Karsten Klees retweeted
I agree with Sawyer's perspective, but that's unfortunately not how most people make their car buying decisions, whether they like to admit it or not.
Many TSLA investors think that the problem is either:
1. that Elon has been misrepresented in the media, or
2. that consumers should strictly focus on the quality of the product and ignore Elon's involvement in political matters.
But both of those takes show a complete misunderstanding of consumer psychology.
On the second point, if people were guided in their purchasing decisions strictly by the quality and/or the value-for-money of the product, then there should be a maximum of 2-3 products taking up all the sales in any product class, and nobody would be buying any of the others. Clearly, that's not the case; many items — and particularly cars — are emotional purchases, where design, style and brand image are important factors.
On the first argument, about Elon's public image: try to think of a really despicable person. Not Elon. Imagine a public individual you really, truly resent, someone who, in your view, embodies everything that's wrong with humanity. Maybe someone who's known to be abusive towards their spouse or children, or someone who exploited and/or harmed some people for their own benefit, etc. Maybe someone you're pretty sure has committed some illegal or criminal stuff but was never prosecuted. Would you buy anything from that person? Would you even consider the qualities of the product of service they're selling?
That's how many people, maybe the majority — particularly in Europe — see Elon. This is not a debate on whether their perception is correct or not, or whether they've been manipulated; this is simply a statement of fact: many people currently believe Elon is an immoral oligarch devoid of any personal merit who's using his wealth to influence politics both in the US and in foreign countries, and that his only goal is to enrich himself further. I see and hear this daily. Those people don't care about the safety performance of the refreshed Model Y, or its improved energy efficiency, or its better ride comfort; the simple mention of "Tesla" triggers a negative emotional reaction, and no reasoned argument as to why a Model Y is the ideal car for their needs will overcome that "ick" factor.
Now, we can spend a lot of time calling these people idiots or brainwashed sheep or whatever, but it won't change a damn thing. It definitely won't change their minds. The Tesla brand has been damaged by Elon's incessant political comments (amplified daily by the mainstream media for clicks), and a recovery of Tesla's public image will take a long time.
I'm buying a @Tesla Model Y because I want a vehicle from a brand that makes the safest cars on the road. I want easy access to a reliable charging network. I want to buy this car from a company doing more than any other for the environment and to push sustainable energy forward. I want the best overall EV ownership experience.
There are thousands of people who made this best-selling car possible, and I want to support them and Tesla's mission.
At the end of the day, I want to buy the best product, period.
Karsten Klees retweeted
🚨NEWS: $TSLA Affordable Model Q: Reportedly Launching Under $30K in 2025
According to reports shared during a Deutsche Bank investor conference by Tesla’s Investor Relations head, Travis Axelrod, the company plans to release this entry-level electric vehicle in the first half of 2025.
After subsidies in North America, the Model Q is expected to cost less than $30,000, making it Tesla’s most accessible model yet.
The Model Q, internally codenamed is designed with a compact build, measuring approximately 3,988 mm in length—about 15% smaller than the Model 3.
It also weighs 30% less, which significantly enhances its efficiency. Tesla plans to offer two battery configurations: a 53 kWh and a 75 kWh lithium iron phosphate option.
The top-tier model will boast a remarkable EPA-rated range of up to 500 km. Customers will have the choice of single-motor or dual-motor setups for varied performance needs.
Tesla’s ambitions don’t stop at affordability.
The company aims to expand its footprint in China with more versatile models, including a long-wheelbase, and a three-row Model Y.
This strategy is part of its plan to achieve a 20-30% sales growth by 2025.
With the Model Q, Tesla continues to push the boundaries of electric vehicle accessibility and innovation, setting a new benchmark for value in the EV industry.
Karsten Klees retweeted
I'm convinced Tesla will start selling Cybercab in the next 6 months as a drivable vehicle but without a steering wheel and pedals.
Instead, they'll use a joystick or some new form of interfacing with the vehicle.
Too many things are starting to line up, and all of it is enabled by the progress Tesla has been making with FSD, especially v13.
There are three major variables why I think this will be the case:
- Tesla's stated strategic goal, including more affordable vehicles starting in 2025
- Steering wheels and pedals are not required by law
- Progress with FSD and what that means to be a "driver" in a car
First, Tesla has a stated goal to reach 2M units per year run rate by the end of 2026 for the Cybercab.
This number makes absolutely no sense if the Cybercab is a car that is completely driverless. There aren't enough jurisdictions around the world that would have enough demand to satisfy the supply of 2 million driverless cars per year.
There's simply not enough driverless miles available for this many units to be put to work and it be a profitable endeavor. Nowhere near close.
This would mean that Tesla would be cranking out millions of cars per year without being able to sell them or put them to work, losing out on tens of billions of dollars.
This is strategically insane.
However, if Tesla COULD sell the same exact car, but with a way to actually drive it with little to no modification, then that problem is solved.
Enter the joystick.
By law, all major car markets DO NOT REQUIRE a steering wheel or pedal to control a car. It just happens to be the format that a) companies have employed and b) customers are accustomed to.
As long as an automaker can prove that their new method of controlling a vehicle, be it a joystick or whatever else, meets the required safety standards by regulators around the world, that input system will be approved by each jurisdiction. This is known as homologation.
This is also the reason why handicapped drivers missing limbs can operate vehicles.
However, this is a major risk that automakers would have to take on. What if the system doesn't work as well as intended? What if customers hate it? They wouldn't be able to sell their car, and cars are VERY expensive to develop and manufacture.
That risk would have to be weighed against a decades-long run of the steering wheel and pedals being the standard. Not to mention that an entire 3rd party supply chain has been built around it to bring costs down on this input format.
Not only that, but there's also decades of safety data that has been collected around steering wheel and pedals being a good and safe way to control a vehicle.
Moving away from a steering wheel and pedals would be an insane move for Legacy Automakers, who are risk adverse and want nothing more but to iterate year in and year out while protecting profits for shareholders.
But Tesla? Tesla does nothing but take risks day in and day out, and a brand new way to interface with the driving of a car would be right up Teslas alley.
Not only that, it makes all the sense in the world to have a joystick instead of traditional steering wheel and pedals given that a Tesla owner barely has to drive anyway.
Enter FSD.
With Tesla's latest Full Self Driving version, v13, miles between critical interventions are starting to creep into the thousands of miles. This means that a "driver" could go weeks or months before having to take control of the car for whatever reason.
The pace of improvement for Tesla in relation to the system's performance has been staggering. It is becoming obvious that Tesla has developed a system that WILL drive itself under any circumstance for hundreds of thousands of miles without human input in the next year or two. End of the decade at the absolute latest.
What this means is that in many cases, a Tesla vehicle will be driving itself for thousands of miles between interventions as early as this month while carrying the extra weight and cost of a steering wheel and pedal system.
This system, including the steering column, would weigh at least 20 pounds and cost at least $750 in parts.
In the automative world, this means a huge hit to profits, and extra weight that impacts range, which means that Tesla would have to add more batteries - which is the most expensive part of the car - to offset that weight.
It makes no sense to carry around all that weight and cost if the car is driving itself 99%+ of the time.
A joystick would weight no more than a quarter of a pound, and cost no more than $20, saving HUGE amounts of money and weight.
Fortunately for Tesla, all the components required to control a car can be handled by a joystick TODAY. All input systems in the latest Tesla car today, the Cybertruck, are already electronically controlled.
