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

[Drumheller]

Do we have a pool going as to what we will see on FSD day? What is the plausible range of events?

Worst case - basically just showing AP and advanced summon
Best case - L3 self driving through a closed course with non-stationary objects? (say they have people walking around in front of the car)?


I never said the kettle wasn't black. Of course, that very saying supports that as well. Both are black.


OT, that's not a lie. Alt right people latched onto Pepe, especially after the Clinton stuff. If you don't understand how the meaning of memes can change as various groups co-opt them I don't think you understand memes at all. Even the creator of Pepe talked about how he didn't like that it became a symbol for the alt-right/Trump/White supremacists etc. The world is a complicated place.

The meaning of words, phrases and symbols is entirely determined by what the consensus within the society in which they exist says they mean. In current society, the clown meme apparently indicated the user self-identifying as alt-right. Whether that’s been propagated by the media, an evil dictator or mind controlling aliens doesn’t really matter. Rail against that as you will, said meme will only mean something different when the rest of society deems it to.

If it makes you feel any better, this only goes as far as language. Language is arbitrary but objective reality does, of course, still exist outside it.

You're both missing the point. It ISN'T the majority of people saying the memes mean *this*. It's the media. The majority of people that use memes understand what they actually mean and understand that a tiny minority using them for another purpose does not change the meaning. That's the media.

There are millions of people passing memes every day to friends. But, then again, thanks to the media, people outside of the USA actually believe there are millions of alt-right people. LOL.

 

[Skryll]

OT:

The meaning of words, phrases and symbols is entirely determined by what the consensus within the society in which they exist says they mean. In current society, the clown meme apparently indicated the user self-identifying as alt-right.

The real interesting aspect to me is what he said earlier: 'The clown meme represents, "This world has become a joke". People use it to express that viewpoint.'. Which is basically something that has been 'channelled' funny enough from the 4chan community into an alt-right movement.

The 'I dont give a *sugar* about anybodies values, its all just hilarious to me' troll crowd that is too afraid to have a real binding dialog or face consequences for their and others actions, probably out of helplessness. I had a friend like that which I lost to the Trump movement, that fits squarely into that clown meme 4chan crowd attitude.

So maybe it is a sign of waking up and realizing you have been duped and channeled when you say you don't want to be identified with alt-right.

Hope that happens to my friend at some point, too.

Best of luck to you all.

 

[EVNow]

The way that Tesla defines advanced summon means that I cannot use it to summon my car in the work parking lot to pick me up at the door. So when Musk says "feature complete by end of 2019" what I hear is "meets some subset of features not currently defined" rather than "full autonomy."

One thing Tesla should do on Monday is to specifically say what their "feature complete" includes and examples of what it doesn't. So, we (and market & media) know exactly what Tesla is promising. Otherwise I expect when they deliver what they had planned to deliver, media will still be saying they didn't deliver what they promised.

 

[EVNow]

You're both missing the point. It ISN'T the majority of people saying the memes mean *this*. It's the media.

Prove it.

 

[bdy0627]

You're getting a lot of grief for this comment, but I totally agree. Elon's talking like FSD is almost ready except for some red tape and additional validation, but my Model 3 still can't autopilot down the recently repaved and extremely well-marked (but curvy) Highway 68 by my house without crossing over the yellow lines. I really can't think of a more ideal test for AP (bright daylight, perfectly black pavement, wide road, bright white and bright yellow markings), but the car literally can't stay in its lane, even at 10-15MPH below the posted speed limit of 40MPH. Now it has been getting better - it would go way further over the yellow line 3 months ago and even more 6 months ago - but it doesn't inspire confidence that FSD is right around the corner if the car can't handle these ideal conditions. And that's just one example. I have dozens more where AP has performed poorly where you'd expect it to have no problem.

So while I agree that Elon has eventually done all the big things he's said he's going to do (and many of those things are world-changingly amazing), I'm skeptical about FSD being anywhere near releasable, even with the 3.0 neural nets. If the fundamentals aren't 100% rock-solid under ideal conditions, how's it going to do with all the poorly marked roads and millions of other corner cases it would need to deal with to be FSD?

