?

This pole strike should have been predictable then. Maybe trying to speed up the vehicle/performance. Desperate move?

My playful side wants to go paint the pole to see what happens next time (I'm always up for an Experiment). Something like a totem pole, with meaning as the guardian of FSD. This is borderline poetic! Sorry, but I am from BC.

The word totem refers to a guardian or ancestral being, usually supernatural, that is revered and respected, but not always worshipped. The significance of the real or mythological animal carved on a totem pole is its identification with the lineage of the head of the household.​

Behold the Great Pole.

Feel the almighty power as it sits perfectly still on the street, staring down anyone who dares to touch it. It is the way of the people together with the City and the Earth. We shall resinate to take back our streets. People will walk and bike unharmed, breathing fresh air once again.

Hail to the Great Pole!

 

[alexxs88]

Not sure why people go on about the quantity of training data. There's basically no proof that more data (beyond a certain amount) equals better results. It's obviously highly dependent on the model, otherwise the number of disengagements per mile would linearly go down as more data is gathered. Which obviously didn't happen. Not to mention that there are regressions on each iteration, meaning training data isn't leveraged to the full extent.

I also think people are celebrating way too early any waymo failure. When / if fsd ever becomes unsupervised, any of the tens of thousands of critical disengagements that occur over the fleet now has the potential of exposing conceptual flaws orders of magnitude worse than Waymo's pole.

 

[MartinAustin]

Ha.... or like "a 5M Steel Ball in the road"?

I think it's time to get MetalBallStudios involved, and render simulations of Waymo cars hitting different sizes of objects

Wait, isn't this thread about Tesla stock? I wonder if the market makers have anything left in their toolbox or they are giving up at $195

 

[Usain]

Not sure why people go on about the quantity of training data. There's basically no proof that more data equals better results.

It's about quantity and quality. But more data does indeed equal better results.

Look at Karpathy's talk below and pay special attention to the slide at about 25:45 (see below set at proper time).

He presents the LLM scaling law which says performance of an LLM is a smooth, well-behaved, predictable function of N, the number of parameters and D, the amount of training data. He also mentions that there is no known limit to this scaling law.

I see no reason that the LLM scaling law would not apply to Tesla's end-to-end FSD.

It's obviously highly dependent on the model, otherwise the number of disengagements per mile would linearly go down as more data is gathered. Which obviously didn't happen. Not to mention that there are regressions on each iteration, meaning training data isn't leveraged to the full extent.

Obvious? We the public don't have good data on disengagements per mile. But Elon does. Elon is claiming 5x to 10x intervention improvement between 12.3 and 12.4. I don't think he's lying. He is basing this on internal metrics.

I also think people are celebrating way too early any waymo failure. When / if fsd ever becomes unsupervised, any of the tens of thousands of critical disengagements that occur over the fleet now has the potential of exposing conceptual flaws orders of magnitude worse than Waymo's pole.

I agree with you here. There is no reason to celebrate any failure. And when (not if) FSD finally goes unsupervised there will be plenty of risk.

But I do think it is fair to say that it looks like Elon has chosen an approach that is superior to Waymo and Tesla has a much better chance at success.

 

[alexxs88]

Altman explicitly said he doesn't expect training data to matter for LLMs beyond a certain level to which we're very close now for example. So him and Karpathy disagree on that. I don't believe there's anyone on this forum that has the required insight or knowledge to contradict any of those two. I'll say that waymo is the only company that's attempting RT which was never resource constrained and which was always able to choose the best solution for their end goal. Autonomous driving. Tesla, in its entire existence, rarely had this luxury. And definitely not when it mattered.

What we know for sure is that there is a vast cohort of people using Waymo fully autonomously with great results right now. Today. With zero perceivable interventions, since it's not realistic to think a human operator could give even a hint of input to a moving car which is approaching a situation without a large degree of confidence.

 

[alexxs88]

But I do think it is fair to say that it looks like Elon has chosen an approach that is superior to Waymo and Tesla has a much better chance at success.

There's nothing in what we're seeing now that would suggest that's the case. Not to mention that we're not even two months into the new "FSD is coming soon timeline" and we're already 6-7 weeks behind, with 12.4 still not released. Even though it was supposed to be coming "next week".

I genuinely can't understand how people are making any sort of projections. We're talking about self driving functionality, not a new theme for an e-commerce website. I'd expect the CEO of the company developing such a system to only say "it might come next week" only if it's been flying through testing. Which was obviously not the case. It looks to me that either the CEO was lying (I'm probably in the minority but I don't believe he was) or the people managing the project are so afraid of getting kicked off that they are telling Musk what he wants to hear. If anyone has a better explanation of why 12.4 was not only indefinitely delayed, but also seems to have been performing much worse than 12.3, I'd love to hear it.