A lot of cars already employ brake and drive by wire, including Teslas, which means that the braking and accelerating are electronically controlled. There's no mechanical link between the pedals and the engine and/or brakes.
And on top of that, Tesla has now introduced steer by wire with its Cybertruck, which removes the mechanical link between the steering wheel and the front wheels.
All these electronic inputs can be migrated to a joystick (forwards for accelerating, backwards for braking, left and right for steering). You can place buttons on the stick itself for blinkers, flashing lights, etc.
On their existing cars, Tesla has already moved all controls away from stalks and buttons and into the screen and steering wheel. The precedent is already set for overhauling the input system.
I mean for goodness sake, just look at the Cybertruck. Many thought Tesla wouldn't be able to get approvals to sell that thing, and yet the US is being overrun with massive metal triangles on wheels like there's no tomorrow.
Tesla sells more Cybertrucks than the rest of the EV truck market COMBINED - and will only extend its lead in 2025.
And so it all starts to align - Tesla can start selling Cybercabs next year starting at $29k, sell it with a joystick so that it meets all regulation, start front-loading the market with Robotaxis that will be turned on as jurisdictions allow it, and flood the market with the optimal way to get people from point A to point B at the lowest cost per mile possible.
Genius.
In conclusion, Tesla has reinvented what it means to be a "driver" in the car. With its progress in FSD, Tesla owners will be able to go thousands, if not tens of thousands of miles between interventions with their vehicles by next year.
In that world, it makes no sense to carry around all the extra weight and cost of a steering wheel and pedal system. Instead, that system should be handled by a joystick or similar to still allow for drivers to intervene where needed and meet all regulation standards around the world.
Not only that, but this is the only way Tesla has a realistic shot of fulfilling their goal of making 2 million Cybercabs per year run rate by the end of 2026.
By doing this, Tesla will start selling their optimal self-driving solution into the market that requires no additional modifications, given that a Joystick meets all "drivable" requirements and costs and weighs next to nothing.
Finally - Tesla can start making money on their next-generation product TODAY, vs their competition like Waymo and others that are still bleeding money because of their very high costs and very low volumes relative to Tesla.
Tesla can only do this because the cost of their 8-camera and AI brain system is a fraction per unit vs its competitors.
No one else can make a car and sell it for $29k, sell millions of them, and sell them profitably.
It's truly game, set, match.
I kept doubting Tesla and its executives when they said the Cybercab will never be sold with a steering wheel and pedals. The strategic timing and release of the vehicle makes no sense without a drivable version.
Turns out they are telling the truth. They're just not saying that the car will be drivable with a freaking joystick.
The cherry on top will be when Tesla opens up reservations completely randomly in the middle of next year with near-immediate delivery, absolutely breaking the internet.
Can't wait for all of this to be completely wrong and for me to look like a complete idiot.
Karsten Klees retweeted
Replying to @stekkerauto
AI says:
The self-defeating narrative around EVs (electric vehicles) in Germany, particularly among auto executives and media, is indeed puzzling when viewed from a long-term survival perspective. Several factors may explain this paradox:
1. Inertia and Legacy Thinking
• The German auto industry is deeply rooted in internal combustion engine (ICE) technology, with a century of expertise, infrastructure, and global dominance. Shifting to EVs means abandoning or diminishing these core competencies, which is a difficult pill to swallow for established players.
• Executives may perceive EVs as a disruption rather than an evolution, fearing the loss of their historical competitive advantage.
2. Economic Concerns
• The auto industry in Germany is a major employer, and the transition to EVs involves fewer moving parts and a simpler supply chain compared to ICE vehicles. This means potential job losses in manufacturing, parts supply, and maintenance.
• Resistance could stem from fear of destabilizing the economy in regions heavily dependent on ICE-related jobs.
3. Cultural Attachment to ICE Cars
• German engineering excellence, particularly in ICE vehicles, is a point of national pride. Brands like BMW, Mercedes-Benz, and Volkswagen are icons of performance and luxury. The transition to EVs could feel like an erosion of identity.
• High-performance EVs, like Tesla, have disrupted this narrative, leading to a defensive posture rather than embracing the challenge.
4. Short-Term Profit Focus
• EVs require significant upfront investment in R&D, factories, and supply chains (e.g., battery production). While these investments pay off long-term, the initial cost pressures may deter executives who are judged by quarterly profits and shareholder returns.
5. Media Influence and Resistance to Change
• Some German media outlets may cater to skepticism among the public, exaggerating challenges like range anxiety, battery recycling, or high costs.
• This skepticism feeds a feedback loop where auto executives feel less urgency to adapt, believing the market isn’t ready.
6. Political and Lobbying Dynamics
• The German auto industry has historically wielded significant political influence. Policies favoring ICE vehicles—like slow adoption of stricter CO₂ regulations—have delayed EV progress.
• Lobbying efforts may reflect a desire to extend the profitability of ICE vehicles for as long as possible.
7. Underestimating the Competition
• Many German automakers underestimated how quickly Tesla and Chinese manufacturers would rise, dismissing them as niche players for years.
• By the time they recognized the shift, competitors had already established significant market shares and brand loyalty.
8. Fear of Losing Control
• EVs are as much about software as hardware. Traditional automakers risk losing dominance to tech-focused companies that excel in software integration and user experience.
• Germany’s legacy automakers are used to controlling the entire ecosystem—from design to distribution. EV disruption threatens this model.
Why Undermine Your Survival?
This behavior reflects a classic case of The Innovator’s Dilemma—a reluctance to cannibalize your own lucrative business even when the alternative is being cannibalized by competitors. However, this delay tactic is self-defeating. As global EV adoption accelerates, driven by climate policies, consumer demand, and falling costs, the longer automakers resist, the harder it will be to catch up.
Moving Forward
While some German automakers are now pivoting (e.g., Volkswagen with its ID series), the narrative shift must be faster and more decisive. Public perception, policy support, and a willingness to embrace change are key to ensuring survival in the EV era.
Karsten Klees retweeted
Kommt nach KI jetzt Singularität? 😳
@mherger
#ki #AI #superintelligence #superintelligenz #alberteinstein #tesla #xai #marioherger
Karsten Klees retweeted
Ein Autohersteller verpennt mehrere Trends und hat völlig vergessen, was Innovation bedeutet. Den Grund sucht man aber ausschließlich in der Politik. Mein Kommentar zu den Problemen bei Volkswagen.
Karsten Klees retweeted
Waymo is netting a revenue of $ 75.000 🤑 a year per vehicle
in 3 cities Los Angeles, San Francisco, Phoenix.
Can you imagine earning this type of revenue with your Tesla ........ after an software update
Only lees then 20 days and we will probably learn more where we can expect to start with the first Tesla robotaxi's
Karsten Klees retweeted
🚨 FSD Supervised is coming to Europe in 2025 and the end of the Autopilot “nag” might be worldwide! 🚨
As we look ahead to Tesla’s FSD Supervised launch in Q1 2025 for Europe and China, I wanted to dive into how these developments tie into the ongoing work of the @UNECE GRVA and what this means for Tesla drivers.