I want to believe, but the 0.0002% of Tesla's "data" that I've experienced is not inspiring confidence.

That's really poor. My experience has been much better in terms of lane keeping. It hasn't crossed a yellow line in a long time. Maybe the FSD team isn't getting enough feedback from the vehicles like yours that are not doing well.

 

[SOULPEDL]

I have been reading these jiucy links in posts and need to eat soon - addicted! OMG, my wife told me "step away from the phone" just now.
Reddit, Waymu user, Monday's prediction article... very rich and exciting!

 

[Drumheller]

OT:

The real interesting aspect to me is what he said earlier: 'The clown meme represents, "This world has become a joke". People use it to express that viewpoint.'. Which is basically something that has been 'channelled' funny enough from the 4chan community into an alt-right movement.

The 'I dont give a *sugar* about anybodies values, its all just hilarious to me' troll crowd that is too afraid to have a real binding dialog or face consequences for their and others actions, probably out of helplessness. I had a friend like that which I lost to the Trump movement, that fits squarely into that clown meme 4chan crowd attitude.

So maybe it is a sign of waking up and realizing you have been duped and channeled when you say you don't want to be identified with alt-right.

Hope that happens to my friend at some point, too.

Best of luck to you all.

The memes have nothing to do with Trump. Not everything is about Trump.

Memes are started because they're funny. They morph over time. Like anything that is popular, fringe groups adopt them in an attempt to be relevant.

It's difficult to have a real dialog because most people have already formed their opinions due to media influence, not due to factual and logical analysis. Think about the frustration people in this thread feel when trying to counter FUD. That same frustration applies to this situation with memes. Just because the media says it, doesn't make it true. When the media publishes FUD about Tesla, people in this thread are quick to point it out and take it apart. However, if the media discusses anything other than Tesla, then they must be telling the truth and anyone objecting to it is wrong?

 

[juanmedina]

What level of automation FSD is supposed to be?

I purchased FSD and I think Level 5 autonomy is far off but I can see level 3 and 4 being around the corner. I would be really happy with Level 3 in the short term. The current AP and FSD which is what I have already impresses the hell out me. I just drove 450 miles in NOA with zero disengagements.

 

[TradingInvest]

Interesting insight on the state of Autonomy from the Founder of Creative Destruction Lab and AI expert : teslamotors

I work for one of the major customer experience companies (Enterprise Software) in the world and just got back from our annual customer conference in Las Vegas. We have a lot of AI and machine learning in our product portfolio and have everything from voice analytics that can detect sentiment (along with key words and phrases) for marketing/training/prediction of behavior etc (for contact centers or otherwise discerning the voice of the customer), to real time authentication, financial transaction analysis (think financial crime and compliance) and robotic process automation (attended and un-attended software robots with real time virtual assistants) and other things which also leverage machine learning. Most of the breakout sessions revolved around customers deep diving into the use cases and/ or technology to assist their businesses, but one session in particular caught my eye - "our automated future - learning to adapt and thrive in the era of intelligent machines" by Dr. Ajay Agrawal. He opened by describing the history of the University of Toronto (I had no idea) as one (if not the) pre-eminent centers for AI research (especially for neural networks and deep learning) and listed off a "whos who" of researchers who have spun off into positions at major companies around the world. Of course when he mentioned Andrej Karpathy, my ears pricked up. Very interesting lecture. Couple of things that he said that after hearing I was like - oh of course. First - whoever gets there first, in terms of application of AI to a vertical market (like autonomous driving) wins. Like Google did with search, they were the first to really get it right - and now because they have SO many users reinforcing and providing data to refine and continue to build they keep getting better and better. As he described it, AI is simply a prediction machine, and when the software makes a prediction, the software then makes an analysis of risk of error (what happens if I make the wrong/right decision, what happens if I don't make a decision), makes a prediction and then gets the result which provides a feedback loop to improve the model. Obviously this is vastly oversimplified, but in his example with google, you type in a query - the software predicts what you want to see, provides the results and you click on the one you want. Based on what you clicked on, it gets feedback as to how well it predicted and so on. The pure brute force approach works well on simple models, but when things get very sophisticated (like with driving) its limited by data. That made sense to me as well. He referenced Tesla and Waymo a few times in the lecture and used FSD as a use case example a lot and he knows Karpathy (as well as a number of the major players in the AI/NN research and commercial fields). One of the KEY things he brought forward when it comes to the development of NN and AI, in all applications, but especially the most difficult applications is the value of the DATA - which he said has been described as "the new oil" - super valuable, but must be refined into the things we really want (gasoline, plastic, chemicals) which then drive profitable activity. Neural Networks (NN) are fairly well understood - and are simply refining that data into a product which can be commercialized. I'd never heard it described that way. The other key thing he brought forward was that much like computational advances can be described as reducing the price of arithmetic to zero (or the internet driving the cost of searching and communicating to zero), AI can be described as the reduction of the cost of prediction (to zero), and as the cost of prediction is driven down, its the DATA which is injected into the prediction machines that has the real value as a complimentary product (in economic terms) which will increase significantly if not exponentially.


