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

[SOULPEDL]

You are not making much sense.

I use FSDb all the time too. Have been since it came out - infact used EAP extensively before that.

That doesn't mean "robotaxi" is close by. You disagreed with a comment it might take 10 years - but as you confess - you have no idea when (or if) it can be achieved.

(Edit: I forgot, this might be dated. It's Johnny Carson making a guess what's in the envelope. But for reasons stated (2 now) my guess will be sooner than 10 years. Capable this year, released somewhere by '24 but they may wait for the Gen 3 and go all out. So I could be wrooon... n..ghu. (said like The Fonze).

Perhaps you are frustrated like my brother who expected that $30K annual income - I was under the same belief and it's disappointing. But AI is clearly accelerating, Dojo 10X ('23) then 100x ('24) per Elon's estimates. So I'm in with pretty high confidence by next year.)

 

[2daMoon]

View attachment 922763

(Edit: I forgot, this might be dated. It's Johnny Carson making a guess what's in the envelope. But for reasons stated (2 now) my guess will be sooner than 10 years. Capable this year, released somewhere by '24 but they may wait for the Gen 3 and go all out. So I could be wrooon... n..ghu. (said like The Fonze).

Perhaps you are frustrated like my brother who expected that $30K annual income - I was under the same belief and it's disappointing. But AI is clearly accelerating, Dojo 10X ('23) then 100x ('24) per Elon's estimates. So I'm in with pretty high confidence by next year.)

That's not Johnny Carson,

that's Carnac the Magnificent.

 

[Ogre]

@SOULPEDL @2daMoon Y'all just aged 10 years in one meme.

 

[EVNow]

Perhaps you are frustrated like my brother who expected that $30K annual income - I was under the same belief and it's disappointing. But AI is clearly accelerating, Dojo 10X ('23) then 100x ('24) per Elon's estimates. So I'm in with pretty high confidence by next year.)

You are absolutely wrong.

When I bought FSD I knew it is experimental and might be a decade before anything approaching L3 is achieved. I’ll post my comment to prove it.

Disengagement rate is still about 1 in 10 miles. When it approaches 1 in 1,000 miles, get back to me (assuming there are 10x the number of disengagements than are legit).

ps: See this from Apr-2019. Interesting times … arguing FSD with neroden.

ps : I don't even consider robo taxis a near term possibility. We can come back to this issue in 2030 and see where we are.

 

[FSDtester#1]

You are absolutely wrong.

When I bought FSD I knew it is experimental and might be a decade before anything approaching L3 is achieved. I’ll post my comment to prove it.

Disengagement rate is still about 1 in 10 miles. When it approaches 1 in 1,000 miles, get back to me (assuming there are 10x the number of disengagements than are legit).

We don't pee in water fountains, we use the bathroom.

Fsd discussions, if they aren't related to the stock price going up or down in the near future belong in one of the 5000 Fsd threads. Either way, I will put both of your predictions in my excel spreadsheet and the winner gets a bottle of Burnt Hair Cologne. Only because I love both of you.

Mods: You can give me 1000 TMC credits for helping out.

 

[Singuy]

You are absolutely wrong.

When I bought FSD I knew it is experimental and might be a decade before anything approaching L3 is achieved. I’ll post my comment to prove it.

Disengagement rate is still about 1 in 10 miles. When it approaches 1 in 1,000 miles, get back to me (assuming there are 10x the number of disengagements than are legit).

Disengagement rate is a useless metric for a non-geofenced car and I don't know why people keep using this as a way to compare autonomous vehicles. There are routes FSDb will have a disengagement 100% of the time within 100 ft and there are routes you can drive for 1000 miles and not have any disengagements.

However talking about L3. I see that Hyundia's newest EV SUV will have L3. I suspect after Mercedes, this is the new BS marketing trick to make their cars seem more capable than they are with the use of geo fencing + extremely limited capabilities. However like bits for gaming consoles and memory sizes for graphics card...the general public will see L3 and say how it's better than L2.

 

[Knightshade]

However talking about L3. I see that Hyundia's newest EV SUV will have L3. I suspect after Mercedes, this is the new BS marketing trick to make their cars seem more capable than they are with the use of geo fencing + extremely limited capabilities. However like bits for gaming consoles and memory sizes for graphics card...the general public will see L3 and say how it's better than L2.

Mercedes L3 system is absolutely very limited. Low max speed, and only on highways- it's very much the same sort of traffic-jam-only system Audi claimed they'd bring to market years ago then never did-- But Mercedes actually has.

Hyundais system will be similar, but with a higher top speed (50 mph rather than 37 for Mercedes)

That said, they WILL (already are in Mercedes case) be functional, sold-to-consumers, L3 systems.