You might remember my earlier posts about the GRVA’s workshops on autonomous driving systems, like the test drives of FSD v12 in Troy, Michigan and the hands-on feedback from experts like the Swedish Transport Administration. There’s been a lot of movement on the DCAS (Driver-Control Assistance Systems) front lately, and it’s all pointing toward one thing: we are getting closer to a world wide release of FSD (Supervised)🙌
The Path to FSD (Supervised) wide release: Key Updates
With Tesla’s announcement that FSD Supervised will be rolling out in Q1 2025 for Europe, the latest regulatory updates couldn’t be more timely. The WP.29/GRVA is making rapid progress, particularly with the DCAS legislation (UN Reg. No. 171) that is laying the groundwork for FSD Supervised in Europe.
The GRVA's System-Initiated Manoeuvres (SILC) guidelines are a big deal. These updates will allow FSD (Supervised) to make autonomous lane changes, merging, and other manoeuvres without needing prior driver confirmation. It’s a massive step forward for hands-off driving on European roads.
Even bigger is the conversation around HOR (Hands-On Request). One of the most exciting developments is the proposed amendments to withhold HOR on highways—essentially, getting rid of that annoying “nag” telling you to keep your hands on the wheel. This could mean no more HOR, at least on highways, unless the system detects a boundary. 🚫✋
#FSD in Europe: What to Expect in 2025
So, how does all this regulation work connect to Tesla’s upcoming release? Well, the DCAS framework that’s being finalized will directly influence @Tesla's FSD Supervised in Europe. Tesla will need to comply with the new UNECE requirements, including guidelines around system transparency (clearly showing what actions the car will take) and driver intervention capability.
For us, this means we’ll see a system that’s not only technically compliant but also incredibly advanced. In fact, FSD Supervised is expected to launch with no more mandatory driver confirmation for lane changes and other actions—a major win for Tesla’s autonomy goals. The HOR may still apply in some situations, but Tesla’s advanced Eye-on-Request monitoring might eventually eliminate it altogether, just like we’re hoping for in the U.S.
The Future: What’s Next for FSD?
I’ve been following this regulatory journey for a while now, and one thing’s clear: the UNECE and Tesla are working faster and more efficiently than ever before. Just like I mentioned in my latest regulation post, we’re seeing regulators not only keep up with the rapid pace of innovation but even helping to speed it up.
The upcoming GRVA sessions in September will finalize some of these critical changes, and it’s only a matter of time before we’re driving with FSD Supervised—and yes, potentially without the nag altogether! The way things are moving, 2027 might be the year we see FSD Unsupervised hit European roads. Imagine that—full autonomy, no driver required. 🚘🤖
As we get closer to these milestones, I’ll keep you posted on all the latest developments. For now, get ready for an exciting 2025. FSD is coming, and it’s coming fast! ✌️
Disclaimer: I am not a regulatory professional so I can make interpretation mistakes.
#Tesla #FSD #FSDSupervised #DCAS #UNECE #AutonomousDriving #NoMoreNag
Karsten Klees retweeted
Despite the protests against Tesla, the community representatives of #Grünheide just decided in favor of the new development plan for the expansion of #GigaBerlin for the infrastructure and the new rail yard!
teapawt.net/tobilindh/status/17683…
After the vote of the citizens of #Grünheide against clearing another 100 ha at #GigaBerlin, a new sketch for an adjusted development plan has been published.
This would make all planned infrastructure projects possible, while saving 70 ha of forest.
gruenheide-mark.de/news/1/92…
Karsten Klees retweeted
The end of the @tesla Autopilot nag might be worldwide! Find out more 👇
UNECE GRVA UPDATE May 13th DCAS legislation (FSD Supervised)
As we are gearing up for the next (19th & 20th) sessions of the GRVA in resp. late June and September I thought I’d share and post my findings.
Last time I shared something was about the GRVA agenda for may and june and the GRVA going on a trip to Troy, Michigan 🇺🇸 to testdrive FSD.
We’ve seen some cool posts by @Tslachan here on X of FSD v12 being demoed to an advisor to the Swedish Transport Administration who was very impressed by the safety of v12. This all thanks to the tireless efforts and persistence of @VImpeMarc and the development of DCAS, UN Reg. No. 171.
Among the topics today and the coming days are ofcourse the workshops on scenarios and test results of a deep analysis of a Model 3 with FSD v11 and a screening of a Model 3 with v12. But there was more!
The most notable are amendments to the DCAS legislation regarding System-Initiated Manoeuvres (which make up the heart of FSD) and last but not least the withholding of HOR.
Withholding of HOR is a very cumbersome way of saying, NO MORE NAG! At least to some degree. HOR stands Hands-On Request, aka the nag.
One of the proposed amendments is to stop hands on requests on highways unless a system boundary is detected. To me this reads as a step in the right direction to get rid of the nag.
The other is about the system initiated manoeuvres, or SILC in UNECE terminology, which stands for system initiated lane changes. The proposed amendments and comments by different countries are IMO a really good step in the right direction. Because there are a lot of proposed situations where SI manoeuvres will possibly be possible under DCAS, i.e. positive news regarding FSD supervised!!
All together, as I said in earlier posts, the regulators are getting better and quicker. The speed of developments and positive signs we’re seeing around us are becoming clearer by the day.
A safer and more comfortable future of transportation is right at our doorstep, we only need to open the door!
I’ll try to keep you posted in the coming weeks! ✌️
wiki.unece.org/display/trans…
#FSD #Supervised #FSDBeta
@teslasi @aikisteve @TeslaBoomerMama
Karsten Klees retweeted
Richte dir eine Erinnerung für meinen kommenden Space ein! teapawt.net/i/spaces/1OdKrjpmWvAKX
Karsten Klees retweeted
We could make some small improvements to the older versions of FSD, based on the updated DCAS regulations at the end of the year. But unless there is an updated allowance for “system initiated” maneuvers, the full benefits of systems such as Supervised FSD are not type approvable.
The way the updated regulation works is that the driver would need to approve of almost every significant maneuver or set of maneuvers. Practically I’m not even sure how that might work or be designed, but it sounds like a massive pain/distraction for a driver - the opposite of the point of both the driver assistance and driver monitoring benefits of Supervised FSD.
Karsten Klees retweeted
Autonomous Driving, Social Turnaround
I know that the thousands of followers I've accumulated aren't here to be bored with automotive stuff. I'm aware of how fashionable it has become to hate on Elon Musk.
But I have a serious handicap: I understand what they're doing at Tesla. “Unfortunately,” I happen to be an automotive expert in my day-to-day life. Mea culpa.
It's precisely for this reason that I grasp the society-shaping change that happened last week. And regardless of the "Elon bad" sentiment, I want to share with you what happened, why it happened, and how, so you might see the future a bit sooner than others.
It's about autonomous driving. Last week, the wicked Elon made the Full Self Driving (FSD) autonomous software's latest "supervised" version 12, previously in beta, freely available to all Teslas sold in America.
Does this news even matter? After all, the FSD beta has been accessible for at least four years to those who signed up for the beta program, and the world hasn’t turned upside down, despite the release of four versions from 8.0 to 11.0. None of these changed the course of world history.
But last week's release, version 12, will.
Why? What's the difference?
The difference lies in the fact that until now, no one's autonomous software (Tesla, Waymo, MobilEye, Cruise, etc.) understood the world around it, despite using neural networks for years for specific tasks (e.g., vision).
Instead, the programmers at these companies, understood the world and tried to code what's arguably uncodable – with varying degrees of success.
Let's face it, previous autonomous solutions were downright dumb. Their dynamic situation recognition was barely above ground level.
