Latest / eMotors: Electric Revolution / Musk's Deceptive Claims on Tesla Full Self-Driving
Transcript
- 0:09Welcome back to the Deep Dive. Today we are strapping in for, I
- 0:13think, a very necessary reality check.
- 0:17Our mission is pretty direct. We're going to unpack the
- 0:20claims, the performance, and most importantly, the data
- 0:24around Tesla's Full Self Driving, or FSD.
- 0:27And the whole robo taxi network that's always just around the
- 0:29corner. Exactly.
- 0:30We're diving deep into the gap. And believe me, it's a it's a
- 0:34huge consistent gap between the high octane rhetoric and the,
- 0:38you know, the technical reality of what's actually on the road
- 0:41today. OK, let's get right into that
- 0:42central issue, because this isn't a new story.
- 0:46It feels like a repeating loop. It is.
- 0:48For years now, I mean going back to at least 2018, the promise
- 0:52has been that Tesla would solve full self driving quote by the
- 0:55end of the year. Or next year for sure.
- 0:57Right, next year and that timeline just keeps resetting
- 1:00and that hurdle has well, it's never been cleared.
- 1:02Which is why we have to establish the baseline right now
- 1:05for everyone listening. Despite the marketing, FSD and
- 1:08Autopilot are officially classified as SAE Level 2.
- 1:11Level 2. Advanced driver assistance
- 1:13systems, Yes. This is not autonomy.
- 1:15It's a system that requires, you know, constant human
- 1:19supervision. It's nowhere near the level 5
- 1:22the truly get in and go to sleep autonomy that gets predicted.
- 1:26And that distinction, level 2 supervise versus level 5 totally
- 1:30autonomous, that is the perfect pivot to our first section.
- 1:34The history. The history of missed deadlines.
- 1:36Because look, it's one thing to be optimistic about a complex
- 1:39problem. It is something else entirely to
- 1:41be this persistently wrong for over a decade.
- 1:44What's really fascinating here is just how far back this
- 1:47pattern goes. I mean, way back in 2013 we saw
- 1:50the first big prediction. What was that one again?
- 1:52That 90% of miles driven would be autonomous within three
- 1:55years. So by 2016, Nope.
- 1:57Didn't happen, not even close. Then we got the famous one.
- 2:00October 2016, The coast to coast demonstration.
- 2:03Ah. Yes, the plan to go from LA to
- 2:06New York City without a single human touch?
- 2:08That was supposed to happen by the end of 2017.
- 2:10Another mess. A big mess.
- 2:12But maybe the most memorable promise, one everyone talks
- 2:15about, came in April 2019. The robotaxis.
- 2:18The declaration that there would be over 1,000,000 autonomous
- 2:21robotaxis on the road by 2020, that was just a monumental
- 2:26prediction. And of course, it never
- 2:27materialized. Not.
- 2:28At all. I mean, at some point you have
- 2:30to ask, doesn't the sheer weight of these missed goals hurt
- 2:33confidence even among the biggest supporters?
- 2:36Well, it raises a really important question about intent.
- 2:39And even recently this isn't just ancient history.
- 2:42In July 2023, there was this public acknowledgement.
- 2:46Oh wait, remember this? Musk admitted his predictions
- 2:48had been, you know, optimistic. He even compared himself to
- 2:52quote the boy who cried FSD. But then what did he do
- 2:56immediately after? He immediately followed it up by
- 2:58predicting better than human autonomy by the end of that same
- 3:01year, 2023, which again was missed and this.
- 3:04Whole history of these, let's call them extremely positive
- 3:08statements. It eventually ended up in court.
- 3:10It did. That's a vital part of the
- 3:11story. In February 2023, a lawsuit was
- 3:15filed by investors. They alleged securities fraud
- 3:18over these misleading statements.
- 3:19And what? Happened with that.
- 3:20It was actually dismissed in September 2024.
- 3:24The judge ruled Musk's statements were, and this is the
- 3:26key phrase, corporate puffery. Corporate puffery.
