Walter Isaacson’s Elon Musk biography is set to be published on Tuesday, and a new preview of the book illustrates details about Tesla’s development of the upcoming Full Self-Driving (FSD) version 12.
In an additional preview of his Musk biography for CNBC, Isaacson discusses the use of AI in the development of Tesla’s FSD v12, in a shift that took place within the last several months. Isaacson talks about Tesla’s recent development of the upcoming FSD v12, which he and Tesla demonstrate has moved away from a “rules-based” approach.
Notably, FSD v12 is expected to use billions of video frames from real-world driving incidents to train its neural network rather than using thousands of lines of code like previous versions. In a conversation with Musk last December, Tesla Autopilot employee Dhaval Shroff had likened the concept to the popular chatbot ChatGPT, instead for use with driving.
“It’s like ChatGPT, but for cars,” Shroff said. “We process an enormous amount of data on how real human drivers acted in a complex driving situation, and then we train a computer’s neural network to mimic that.”
Surprisingly enough, Tesla only shifted toward this “neural network planner” approach recently. By the beginning of this year, however, the neural network had already analyzed 10 million video clips based on the best-case-scenario drivers the system had access to. Musk instructed employees at the company’s Buffalo, New York facility who were in charge of analyzing the footage to train the AI on things “a five-star Uber driver would do.”
Moving from a rules-based to a network-path-based AI approach allowed FSD to use human driving data to avoid obstacles, even if breaking some rules was necessary. Shroff helped demonstrate the idea to Musk with a demo featuring trash bins, debris, and upturned traffic cones, which the car handled surprisingly well.
“Here’s what happens when we move from rules-based to network-path-based,” Shroff explained. “The car will never get into a collision if you turn this thing on, even in unstructured environments.”
Musk quickly took to the idea, as can be seen in a recent livestream of Tesla’s FSD v12 software in Palo Alto with Autopilot software director Ashok Elluswamy. He has repeatedly spoken about the upcoming software version’s impressive driving results, despite one small moment in the drive where the car almost ran a red light.
In any case, Musk could argue that the red-light moment is a good case for the need for self-driving software to continually learn. Given that it will constantly be trained from the video data generated by camera footage from real-world drivers, it should theoretically make it safer over time, according to Musk.
During development, Musk also reportedly latched onto the fact that it took over a million video clips for the neural network to begin performing well, though he looks forward to what significantly more data will do for FSD.
Still, critics and regulators have expressed concerns about the faults of human drivers training AI-based driving systems, and Tesla has repeatedly been questioned by the National Highway Traffic Safety Administration (NHTSA) about its Autopilot and FSD beta systems.
According to Isaacson, Tesla plans to release FSD v12 as soon as regulators approve it. Meanwhile, an ongoing study by the National Highway Safety Board is looking to determine if self-driving cars should be permitted to imitate human driving actions that blur traffic rules, such as creeping up at stop signs.
Musk said in April that he expects Tesla to reach full autonomy within a year, though he has also been known to share ambitious targets for the software in the past.
You can read Walter Isaacson’s full account of the development of Tesla FSD v12 here, in a CNBC preview of the upcoming Elon Musk biography.
What are your thoughts? Let me know at zach@teslarati.com, find me on X at @zacharyvisconti, or send your tips to us at tips@teslarati.com.
News
Tesla Summon got insanely good in FSD v14.3.2 — Navigation? Not so much
There were two new lines of improvements in the release notes: one addressing Actually Smart Summon (ASS), and another that now allows drivers to choose a reason for an intervention via a small menu during disengagement.
Tesla Full Self-Driving v14.3.2 began rolling out to some owners earlier this week, and there are some notable improvements that came with this update.
There were two new lines of improvements in the release notes: one addressing Actually Smart Summon (ASS), and another that now allows drivers to choose a reason for an intervention via a small menu during disengagement.
Overall operation saw a handful of slight improvements, especially with parking performance, which has been the most notable difference with the arrival of FSD v14.3. However, there are still some very notable shortcomings, most notably with region-specific signage and navigation.
Tesla Assisted Smart Summon (ASS) improvements
There are noticeable improvements to ASS operation, which has definitely been inconsistent in terms of performance. Tesla wrote in the release notes for v14.3.2:
“Unified the model between Actually Smart Summon, FSD, and Robotaxi for more capable and reliable behavior.”
As recently as this month, I used Summon with no success. It had pulled around the parking lot I was in incorrectly, leaving the range at which Summon can be operated and losing a signal while moving in the middle of the lot.