You're probably familiar with the self-driving taxis that drive into dead-ends and then just sit there, clueless. Or Cruise's infamous accident where although it wasn't the car that hit a pedestrian, but it failed to recognize a lying body and drove over it multiple times before "skillfully" parking.
Of course, everyone's autonomous software operated this way, including Tesla's, because it never occurred to anyone that there could be a different approach. That maybe the car could actually be smart.
This wasn't even remotely obvious until December 2022; it seemed utterly impossible. Everyone was fabricating all sorts of things, like Tesla with its occupancy network, heuristics, and other magical, yet capability-limited "solutions."
Then, at the end of 2022, ChatGPT appeared and showed the world that intelligent neural networks could indeed exist. I'm not saying it's infallible, but it's intelligent.
ChatGPT also makes huge mistakes, but not because it doesn't understand the world, but because it was taught inaccurately or nonsense about a particular subject.
ChatGPT's knowledge doesn't have a theoretical upper limit.
Let me repeat 'cause it's important:
ChatGPT's knowledge doesn't have a theoretical upper limit.
If humanity discovers new things and documents them, ChatGPT, after a new learning phase, will know the novelty just as well as the old stuff.
How does ChatGPT work, and why is this important for us?
Let's not debate whether a CSV file that can be opened in Excel is a sentient being. In my view, no text file is a sentient creature, but I might be alone in this.
The fact is, ChatGPT is just that, a massive text file that describes the "weights," or threshold values, of artificial neurons, triggering them to fire.
I won't delve into the details now, but suffice it to say that behind the intelligent text responses is a huge file filled with numbers, not text.
What's interesting, and relevant here, is how a file filled with numbers managed to gain such intelligence.
The idea behind ChatGPT is simple yet profound. The folks at OpenAI tried to teach a neural network to predict the next word in a conversation. Like the big brother of autocorrect that suggests words instead of letters.
Sam Altman must still be picking up his jaw because this seemingly simple idea created understanding within the neural network! Knowledge was created in a mere CSV file!
The most straightforward and accurate way to predict the next word in a text or conversation? Understand the text! Bingo, right?
The network stumbled upon this solution itself because it was penalized for mistakes until it developed a structure that prefers not to err to avoid punishment.
The breakthrough was miraculous: ChatGPT clearly understands human civilization. While we don't understand exactly what's in that CSV file that makes it intelligent, we know the process that smartened up the CSV: simply ask it to make predictions and penalize errors.
Hmm. Could this principle be applied in other areas? Like vision? Or autonomous driving? If we trained a neural network to predict the real world's next move, would understanding of the world emerge?
The answer is a resounding and massive YES. This is what's happening.
In hindsight, it's clear that all previous autonomous software should have been discarded the moment ChatGPT emerged, regardless of the billions of dollars companies had invested in their development. A dumb system with a theoretical upper limit of knowledge, is worth replacing with a system that can theoretically expand its knowledge indefinitely and is conscious and intelligent.
Enter the "sunk cost fallacy." This economic concept describes the phenomenon where outdated technology, acquired or developed at a high cost, continues to be used simply because it was expensive to obtain or develop. However, this decision is irrational; every market player would be better off discarding the old junk. But they don’t.
To this day, Waymo, Cruise, MobilEye, and the Chinese companies have not discarded their decade-old, polished, but as of 2024, clearly flawed, unusable, outdated autonomous code because "there's already billions of dollars in it."
Tesla, under the wicked Musk, was the only one to completely rewrite its previous solution in light of ChatGPT, reportedly throwing out 300,000 lines of C++ code and replacing the driving logic with a neural network-based system.
Thus, the Full Self Driving v12 was born, which not only sees and understands but also predicts the next step of reality.
And if its current knowledge leaves something to be desired, because as of now v12.3.3 scrapes the rear wheels in corners, it's not a limitation of capability but of the training framework and the input data, which, by the time this article is published, will have been corrected.
Okay. But what does all this mean?
What does the creation of an autonomous driving system with theoretically unlimited capabilities mean? One that sees with 8 eyes instead of 2, never tires, is never inattentive or angry, certainly not drunk, and doesn't ask for a salary?
And is statistically proven to drive much safer than any Homo Sapiens?
Massive social change.
Specifically, the extinction of the driving profession. This means the disappearance of millions of jobs. Which, in turn, leads to economic chaos, machine smashing, protests.
And something else: a Tesla monopoly.
The competition will catch up to Tesla in autonomous driving just as other space companies caught up with SpaceX in launching and safely returning rockets from orbit.
A Tesla monopoly means unimaginably high profits and stock prices.
It doesn't matter what the Wall Street smart alecks say about FSD v12; their opinion is irrelevant because they don't understand the whole picture. Their ignorance will only last on the stock market until reality knocks it out. How many years will that take? I think at most two. But who knows?
After all, wicked Elon needs the money. He plans to build a city on Mars, out of his own pocket! He need a monopoly!
So that's it. Should I write more like this, or should I take this and go away? Your choice!
@eavidan @BizOptimizer @Dangerbug @scottconso @BLSmith2112 @Chamber32Doors @AzharTheGreat @universexy69 @DogeMarineP @wholemars @FudBos @RealJoeDee1 @medicgordus @SaltasticTakes @DominikFilkus @FSSeychelles @EagleRider4 @a3voices @innovateathome @gntraveler @MarkMCruden @SKircher6 @Alexand51194866 @visitseychelles @TheCyberCamper @MomOf6x @theschnell3 @buylowbro @Th3Wlzard @AntonyCassidy1 @LOMWILD @LeoBerSt @shengyuhuang6 @challinor_dave @cosastesla @AURELIUPORUMBE1 @mike_bidde60926 @cosydell @_WhyCookie_ @TrendSpider @TeslaAlgo @elonmusk @fritz_rauer @ElectricTo11 @TeslaBoomerMama @therealCoinmon @TC14156853 @defymundane @csanyi_andras @WR4NYGov
Karsten Klees retweeted
OpenAI's rebuttal to @elonmusk's lawsuit raises TONS of questions.
First, it was the board of OpenAI kicking Sam Altman out for 'lack of transparency'. Then, it was Sam coming back to OpenAI and firing most of the board.
Then, it was Elon's suit of OpenAI for being neither open or non-profit. And now, we've got a blog post from OpenAI refuting Elon's lawsuit.
The drama seems to just be getting started, folks.
Let's start with an email that was sent to Elon on 1/31/2018. Who this email came from is redacted (let's call them Unnamed), but this person proposed the idea of OpenAI attaching itself to Tesla so that Tesla can be the 'cash cow'.
I've highlighted the juicy bits, but I strongly recommend you read the entire email. OpenAI claims this email was sent to Elon, which was then forwarded to Ilya and Greg at OpenAI.
From where I stand, OpenAI was having a very difficult time getting enough funding in order to compete with Google in its current form as a non-profit, open source start up.
Unnamed threw out the idea of OpenAI going for a for-profit pivot. However, they highlighted risks that going this route could fall flat on its face - things like company culture and greedy investors getting in the way of development in order to compete with Google.
Unnamed proposed for 'OpenAI to attach to Tesla as its cash cow'.
It's unclear how this "attachment" would work exactly, but the goal was for Tesla to have access to OpenAI's technology, which would allow it to develop self-driving technology quite quickly, which could allow Tesla to generate massive amounts of profits and revenue in a relatively short period of time.
This would in turn allow Tesla send a portion of the profits from said technology to the OpenAI team in order to invest in its compute hardware.