The light bulb went on in my head.


So after the lecture I had a chance to talk to him for a while - (which was amazing in retrospect - I didn't realize quite how big a deal he was until after and only about 20 people attended his session (there were 3k+ people at our event) and I got the time to talk to him about Tesla and Karpathy's approach. I opened asking about Waymo and Tesla - and how it seemed pretty clear based on the disengagement reports that Waymo appeared to be head and shoulders beyond Tesla, even though Elon seems to be VASTLY confident (over?) about Tesla's capabilities. I asked about how the data is gathered - basically the lidar vs machine vision debate and whether or not that was part of it. His basic take was that he felt like Waymo seems like they are ahead now because they are likely using much more 'hard' code to do things which works well in in the "middle" or more typical use cases but what happens is that they "top out" in terms of performance and defining the corner cases rather than taking the time to let the AI/NN learn on its own. He says he use of simulators to teach the AI works well, but is limited in terms of how much data and the types of use cases that can be simulated in a sophisticated application like driving. This is why they appeared to jump ahead, but have now stalled or slowed down because their data inputs are limited both in use cases and in volume of the data. He believes and agreed that he thinks Tesla's approach will ultimately win, and then own the FSD market simply due to the amount of data that they are able to feed into the neural network and that what they have seen in their AI/NN research is a slow and steady improvement, marked by HUGE jumps forward in progress and then continued refinement and improvement at a fairly predictable rate. Tesla obviously has a HUGE advantage here in terms of the data collected from the cars and the feedback it gets from drivers back into the NN as to whether or not the 'prediction' model was correct or not.

Much like google, now that they have so many users providing data into the feedback loop, they will continue to get better and better to the point of where new competitors just cant enter or get to the point of where Tesla is or will be.

As for the camera vs lidar debate:

As he put it - what he sees with machine vision with cameras is amazing and likely to kill or fully depreciate many areas such as imaging diagnosis (think MRI or Xray scans) and that its pretty clear cameras are the way to go although Lidar has its place for sure and is basically just another method to do the same thing. The difference is that machine vision would deliver the highest amount of flexibility in terms of what can be visualized and discerned, smoke/weather/haze not withstanding. He felt like a hybrid model was likely best and where things would end up at Tesla in time, given the reduction in cost of Lidar over the last few years -- think AP 4.0 hardware. Having both would provide redundancy and the ability to 'see' in compromised environments (dark/smoke/fog etc) that camera only applications have limitations on.


I also asked about why the versions seem to regress in terms of performance - for example in my car the latest version with Nav on AP which is smoother on the highway and does some new amazing things better than ever, now cant take a typical offramp curve without ping ponging all over the place when its been smooth as silk for 4-5 versions. He said that sometimes small adjustments in the model / or the way the NN/AI is learning can have some unexpected results in corner cases which then have to re-adjust themselves over time or via tweaks by the software team. All part of the process. I was pretty encouraged by this - and will be excited to see what Tesla shows us on the FSD investor event coming up.