- 3:30So wait, the judge is basically saying it's OK for a CEO to make
- 3:33promises about technology that don't come true as long as
- 3:36they're kind of vague and optimistic?
- 3:38That's essentially it, yes. The ruling saw them as vague
- 3:41corporate optimism, not as concrete technical claims
- 3:44designed to defraud. It set a really fascinating
- 3:47legal precedent. OK, so if the history of FSD is
- 3:50defined by corporate puffery, then the Austin Robotaxi launch
- 3:55feels like it's perfect expression.
- 3:56Let's get into Section 2 and look at what that actually is.
- 3:59This is where we see the definition of robotaxi just
- 4:02stretched, you know, past its breaking point to fit a
- 4:04narrative. It's misleading.
- 4:06From what our sources say, the service has only a few vehicles,
- 4:09maybe around 30. And they only operate in a tiny
- 4:12pre mapped area, a geofence. And, and this is the crucial
- 4:15part, it relies entirely on in car supervisors.
- 4:18These are employees with a kill switch ready to take over.
- 4:21And that supervision factor creates this massive
- 4:25contradiction, especially when Musk criticizes competitors.
- 4:29He actually called Waymo's fleet size Ricky numbers.
- 4:32Right. But let's just look at what we
- 4:34MO is actually doing. They have about 2500 fully
- 4:37autonomous vehicles. And crucially, without employees
- 4:40sitting inside them. It's true, unsupervised level 4
- 4:44operations in multiple cities. You compare that to Tesla's,
- 4:48what, 30 or so supervised cars in Austin?
- 4:51It's not even the same league. And then there's the Bay Area,
- 4:54where Tesla says they have over 100 cars running.
- 4:57But it's not an autonomous service there.
- 4:58No, not officially. It's just a ride hailing
- 5:01service. The drivers have to be in the
- 5:02driver's seat, and Tesla hasn't even applied for a full
- 5:05autonomous driving permit in California.
- 5:07OK, but here is where it gets. I mean truly shocking according
- 5:11to the source material. The safety numbers.
- 5:13Oh, this is a staggering data point.
- 5:16Tesla's robo taxi crash rate, even with those human
- 5:19supervisors ready to intervene, is almost double waymo's.
- 5:22Double the crash rate of Waymo, which is running completely
- 5:25unsupervised. How is that even possible if
- 5:28your supervised system is crashing twice as often as a
- 5:31competitor's unsupervised one? I mean, the whole argument for
- 5:34your tech leadership just evaporates.
- 5:36They completely dissolve. Twice the crash rate with a
- 5:39human safety net that just destroys the entire premise that
- 5:43FSD is fundamentally safer. It does, and this this high rate
- 5:47of failure under supervision, it leads us to wonder how they
- 5:50might try to spin this. The source suggests Musk might
- 5:53just try to change the definition of robotaxi.
- 5:56What would that even look like? A fuzzy definition.
- 5:59It means maybe hiring thousands of Uber style drivers, putting
- 6:02them in FSD enabled cars, and then just calling them
- 6:05robotaxis. Or they might start counting
- 6:07every single consumer who pays for the FSD beta as part of
- 6:11their robotaxi network. It completely blurs the line
- 6:14between a real autonomous service and just a driver using
- 6:18a level 2 system. So they could inflate their
- 6:20fleet numbers into the hundreds of thousands while offering zero
- 6:24true unsupervised rides that move using a questionable
- 6:28methodology to claim you're winning.
- 6:30That is the perfect segue to our third section.
- 6:33Dissecting the safety data. The data that Tesla itself
- 6:36reports, they always say that FSD is safer than humans, but
- 6:40what's wrong with the numbers they use to back that up?
- 6:42This is so important for critical thinking.
- 6:45When you hear that claim, you need to understand there are
- 6:48three major, I would even say fatal flaws in their
- 6:52methodology. OK, let's break it down.
- 6:54What's the first flaw that just undermines the whole thing?
- 6:56The first is how they even count a crash.
- 6:59It's entirely self reported and they only count an accident if
- 7:02an airbag or a restraint system goes off.