I'm assuming this sort of deal would give Tesla some form of exclusive licensing of OpenAI's technology, but tough to tell. This would imply OpenAI likely wouldn't do business with any other entity except for Tesla if all parties had agreed.
According to the email, Elon agreed that 'Tesla is the only path that could even hope to hold a candle to Google.'
The blog states that in 2017, OpenAI and Elon 'decided the next step for the mission was to create a for-profit entity. Elon wanted majority equity, initial board control, and to be CEO.' Afterwards, 'We couldn't agree to terms on a for-profit with Elon because we felt it was against the mission for any individual to have absolute control over OpenAI.'
It was at this point that the above email was forwarded by Elon, claims OpenAI.
OpenAI's blog then states Elon soon chose to leave OpenAI saying 'that [OpenAI] probability of success was 0, and that he planned to build an AGI competitor within Tesla.' It also says that '[Elon] told our team he was supportive of us finding our own path to raising billions of dollars.'
Then, there's this email that OpenAI attached from 2016 where Ilya said: 'As we get closer to building AI, it will make sense to start being less open. The Open in openAI means that everyone should benefit from the fruits of AI after its built, but it's totally OK to not share the science."
Here's the email:
Elon replied 'Yup' to the entire email.
All of this is to say - OpenAI claims that Elon and OpenAI were both aligned in making the entity into a for-profit, and that as AGI got closer, the entity would become increasingly 'more closed'.
Directly from the blog post: 'We're sad that it's come to this with someone whom we've deeply admired - someone who inspired us to aim higher, then told us we would fail, started a competitor, and then sued us when we started making meaningful progress towards OpenAI's mission without him.'
Now, what's really interesting in all of this from my standpoint, is that Elon's suit claims that Sam approached him 'to form a non-profit AI lab that would try to catch up to Google in the race for AGI.'
With this claim, Elon is putting the onus of OpenAI's founding, including starting as a non-profit and open source, on Sam.
In addition, in 2017, which is before the email I attached about Tesla being the 'cash cow' for OpenAI, Elon claims that he made it very clear that he would only continue to fund OpenAI if the parties committed to stay at the firm as a non-profit.
It seems like as time passed, it became clear that OpenAI would need to switch up its ability to raise funds in order to compete with Google, and some form of for-profit arm was basically inevitable.
Given all this, and knowing that both sides are going to duke it out in court, I'm going to take a stab and lay out what I 'think' happened, which is what ultimately brought the suit by Elon forward and prompted the response by OpenAI:
- Sam approached Elon to form a non-profit AI group so that they can tackle Google's accelerating AI ambitions, especially as it related to safety and ensuring AGI wouldn't turn into a massive disaster. Elon would provide funding, and Sam and Greg would put together a team to get to work.
- As time went on, it became increasingly clear that OpenAI needed to raise way more funds in order to compete. Short of becoming a for-profit venture, it would be essentially impossible to keep up with Google due to how much compute hardware was needed.
- Sam and Greg started to figure out how to turn OpenAI into a for-profit venture, which frustrated Elon due to the original proposition by Sam. Over time, they were able to all align that OpenAI would need way more funding.
- At this point, Elon proposed either being in full control of OpenAI as a for-profit venture, or bring OpenAI under Tesla's wing so that it can bank roll the entity's ambitions, and at the same time, enable Tesla to gain exclusive access to AI technology that would allow it to have a giant competitive advantage with self-driving tech and be a counter-force to Google's ambitions.
- After OpenAI team felt uneasy about giving Elon all the control for OpenAI, Elon left OpenAI and decided to start a Google competitor on his own. Elon continued to fund OpenAI's ambitions as he perceived them to be good.
- The OpenAI team then started making moves becoming a for-profit, the biggest move being partnering with Microsoft, which over time birthed GPT4 and made OpenAI into what it is today. This allowed OpenAI to gain a giant advantage and supercharge its development towards AGI.
- Then, in 2023, all hell breaks loose at OpenAI and Sam gets fired from the board for his 'lack of transparency'. In a shocking turn of events, Sam gets his old job back and is able to completely replace the board so it more closely aligns with its for-profit motives. To this day, no one knows wtf happened.
- By this point Elon has watched OpenAI, which he personally funded to the tune of $45 million, turn from what was pitched to him as a non-profit, to a for-profit, that has a quasi-exclusive relationship with Microsoft, that likely divorced itself from its founding premise the second Sam was hired back and the board was gutted after he was fired.
- Elon sues.
From OpenAI's perspective, they view Elon as someone who is no better than they are - he was part of said non-profit, and decided to leave and start his own for-profit in order to compete with Google once the OpenAI team was unwilling to agree with him either having majority control of OpenAI, or taking it under Tesla's wing.
However, if I were a betting man, I don't think Elon sues if Sam if he was never fired from OpenAI.
The reason why we're all here, I think, is because Elon doesn't trust Sam - right or wrong.
I think this whole lawsuit has less to do about OpenAI being a for-profit or open source, but more to do with laying everything out in the open so that everyone knows what the hell happened at OpenAI as it transformed into a for-profit company, formed a cozy relationship with Microsoft, and the subsequent firing and rehiring of Sam Altman, which caused a massive shake up at OpenAI's board.
The other piece here is that the advent for a non-profit, open source AI company is about as dead as dead can be. We increasingly need to come to terms that the most powerful technology that will likely be developed by human kind is being developed by a handful of people in 3 different silos.
Google.
OpenAI/Microsoft.
Elon Musk/Tesla/xAI.
How these people decide to leverage said technologies will have a GIGANTIC impact on the future of civilization.
Wild stuff. I'd love to get your take in all of this.
Karsten Klees retweeted
Richte dir eine Erinnerung für meinen kommenden Space ein! teapawt.net/i/spaces/1jMJgmANkYPKL
Karsten Klees retweeted
Tesla’s Monopoly Inches Closer.
New industry data is showing that Tesla is trending towards a Monopoly. As an example, Ford is losing $1.6B per quarter on EVs, and Lucid and Rivian, both of which are EV-only start ups, are close to running out of cash and have effectively stopped growing. These results prove that Tesla has a giant lead in making electric vehicles at scale that are both affordable AND profitable, with the kicker being these same EVs could become next-generation transportation units with the advent of AI-powered self-driving technology.
This transition is slated to have the same impact to the car industry that the iPhone had on mobile telecommunications and computing - yet most are still dismissing that this is a possibility.
Let me walk you through why most people are in for a giant surprise as we reach the end of this decade, and it all starts with Tesla’s ambitions in entirely rethinking what a car should be, versus what it is today.
Tesla believes it can get many of the cars in its over 5 million unit fleet, and all new cars being sold today, to become self-driving “Robotaxis” by leveraging the existing 8-camera + on board computer hardware suite that is shipped with every Tesla car today. A software update, potentially with its v12 iteration, will upgrade these cars from manually driven cars to cars that can drive themselves without the need of an attentive driver. The company is beginning to trial this software with its customer base, hoping to have a finalized version by the end of the year.
On top of this, Tesla has been on a mission to dramatically bring down the cost and affordability of an electric vehicle. Thus far, this strategy has allowed Tesla to have the best selling car in the world with its Model Y. With EV tax credits in the US, the Model Y starts at $35k, which is $10k less than the average new car price in the US. This is quite an achievement, especially since Tesla unveiled its Model 3 sedan in 2016, which was slated to launch with a $35k starting price. 8 years later, Tesla is offering the same starting price, with EV incentives, in an SUV form factor and 8 years worth of inflation.