If Tesla genuinely believes their system is better, people will ask why can't they offer something similar? It's useless for local roads and people who live in low-traffic areas doing 70 on the highway, but it's legitimately valuable to folks who might sit in slow jammed highway traffic for a lot of hours a year and there's many millions of such people, including in Teslas largest US market, California.

 

[AndrewZ]

It's also worth noting that Mercedes and Hyundai are using implementations of tech from NVIDIA and Mobileye (I believe) respectively, and each have configurations that are capable of attempting city street autonomy like FSD. Whether those companies have any interest putting that stuff in consumer's hands, that's a very different thing, but they both have test fleets out there doing similar development and you can find videos on YouTube.

 

[Singuy]

Mercedes L3 system is absolutely very limited. Low max speed, and only on highways- it's very much the same sort of traffic-jam-only system Audi claimed they'd bring to market years ago then never did-- But Mercedes actually has.

Hyundais system will be similar, but with a higher top speed (50 mph rather than 37 for Mercedes)

That said, they WILL (already are in Mercedes case) be functional, sold-to-consumers, L3 systems.

If Tesla genuinely believes their system is better, people will ask why can't they offer something similar? It's useless for local roads and people who live in low-traffic areas doing 70 on the highway, but it's legitimately valuable to folks who might sit in slow jammed highway traffic for a lot of hours a year and there's many millions of such people, including in Teslas largest US market, California.

The reason is cost.

Hyundai/Mercedes understand their system is not enticing enough to help sell cars, but Nvidia/Mobile eye is not cheap. So mobile eye most likely signed a deal with the company in which they will take a lot of the legality burden off Hyundai/Mercedes and allow them to advertise L3 as an enticing feature to sell cars. Tesla's FSDb and autopilot are enticing enough to sell cars, and adding L3 they believe most likely will just increase cost and not sell any more additional cars because their customers thinks L3 from competitors are BS anyways.

 

[unk45]

CNN International seems generally less stupid.

Oh No.. exactly the same.

It is just that on the subjects that we don't that deeply we tend to think they are being more neutral and fair. Not even close. Lazy biased reporting is pretty much the same on international bureau too. They go out of the way to hire reporters who will write slanted view of the world that Americans cherish. I know..

Maybe watch CNN Brasil that is totally different than CNN International. No question that Television/cable ‘news’ is built to generate revenue. By definition they cannot be impartial.

 

[SOULPEDL]

I bet we have the same debate regarding Optimus vs the competition - it's quite the similar challenge. (Discussion reminds me of when they first tried to cross the Atlantic... by boat... Remember that! slaps knee).

So let me get my Carnac the Magnificent hat on to say... I believe there will be more than one legit position replaced by Primus in the factory by next year. Easily. And by 2030, they will be unionize after their first strike for pushing their batteries too far and making it harder to lift things.

 

[Nocturnal]

Mercedes L3 system is absolutely very limited. Low max speed, and only on highways- it's very much the same sort of traffic-jam-only system Audi claimed they'd bring to market years ago then never did-- But Mercedes actually has.

Hyundais system will be similar, but with a higher top speed (50 mph rather than 37 for Mercedes)

That said, they WILL (already are in Mercedes case) be functional, sold-to-consumers, L3 systems.

If Tesla genuinely believes their system is better, people will ask why can't they offer something similar? It's useless for local roads and people who live in low-traffic areas doing 70 on the highway, but it's legitimately valuable to folks who might sit in slow jammed highway traffic for a lot of hours a year and there's many millions of such people, including in Teslas largest US market, California.

Current autopilot is more than sufficient for what Mercedes offers, and it even works in the rain.

It's probably a matter of cost vs benefit. How many more FSD subs will they sell if they offered highway L3 up to 24mph?

 

[aronth5]

It's also worth noting that Mercedes and Hyundai are using implementations of tech from NVIDIA and Mobileye (I believe) respectively, and each have configurations that are capable of attempting city street autonomy like FSD. Whether those companies have any interest putting that stuff in consumer's hands, that's a very different thing, but they both have test fleets out there doing similar development and you can find videos on YouTube.

I hope Mercedes and Hyundai are successful since that might push Tesla to make the few changes needed to support L3 on controlled access highways which Mercedes is touting. And yes everyone knows Mercedes current capability is awful for highways but that could change.

 

[Singuy]

Current autopilot is more than sufficient for what Mercedes offers, and it even works in the rain.

It's probably a matter of cost vs benefit. How many more FSD subs will they sell if they offered highway L3 up to 24mph?

I believe Musk finds L3 and even L4 a gimmick and doesn't see that as true autonomy.

 

[Cosmacelf]

It just struck me that Elon has said multiple times that advanced AI development should be done at a public company and that shareholders can vote him out, and that option is an important one.

Per other reports, Elon is also looking to build a team that competes with openAI after being deeply unhappy about the pivot to closed source.