- 7:05Wait, so every minor Fender Bender, every slow speed bump?
- 7:08Excluded if FSD is engaged and causes a minor collision but no
- 7:12airbag deploys, it's not factored into the public crash
- 7:15rate. It's effectively invisible.
- 7:17So the data set is incomplete from the start.
- 7:20What's the second big problem? That's the road type bias.
- 7:23This is a huge one. Historically, autopilot is used
- 7:26mostly on limited access highways.
- 7:28The safest roads? Statistically the safest roads
- 7:31in the country. But the federal crash data they
- 7:34compare themselves to includes all roads, dangerous rural
- 7:37roads, complex city streets, everything.
- 7:39You're not comparing apples to apples, you're comparing the
- 7:42easiest driving scenario to the average of all driving
- 7:46scenarios. Precisely.
- 7:47And the third issue is what we call driver mix or demographic
- 7:50bias. Meaning the Tesla drivers
- 7:52themselves. They tend to be in newer cars,
- 7:54have higher incomes, be tech enthusiasts.
- 7:57These demographics already crash less than the general
- 8:00population, so they're human driving baseline is already
- 8:04better than the national average they're comparing against.
- 8:07So even the human data in the comparison is skewed in their
- 8:10favor. OK, beyond those big three,
- 8:12there's another one, right? The intervention bias.
- 8:15Absolutely. The human in the car is a safety
- 8:17filter. How so?
- 8:18FSD drivers. You know they often choose not
- 8:21to use the system in tricky situations.
- 8:23Heavy fog, snow, complex construction zones, right?
- 8:27You wouldn't risk it. You wouldn't.
- 8:28So those difficult scenarios that lead to crashes for human
- 8:31drivers, they get counted in the human data, but FSD users avoid
- 8:35them, insulating the system from failure and making its data look
- 8:39artificially clean. So if we just for a second set
- 8:43aside all those biases and just accept the data, there's still a
- 8:46core logical error here, isn't there?
- 8:48There is, and it's the most important one.
- 8:51The data, even if it were perfect, should only claim that
- 8:54FSD with human supervision is safer than a human alone.
- 8:58It's not FSD versus human. Is FSD plus a human versus a
- 9:01human? To claim anything else is just a
- 9:03fundamental misrepresentation of what's being tested.
- 9:06Which brings us back to that core question from our sources.
- 9:08Yeah. Is the company leadership
- 9:10genuinely delusional, always thinking they're just about to
- 9:14solve it? Or are they intentionally
- 9:16misleading people? The synthesis from the source
- 9:19suggests it's it's probably a mix of both and it shifts.
- 9:23There's clearly a deep hyper optimistic belief there, but the
- 9:26calculated choices to obscure the truth with flawed metrics to
- 9:30redefine words like robotaxi, that strongly suggests a very
- 9:34powerful motivation to maintain Tesla's public image as the
- 9:37leader. And for you listening to this,
- 9:40understanding these biases is just essential for thinking
- 9:43critically. When you hear a system is safer
- 9:45than a human, you now know the three or four questions you have
- 9:49to ask about the data. And here's the final thought to
- 9:52leave you with one that really captures this whole gap between
- 9:55promise and reality. Yeah, OK.
- 9:56We learned that in June of 2025, Tesla successfully did a fully
- 10:01autonomous drive. No human on board, no remote
- 10:04pilot from the factory to a customer's house.
- 10:07Wow, a true level four or five moment, even if it was just a
- 10:11one off. A brief moment of success, yes,
- 10:13But here's the kicker. The source says that upon
- 10:16delivery, the software on that car was immediately rolled back
- 10:19to the standard supervised FSD version.
- 10:21No way, yes. So that raises the ultimate
- 10:24question. If true, full autonomy can only
- 10:26be deployed for a publicity stunt before needing a human
- 10:29supervisor again. What does that really tell you
- 10:31about its stability and readiness for widespread,
- 10:33unsupervised use? That delta between those two
- 10:37software versions? That's where the real deep dive
- 10:39continues.