Even with30% inflation over the last 8 years, and a larger form factor, Tesla still makes upwards of 20% margin on its EVs, while offering some of the most affordable EVs on the market. Both the Model Y and the Model 3 can be leased for under $400 per month. In that same period, gas cars are becoming more expensive, as legacy automakers continue to prioritize profits over affordability. EV prices are coming down, while gas car prices are going up. Not a good sign for the gas car industry.
This means that the strategy Tesla is deploying of having affordable, profitable, and self-driving EVs that will eventually become the next generation of transportation, which many have thought is overly ambitious and downright pie-in-the-sky thinking, is beginning to bear fruit.
This becomes quite obvious once we start looking at Tesla’s competition, especially in the US.
Starting with Legacy Automakers, we are already well aware that these companies are pausing, delaying, or canceling their EV plans for the foreseeable future, including brands like Ford, GM, VW, and others.
Once we look at their financials, it’s clear to see why. If we take Ford’s Q4 2023 earnings results, their EV division has stalled at 34,000 units sold per quarter for the last 3 quarters, while their average selling prices have dropped by 12% and their losses increased by 50%, from $1.1B lost per quarter, to $1.6B. On dealer lots, Ford EVs have days of inventory of well over 100 days. Dealers ideally want to have inventories of 60 days or less. The story is the same with other legacy automakers and their EVs.
On the other hand, if we look at Tesla’s results, the company was able to grow their sales over the same 3 quarter period by 5%, while generating $2B of income in the fourth quarter. While Ford lost $1.6B from selling 34,000 EVs, Tesla made $2B from selling 484,000 EVs. Again, the story is the same across every legacy automaker.
But this continues to get more grim for Tesla’s competitors if we look at EV startups like Lucid and Rivian.
For the fourth quarter of 2023, Rivian reported deliveries of 13,972 which is a 10% increase over the same period, but loss per car accelerated to negative -$113k per car. That means that for every car Rivian sold in Q4, they lost about $113k. The story gets worse when we look at Lucid, with deliveries of 1,734 in the same quarter, but losing about -$377k per car.
In either case, Rivian and Lucid are both guiding to flat sales in 2024, which means that the levers they need to pull to lower costs and become profitable will not be available to them for the next 12 months. Even though Rivian in their Q4 filing said they are looking to reduce costs by as much as 50%, it will be nearly impossible without increasing volumes.
Auto manufacturing is heavily dependent on scale. If Rivian is already losing more than $100k per car, and they are not expecting to grow much in 2024, then they will not be able to spread out their costs over more cars in order to increase their profits per car. The company is already planning to lay off 10% of its staff to try and reduce costs, but it is difficult to see how this will be a big enough step in order for the company to become profitable.
Another issue arises when we look at how much cash both of these companies have on hand. Rivian has about $8B of cash on hand, while Lucid has a little under $1.4B. At current cash burn rates, Rivian has about a year and a half of cash left, while Lucid has roughly 6 months. Without raising money for their operations, both companies will be forced to either liquidate their assets, dramatically slash their production capacity (further increasing their cost per car), or file for bankruptcy.
In Rivian’s case, it could be likely that they are able to raise cash in order to make it to the release of their next generation, more affordable vehicle, while Lucid’s only real hope is to be taken private by the Saudi fund that owns 60% of the company. To be clear, I’m a big fan of both company’s products, but if you can’t make money selling stuff, you’re not going to be around for very long.
But in all cases above, the biggest issue facing all these automakers is that their scale of production for electric vehicles is not quite there yet. Tesla is making about half a million cars per quarter, while its competition is making less than 10% of that amount.
While almost all of Tesla’s competitors are struggling to enter the EV space, Tesla has reached a 2 million per year run rate of EVs with only 4 models - the Model S, 3, X, and Y. They’ve been able to achieve this with little to no advertising, and up until very recently, with above average prices relative to the broader markets.
A notable exception here is BYD, which has surpassed Tesla in pure EV sales per quarter. However, it is important to note that BYD primarily sells its cars in the Chinese market, sells its cars at a much lower price point than Tesla, and has yet to prove it can make a profit on its EV line up. However, because of the company’s ability to offer EVs at very affordable prices (the company’s Seagull, its compact car, sells at about $11k US in China), BYD has already taken massive market share away from Legacy car companies, especially players like Volkswagen.
What’s of note here is that Tesla is selling about the same number of EVs per quarter as BYD, but without an affordable compact car, and up until very recently, its Cybertruck pick up. However, for Tesla’s ability to reach Monopoly status, it is much less about the Cybertruck and much more about the compact car.
Based on Tesla’s guidance, they are aiming to build an EV platform that reduces costs by as much as 50% by revolutionizing the manufacturing process. This should allow the company to build an EV that’s somewhere around the $25k starting price, before any federal or state EV incentives, that will be sold in the millions of units per year, with the company having a long-term target to reach 20 million cars sold per year.
This means that for the first time in a long time, a US buyer could have access to a brand new car, that happens to be electric, that is under $25k with loads of software-driven features, high safety scores, and driving dynamics that are far superior to comparable gas vehicles due to its EV drivetrain.
Even with this product in the wings, the biggest piece that will virtually secure Tesla’s monopoly is not solely due to its ability to make a very affordable car - but make a very affordable car that can drive itself.
If we look at the auto landscape, there are no players that exist that are solving for self-driving technology that are low cost and high scale. With BYD’s claim in August of 2023 that full self-driving is essentially “impossible”, we are only left with players, outside of Tesla, that are solving for self-driving technology with very expensive hardware. Players like Waymo, Baidu, Zoox, and others rely heavily on LIDAR, Ultrasonics, and a bunch of other sensors that very rapidly increase the cost of admission for self-driving manufacturing.
As impressive as these technologies are, the cost of manufacturing a self-driving car with these sensors runs somewhere in the six digit range, with estimates varying between $120k per car, upwards of $300k. This is because the companies have decided to solve self-driving cars with brute force - throw every sensor you can at the problem, map every road that the cars operate in, and figure out how to make it all work without running over people. Miraculously, many have figured out how to make this work, but at the expense of scale.
It is MUCH more difficult to build many self-driving cars that are this expensive vs a company that can make the same car for under $20k, including all the hardware that’s needed to operate said cars. In addition, the roll out of these vehicles becomes increasingly difficult as every road that’s meant for these cars to operate in has to be mapped, and every little change has to be recorded so the cars don’t accidentally end up in ditches.
This approach sets an artificial cap on self-driving technology. Regardless of how impressive the technology is, if it’s too expensive to make, and too cumbersome to roll out, it is not a viable business in the long term. The breakthrough needed is not the self-driving technology itself - but the manufacturing system that allows you to make self-driving cars that can be a viable business in the long term.
This is Tesla’s bread and butter - creating manufacturing systems that convert state-of-the-art technologies into mass-market realities. And this is precisely what Tesla is achieving with the development of its Full Self Driving technology and its compact car.
The previous 4 models, the S, 3, X, and Y, have allowed Tesla to build up enough data and expertise to make an 8 camera + onboard computer system. This likely costs no more than $2k all in, and the kicker is it’ll drive just as good as a Waymo car at a fraction of the cost and at 1,000x the scale.