TLDR: Tesla's approach, while slow and steady will likely deliver major jumps in FSD performance and in time is the most likely approach to deliver the highest performing FSD product, covering the most use cases, and if first to market will likely result in Tesla owning that market due to the sheer amount of data that is collected and used to refine the model. Also machine vision is more than likely more than adequate for FSD as applied today.


Edit:
The part in bold is more or less confirmed in this video:

The long tail is what has to be mastered and that's exactly where TSLA has the advantage.

It's a good explanation. I agree with Dr. Ajay Agrawal except this part: "... Having both would provide redundancy and the ability to 'see' in compromised environments (dark/smoke/fog etc) that camera only applications have limitations on. "

The right approach for those conditions is to slow down to safe speed level. That's also how human handle it, but human don't have radar to see through smoke and fog. So vision + radar should be able to handle dark/smoke/fog better than human do.

Tesla could add LiDAR for redundancy in the future, I think that possibility is very low at this point. Vision has to be solved for FSD in all conditions. Once vision works, it's unnecessary to add LiDAR.

 

[MarcusMaximus]

You're both missing the point. It ISN'T the majority of people saying the memes mean *this*. It's the media. The majority of people that use memes understand what they actually mean and understand that a tiny minority using them for another purpose does not change the meaning. That's the media.

There are millions of people passing memes every day to friends. But, then again, thanks to the media, people outside of the USA actually believe there are millions of alt-right people. LOL.

You are missing the point. It isn’t the majority of people using the memes that decides the meaning. The meaning is whatever the person seeing them believes it to be. If society writ large sees a given meme(or word, or anything else) and believes it means a certain thing, then it means that certain thing, regardless of how they came to believe it means that.

Your choices of how to deal with that are:
1. Continue using it anyway, with most people assuming you mean it in the way they understand it
2. Try to get everybody else to understand it the way you mean it. Note that this is incredibly difficult and unpredictable.
3. Stop using that word/phrase/symbol.

Right now you seem to be going for #2. Good luck with that.

 

[SebastianR]

If you have to wear a lucky shirt to fly, you're going to want a whole lucky outfit for when Starship starts doing Earth-to-Earth hops

On Earth to Earth: how are they going to resolve the weather issue? It feels like every second SpaceX launch is scrubbed for weather reasons. And if you need weather to be ok both places...?

You've described perfectly where the true fraud is. I find this detrimental effect of WSJ's ongoing effort to destroy #TSLA's SP is sickening. Still, it goes under the SEC radar.

Yea, I sometimes think that Elon taunting the shorts/ talking of the short squeeze just ensured enough attention to have markets manipulated to avoid an epic squeeze. I picture this as follows:

Elon: “you better tip toe to the exit!”
Shorts “you want us to do what? Na, we gonna dance on your living room table while shouting profanities! Watch us! Hold my beer...”

If you control available shares and media narratives there will never be a short squeeze.

 

[scaesare]

Hitler liked Wagner, therefore anybody who likes Wagner is Literally Hitler.

This guy always did seem shady:

 

[dhrivnak]

If the in transit value stays constant, then the profit is unaffected. Otherwise, there would be an infinite cumulative effect. Think of it like stocking a new store, you need to buy inventory, which hits your bottom line, but after that your profit is driven by sales - wholesale cost.

And in addition they had a $932 million debt repayment.

 

[MarcusMaximus]

What level of automation FSD is supposed to be?

I purchased FSD and I think Level 5 autonomy is far off but I can see level 3 and 4 being around the corner. I would be really happy with Level 3 in the short term. The current AP and FSD which is what I have already impresses the hell out me. I just drove 450 miles in NOA with zero disengagements.


View attachment 398581

I’d argue level 3(or at least about as close as it’s physically possible to get to level 3) is already there with current NoA. I did a bit of a deep dive into the actual SAE doc defining the levels recently.

In all foreseeable conditions, the system will alert you if it can’t continue the driving task and continue controlling the car for at least several seconds before handing control back. We see this in heavy rain(warns the user and falls back to normal autopilot indefinitely or until rain stops), sharp curves(ditto), and exiting the area where it’s allowed(warns the user with distance left before it will stop and hand control back).