Also, some credible people on Twitter say GPT 4 is the limit of what can be trained with the current gen Nvidia hardware, and the new H100 systems to scale up will take 6 to 8 months. Will this be as good as a ground up system like dojo? It's all transformers anyways (ducks and covers)

Putting 2 and 2 and 2 together, unless the answer is 5, I am thinking Tesla might get into LLM development. Optimus could really use this. Though I am not sure it can easily compete with the vast amount of data open AI has access to, via GitHub.

Tesla really should be getting into LLM development. I suspect it will very important for Tesla Bot.

Do note that many of Elons tweets against OpenAI are regarding the fact that it isn’t open anymore. And OpenAI is now essentially controlled by Microsoft, not exactly a company Elon has a lot of love for. His latest letter suggesting a GPT development pause is a broadside against OpenAI. It is isn’t game over for Tesla/TeslaBot/Elon, but it sure is annoying. Next, expect an actual government investigation into how a non profit can turn into a for profit. OpenAI is going to get some scrutiny of its own soon.

 

[Webeevdrivers]

I believe Musk finds L3 and even L4 a gimmick and doesn't see that as true autonomy.

Maybe. But seriously. Even Elon has to know that the robotaxi autonomous FSD dream is over, at least for the next decade or two. I personally don’t think it was a malicious deception on Elons or Teslas part, but they need to come clean on this fantasy of robotaxi being a reality with the current hardware software suite. It’s not even remotely close and at the present rate of improvement won’t be for the foreseeable future. I get it. FSD is hard and I’m sure it will happen someday. But Tesla needs to own this and be straight up with folks. I enjoy using FSD but it’s an entertainment feature not any kind of robotaxi feature. I think people would be more understanding if they just came out and explained that it’s just not going to happen on existing cars. Not 3 years ago, not 3 years from now and probably not in 10. Maybe with new hardware someday but not existing cars.

Personally I would still probably buy it again. It’s fun and I like my cars to be fun. Our teslas have always been way fun. But sell it as that. A fun infotainment thing.

Jmho.

 

[Tommy O]

Tesla really should be getting into LLM development. I suspect it will very important for Tesla Bot.

Do note that many of Elons tweets against OpenAI are regarding the fact that it isn’t open anymore. And OpenAI is now essentially controlled by Microsoft, not exactly a company Elon has a lot of love for. His latest letter suggesting a GPT development pause is a broadside against OpenAI. It is isn’t game over for Tesla/TeslaBot/Elon, but it sure is annoying.

There’s zero chance Elon isn’t aware of how important solving these types of problems are but there’s very little publicly known that they’ve been working on it. All I know about is a rumor that Tesla bought an NLP training corpus last year, but that’s a rumor. Maybe Dojo is in part planned for this, but they haven’t said so explicitly. Though their engineers talk about AGI.

Elon is quite vocal on Twitter and in interviews about his motivations for AI risk and has been for many years. I don’t really disagree with anything you’ve said, but calling it a broadside against OpenAI isn’t the phasing I’d use.

AI prognosticators these days tend to fall into 3 camps. Goal post movers, down right deniers, and people like me who see the advances happening weekly and the pace of advancement speeding up. My job has changed dramatically. I spend most of my time now reading about the latest advancements and incorporating them into my work flow. Many of my friends are doing the same.

I’ll be doing a 60 hour round trip drive (am I a masochist?) with a CV expert very soon. The tesla will drive us almost the whole way while I pick his brain.

What a time to be alive.

 

[SmokyPeat]

I bet we have the same debate regarding Primus vs the competition - it's quite the similar challenge. (Discussion reminds me of when they first tried to cross the Atlantic... by boat... Remember that! slaps knee).

So let me get my Carnac the Magnificent hat on to say... I believe there will be more than one legit position replaced by Primus in the factory by next year. Easily. And by 2030, they will be unionize after their first strike for pushing their batteries too far and making it harder to lift things.

Primus? Les Claypool fan?

 

[spacerfc]

Maybe. But seriously. Even Elon has to know that the robotaxi autonomous FSD dream is over, at least for the next decade or two.

Wut? Did you see what's happening in the AI field? We could get to AGI in under a decade, let alone the puny FSD...

If anything, I'd bet Elon is now more certain than ever that FSD will be solved very soon. GPT basically proved that you can get very close to human behavior by throwing enough compute and data into the thing, same approach as FSD.

 

[spacerfc]

Tesla really should be getting into LLM development. I suspect it will very important for Tesla Bot.

It'll be very important for basically everything. But to get into this race, I think Tesla needs to up its game in terms of R&D spending. Currently it's spending about $3B/year which is enough to run circles around legacy motor, but to face off OpenAI and Microsoft, they probably need to increase it by at least 50%, if not double it.