If we fast forward to the end of the decade, and we assume that Tesla’s execution has allowed them to achieve autonomy at the level of a Waymo (or higher), which they can then beam to a fleet of at least 30 million cars, Tesla will easily have the most autonomous miles per year out of anyone in the world by a factor of at least 1,000. And it is the marriage of affordable EVs, that happen to drive themselves, that will allow the company to achieve this.
For a competitor to get remotely close, they would have to do the following:
- Come up with a self-driving system that will cost around $2k
- Put these in cars that cost no more than $20k to produce
- Make millions of these per year
- Make them profitably
Thus far, we have little data that points to any other player being able to make an EV profitably, let alone millions of them per year. The only player that can make many of them, BYD, has concluded that self-driving is impossible. In the meantime, Tesla has launched its v12 self-driving software that is navigating US roads at the level of a Waymo car. Tesla achieved this with about 1 year of development by offloading the code to an AI instead of a team of humans.
As Tesla’s data collecting capacity grows as more and more Teslas populate the world, and as they ramp up their compute capability at its headquarters, the performance of their self-driving vehicles will only get better from here on out.
Given all of this, is it really that unlikely that Tesla will have a Monopoly on transportation? Especially when:
- They are the only player that can make self-driving cars profitably in the millions of units and…
- Because their self-driving cars don’t need a human to drive them, which will remove the most expensive thing from a car today, making per-mile cost of transportation plummet to pennies on the dollar.
The clue here is that Tesla is already beginning to offer licensing partnerships for its self-driving system. The company is clearly confident that they have what it takes to solve for self-driving in a very cost-effective manner. No one is taking them up on that offer, probably because most players are skeptical something like this is even possible. It surely sounds like pie-in-the-sky when a company claims that they’ve figured out how to make cars drive themselves at a fraction of the cost than anyone else can.
Time will tell who is right. I’m willing to bet on the company that has continually achieved feats that most thought impossible. This company probably knows a thing or two about making impossible things a reality. And given the current trajectory of development with its self-driving software and ability to make EVs affordable AND profitable, my guess is that they are much closer to that reality than most realize.
Karsten Klees retweeted
Another excellent (if long!) post by Farzad. The driver/passenger dichotomy is critical to understanding how critically we judge our Teslas vs how we judge something like an Uber or Waymo. FSD 11 is not there yet, but it really looks like v12 might be. Time will tell, but as Farzad says, likely it will no longer be _that_ much time.
Why self driving cars will be everywhere quicker than most expect.
I’ve had a theory about self-driving technology ever since I experienced Tesla’s self-driving software suite (FSD), especially starting with v11. Let me walk you through how I think about this by first giving some context.
Tesla FSD v11 was the first version from Tesla where I had first-hand experience that the company was on track to solve for self-driving technology. It solidly moved from “if” to “when”, mainly for two reasons:
It became obvious (to me at least) that the car was able to “see” everything that was relevant that was happening on the road, which means it is not a hardware problem, but a software problem. The same exact places where the car previously failed massively all of sudden were handled flawlessly.
I noticed that being a passenger while the system was on is a materially different experience than being behind the wheel of a self-driving car, which made me realize that once you reach a certain threshold, solving self-driving for a passenger is materially different than solving for a “driver”.
A lot of the doubt has traditionally arised from Tesla’s approach is mainly due to their choice in hardware. The thought process is that Tesla cars will be able to drive themselves with 8 cameras and a computer, essentially using the same inputs that humans use to drive - two eyes, a neck and body that allows us to move our cameras from side to side (and backwards), and a brain that processes the data we see, which then tells our arms and legs what to do in response to what we see.
Tesla thinks it can get there by putting cameras around the car that simulates our ability to look around, an advanced computer in the car to process the information, and an AI brain at Tesla’s mothership that makes sense of the inputs and subsequently generates the outputs (ie throttling and braking, steering input, etc.) This “AI brain” was largely human code with v11 and previous versions, meaning that humans would need to figure out all the “code” that would be needed for the car to drive itself everywhere. Starting with v12, Tesla has offloaded that work to a neutral net - think of it as a ChatGPT but for driving. The additional bonus here is that Tesla cars require no mapping to understand the roads - they simply look at it and navigate as a regular car would by taking cues from the road and by using navigation maps.
Players like Waymo and others are solving for this equation very differently - they are going all out on sensors (LIDAR, radar, cameras, ultrasonics, etc.) and are generating high-definition maps of the cities that they operate in. This has allowed players to start offering self-driving taxi drives, which is amazing in its own right and should not be underestimated, but at the cost of scale. It is significantly harder and more expensive to scale a technology that requires heavy up front investments in both equipment and mapping. Waymo’s self-driving cars cost somewhere around $250,000 mark. On the flip side, Tesla’s Model 3, which the company thinks has enough hardware to solve for self-driving cars, starts at $39,000. That’s more than a 6x cost difference.
Essentially, Tesla is betting that 8 cameras and a computer will be enough hardware to drive, but it requires way more time and effort to get it to do what Waymo does today because of that hardware and mapping “limitation”. Waymo has WAY more hardware and resources to understand the world, but it’s way tougher to scale the product when you’ve spent 6x more on equipment and need to map every road that exists in the world. In addition, the economics of a Taxi that costs $250k becomes harder to justify than one that costs $40k and could theoretically drive on every road, vs the roads that Waymo has mapped.
If the car can drive itself, but it can’t make enough money to pay for itself, the technology fails regardless of how good it is. This is Waymo’s core struggle.
If Tesla achieves the same level of driving capability as a Waymo car, then in theory, Tesla can upload the software to its 5 million car fleet, allowing Tesla to have a Waymo on every road (that allows it) with 1/6th the cost of Waymo from an equipment standpoint, and 0 cost or effort as it pertains to mapping the roads.
However, the biggest hurdles Tesla has faced with FSD, especially as it pertains to adoption, are the following:
- The car can oftentimes behave in an “inhuman” way - takes a bit too long to cross an intersection, turns feel “unnatural”, decision making around traffic is oftentimes confusing, lane selection is incorrect, etc.
- The car will respond in a materially different way in a perceived “safety critical” situation. In general, people that pay attention will take extra precautions when they face a “weird” situation on the road that might put their safety at risk. A Tesla on FSD can handle these very differently than a human, even though the outcome may be the same (avoid a crash). The challenge is the person can sometimes have little confidence that the car will actually do the thing it’s supposed to do to avoid a crash. This dramatically raises anxiety and stress, which leads to people feeling like they can’t trust the software (understandably so).
- The current up-front price of the software ($12k) is a steep price to pay when the above are present. “Why should I pay that much when I can drive the car better than it can?”
And therein lies the challenge.
I think this is largely a psychological exercise. What we may deem as “better”, within the context of getting from point A to point B, is irrelevant to a computer system. A self-driving computer’s ultimate goal is to get you from point A to point B safely. However, this doesn’t mean comfort in getting from point A to point B is not a big deal - it is the biggest deal and the largest variable to the adoption of self-driving cars. But this equation changes as you move from a “driver” to a “passenger” in a self-driving car.
This is where Tesla’s v12 update comes in, which the company thinks will achieve “Waymo” levels of comfort and confidence, with a fraction of the cost, and the ability to handle all roads without mapping data.
Now, for this to be the case, v12 has to achieve the following:
- It needs to be vastly safer than a human to remove any doubt that the system is safe (10x to 100x safer).