The only deviation is that I believe(though haven’t actually seen evidence of this happening on NoA) in a catastrophic failure(system just has no idea what’s going on for whatever reason), it’ll flash red hands and give control back immediately. The problem with this meaning it’s not level 3 is that it’s physically impossible to meet that criteria. Regardless of how amazing a system you have, there exists some scenario where it cannot maintain control. For an extreme example: some weapon instantly vaporizing the car. Because that’s an impossible standard to meet, I’d consider NoA to be meeting the spirit of the level 3 system definition, if not the exact letter.

 

[JRP3]

The memes are in far more widespread use than just people that claim to be alt-right, but the media isn't reporting on that. I have several on my phone now from friends sending them over the past few days. I have had hundreds of pepe memes over the years.

Is there where you reply, Karen, to say how I'm wrong and I must be alt-right?

Well, if you and your friends send swastika's back and forth, even though they once had different meanings...

 

[Fact Checking]

On Earth to Earth: how are they going to resolve the weather issue? It feels like every second SpaceX launch is scrubbed for weather reasons. And if you need weather to be ok both places...?

It's mostly a non-issue:

• Falcon 9's are sensitive not to "weather", but to "high altitude wind shear", because they are very tall but thin (12 ft wide) rockets with a high "fineness ratio". I.e. they are easy to bend and break.
• StarShip is 9m (30ft) wide, while not much taller - so they won't be sensitive to wind shear like the Falcon 9.

Rockets can actually launch and land in significantly worse weather than commercial airliners, because they have no wings and require no runways.

Orbital rockets derive from ICBMs which can launch in pretty much any weather that doesn't involve large pieces of hail:

 

[Skryll]

That's really poor. My experience has been much better in terms of lane keeping. It hasn't crossed a yellow line in a long time. Maybe the FSD team isn't getting enough feedback from the vehicles like yours that are not doing well.

Same here in the SF Bay Area with my 2017 Model 3 LR RWD that I took delivery of in January 2018, current fw is 2019.12.

Its super reliable on fairly reasonable curves and especially awesome in stop and go single lane two way traffic, not just on the freeway. It amazes me on how well it lets people merge in from the right, used to always have to take over for that but now its pretty rare. Lane changes work alot better now too, less hesitant and slow than it used to be. And Navigate on Autopilot is something where I am usually too much of a micromanager wanting to stay in the soon to exit lane until you have to move over, not merge into thick traffic close to the exit etc, but recently found myself realizing that it was right and am now more inclined to follow its advice. There are some parts where I know it could handle it if it stayed in NoA mode a bit longer after exiting, but it switches back to vanilla AP.

It still has issues with hairpin turns on mountain roads, and generally I am very selective on where I use it also in town, but there are always easy parts where it is super useful.

My conclusion is - there is no full FSD with this HW 2.5, but if HW 3 is dramatically better, it may very well get there this year.

 

[EVNow]

Now it has been getting better - it would go way further over the yellow line 3 months ago and even more 6 months ago - but it doesn't inspire confidence that FSD is right around the corner if the car can't handle these ideal conditions. And that's just one example. I have dozens more where AP has performed poorly where you'd expect it to have no problem.

Interesting - I wonder what we think of ideal are actually ideal for NN or not. May be once the lines aren't so bright and the road isn't as black, it behaves better ?!

BTW, does anyone know how EAP handles turnarounds (even without traffic) ?

 

[mongo]

And in addition they had a $932 million debt repayment.

I think that is on the cash flow balance sheet, not the profit/ loss. Just like taking out the loan does not increase profits.
(however, I am not an accountant, so I could be getting my sheets twisted)

 

[EVNow]

This is also why I think the news (whatever it is) on Monday will have little effect on the stock in the long run, unless it is true full feature FSD, which I doubt.

Tesla has a simpler goal to achieve. Tesla needs to show that they are at the same level as Waymo or Cruise - who according to the market are leaders. Market is simply not convinced - whatever EM might say - that their approach is better than Waymo's. So, they have to actually prove.

That is why live demo on urban roads to participants would be positive - even when "geofenced".