- It needs to give confidence to the “driver” and passengers that the system a) perceives any and all potential dangers on the road and b) acts in a way that measures the people in the car that the car has “got it”. Tesla needs to prove that this is not a hardware limitation, but it’s limited by the car making a decision on what it sees.
- It chooses a route that is acceptable to the people in the car - ie get people to their location by the time you are estimating.
- For wide-scale Robotaxi technology (ie a fleet of driverless cars), Tesla needs to operate these at a much lower cost than its competition (Uber, etc.). Without needing a driver, this should be very easily done.
Now, why do I think this technology will be here much sooner than people expect? Two reasons - computers and AI are vastly better than humans at solving narrow problems, especially if they are based in mathematics and physics, and the experience of being a passenger in a transportation system is vastly different than being in control.
Let’s start with the AI piece by looking at the progress we’ve made with computing for the better part of 3 decades, and the recent advancements in AI.
Computing has made non-creative tasks incredibly easy to solve. Imagine a medium to large corporation existing without excel in this day and age. On the other hand, AI and advanced algorithms are making esoteric problems solvable at a rate that we previously thought impossible.
It started with chess computers beating the best chess players in the world. Now, we have AIs like ChatGPT, MidJourney, and ElevenLabs capable of generating solutions for problems we thought computers would never be able to solve. “Hey ChatGPT, can you write me a Python code for blah?” “Oh yeah and by the way, can you give me a 1,000 summary of how Ghengis Khan conquered the world?” You can ask the same tool this question, and it’ll generate a response with high levels of accuracy.
In addition, as a quick anecdote, I’ve been using ElevenLabs to generate an AI model of my voice and using it on my content. Upon asking my audience if the voice I used for the video is “real” or “AI”, the audience will consistently get them confused for each other. We have reached a point that with enough training data, an AI can simulate my voice, with its quirks and everything, to a point where it’s indistinguishable from my real voice.
And it’s this data piece that Tesla has a massive advantage over everyone else. The video data that Tesla has been able to gather from its fleet of 5.5 million cars (and growing) is what will allow it to solve for self-driving in the same way ChatGPT solved for general-use AI, MidJourney solved for image creation, and ElevenLabs solved for voice creation. It’s a matter of having the right hardware to ingest the data, collecting the data, having the hardware to process it, and having an economic model where it makes sense to “put the data to work”.
If we take Elon Musk’s initial v12 ride from 2023 where he live streamed it to his X audience, and if we take the initial feedback from users that have been using v12 in the wild, the system is already driving better in many instances with about 1 year of development, and the lowest amount of training data and training hardware it’ll have at its disposal, than the previous 6 years of development under v11 and older versions.
In this respect, it appears that Tesla has put in place the necessary variables to finalize its system, which in my estimation will be proved out as we get further into 2024. Time will tell if this is correct.
The second piece is one that I think many folks are overlooking - the experience of being a passenger vs a “driver” in a self-driving vehicle.
If we use Waymo as an example - Waymo requires 0 inputs from the passengers, even though there’s a steering wheel and pedals in the car. In this sense, someone that gets in the Waymo, they have 0 need to a) pay attention to their surroundings or b) be prepared to take over at any point. By default, knowing that this is the case, a person in a Waymo rides in it in the same “mode” they would as if they’re riding in a bus, plane, or an Uber. In a large % of cases, that person will be on their phone, laptop, sleeping, listening to music, or chatting with someone else in the car. What is happening with the car and outside of the car becomes secondary.
On the flip side, with Tesla’s FSD software, there always needs to be an attentive driver ready to take over at any time. This means that by default, the same person that would traditionally be a passenger in a Waymo is a driver in the Tesla in a scenario where the Tesla is driving itself.
Because of this, the level that Tesla has to reach in order to be “acceptable” as a self-driving system is significantly more stringent than a system that is already self-driving. I’ll use myself as an example to prove this point.
When I’m sitting behind a Tesla while self-driving is on, I feel responsible for the actions and maneuvers that the car is doing at any given time. I am constantly judging the performance of the car as if I’m the one who’s driving it. “I would’ve taken a different angle on that turn”. “I would’ve cleared the intersection a little bit more aggressively”. “I would’ve moved to that lane a little bit sooner”. “I wouldn’t have been so hesitant exiting the intersection”. The list goes on.
I’ve made countless drives where the car has taken me from point a to point b, with no interactions from me with the car, where I would’ve made different decisions in getting us there. But the car still got us there with v11 software. The experience is very similar to riding in a car where the driver and passenger both have steering wheels and pedals, and you trust the other person to “make the right decisions”, but you have the power to take over at any time. In the case of Tesla FSD, the system will default to your input 100% of the time. I am terrified to think what it would be like if it was with another person.
And it is this equation that makes the Tesla piece so intriguing - if Tesla is able to achieve “Waymo level” of driving comfort and input with v12, by default, the system is already better than a Waymo because it has cleared a higher bar of acceptability. It has reached “Waymo” level of comfort as a driver, not a passenger. As a passenger, by default, you are less likely to be in a mindset to judge every decision of the car.
And if this is true, Tesla will be able to beam “Waymo level” performance and comfort to millions of cars on the road, which will put it at an insurmountable advantage vs its competitors due to its decision to go with a significantly lower cost, high scale option.
If we look back in history, some of the most memorable and life-altering technological breakthroughs kind of came out of nowhere - at least in the public’s eyes.
A computer beating a Chess Grandmaster at Chess, and a world-class player at Go, proving that high-powered computers can solve even the most difficult human problems.
The internet, ecommerce, and the subsequent birth of the largest companies in the world that are entirely enabled because of the existence of such technology.
AIs based on LLM technology like ChatGPT, and AI tools like MidJourney and ElevenLabs, which made creative work that was initially thought as entirely human into something that can be infinitely iterated and altered (if you have enough money to spend on tokens lol)
And now, with the advent of self-driving technology, my gut tells me that we’re approaching the same level of sudden breakthrough. You have to give props to companies like Waymo, which have figured out how to get a car to drive itself in certain locations without the luxury of having AI at its disposal. A group of folks were able to get together, identify the correct marriage of hardware and software, put in a ton of elbow grease, and made an insanely impressive system that will get people from point A to point B safely and comfortably.
The unfortunate piece here is that because of its complexity, for a company like Waymo to be successful long term, it needs to scale to a point where it can operate anywhere where there is a road. However, the scale will be supremely difficult and expensive to achieve. In a world where a company like Tesla can’t figure out how to scale self-driving technology with a fraction of the cost and no need for mapping, self-driving technology will not be economically viable in almost all use cases.
Given the arc of human technological improvement, this outcome doesn’t seem likely. Instead, the likeliest outcome is for a company to solve self-driving, at a scale previously thought of as “impossible” due to preconceived limitations, and doing so kind of “out of nowhere”.
For a lot of people that follow the space closely, 2024 is the year that will prove if this statement is correct. Will we finally have self-driving cars that can drive anywhere, or will it always be something that is right there, but completely out of reach?
Being an optimist, it’s difficult for me to accept that this technology is not solvable today. I personally think Tesla has the right pieces in place, and in time this will be proven true. And if not, y’all have a post to use for the rest of the decade to dunk on me any time you see fit :)
Karsten Klees retweeted
Richte dir eine Erinnerung für meinen kommenden Space ein! teapawt.net/i/spaces/1DXxyjloewyKM








