This caused me to sprint across the lot to retrieve the vehicle:
It was pouring when I left the gym so I tried to Summon my Model Y
It turned the opposite way and drove out of range, stopping here and forcing me to walk even further across the lot in the rain for it 🤣
One day pic.twitter.com/iD10c8sriB
— TESLARATI (@Teslarati) April 5, 2026
Unfortunately, Summon was not dependable or accurate enough to use regularly. It appears Tesla might have bridged the gap needed to make it an effective feature, as two tests in parking lots proved that Summon was more responsive and faster to navigate to the location chosen.
It also did so without hesitation, confidently, and at a comfortable speed. I was able to test it twice at different distances:
🚨 Tesla FSD v14.3.2 ASS testing part 1
This was a significant improvement than recent tries using ASS. The parking lot was pretty empty but getting it to come to my location in one singular motion and maneuver was encouraging. https://t.co/vF7TS48GGV pic.twitter.com/sYt8tyHgNn
— TESLARATI (@Teslarati) April 23, 2026
Tesla Full Self-Driving v14.3.2 ASS testing part 2 https://t.co/lxfWfnLUxf pic.twitter.com/2R0r3ohI3M
— TESLARATI (@Teslarati) April 23, 2026
I plan to test this more thoroughly and regularly through the next few weeks, and I avoided using it in a congested parking lot initially because I have not had overwhelming success with Summon in the past. I wanted to set a low baseline for it to see if it could simply pull up to the place I pinned in the Tesla app.
It was two for two, which is a big improvement because I don’t think I ever had successful Summon attempts back-to-back. It just seems more confident than ever before.
New Disengagement Categories
This is a really good idea from Tesla, but there are some issues with it. The categories you can select are Critical, Comfort, Preference, and Other.
I think the reasons why people choose to take over would be a better way to prompt drivers, like, “Traveling Too Fast,” “Incorrect Maneuver,” “Navigation Error,” would be more beneficial.
I say this because it seems that how we each categorize things might be different. For example, I shared a video of an intervention because the car had navigated to an exit to a parking lot and put its left blinker on, despite left turns not being allowed there.
I disengaged and chose Critical as the reason; it’s not a comfort issue, it’s not a preference, it’s quite literally an illegal turn, and it’s also dangerous because it cuts across several lanes of traffic and is 180 degrees.
I chose to label this Navigation error as “Critical” while testing FSD v14.3.2
Here’s why:
✅ This intervention wasn’t “preference,” as the maneuver FSD routed was illegal
✅ If a police officer saw this maneuver, it would result in a ticket https://t.co/znhHb4haAo pic.twitter.com/bZOiLwWmQa— TESLARATI (@Teslarati) April 23, 2026
Some said I should not have labeled this as Critical, but that’s the description I best characterized the disengagement as.
Categorizing interventions is a good thing, but it’s kind of hard to determine how to label them correctly.
Inconsistency with Regional Traffic Patterns
Tesla Full Self-Driving is pretty inconsistent with how it handles regional or local traffic patterns and road rules. The most frequent example I like to use is that of the “Except Right Turn” stop sign, which has become a notorious sighting on our social media platforms.
In the initial rollout of v14.3, my Model Y successfully navigated through one of these stop signs with no issues. However, testing at two of these stop signs yesterday proved it is still not sure how to read signs and navigate through them properly.
🚨 Tesla FSD v14.3.2 attempts the “Except Right Turn” stop sign: https://t.co/W5MjAybaNK pic.twitter.com/P6oeUsk4PN
— TESLARATI (@Teslarati) April 23, 2026
Off camera, I approached another one of these signs and felt the car coming to a stop, so I nudged it forward with the accelerator pedal pressed.
This helped the car go through the sign without stopping, but I could feel the bucking of the vehicle as the car really wanted to stop.
Musk said on the earnings call earlier this week that unsupervised FSD would probably be available in some regions before others, including a state-to-state basis in the U.S.
“It’s difficult to release this like to everyone everywhere all at once because we do want to make sure that they’re not unique situations in a city that particularly complex intersection or — actually, they tend to be places where people get into accidents a lot because they’re just — perhaps there’s — and like I said, an unsafe intersection or bad road markings or a lot of weather challenges. So I think we would release unsupervised gradually to the customer fleet as we feel like a particular geography is confirmed to be safe.”
This could be one of those examples that Tesla just has to figure out.
Highway Operation
Full Self-Driving is already pretty good at routine roadway navigation, so I don’t have too much to report here.
However, I was happy with FSD’s decision-making at several points, including its choice not to pass a slightly slower car and remain in the right lane as we approached the off-ramp:
🚨 Tesla FSD v14.3.2 highway operation: generally happy with the performance here, especially behavior near the exit
Love that the car got over in the right lane after its final pass, and stayed there as the off ramp was approaching https://t.co/qVRVhg6XGR pic.twitter.com/1ELwHf2XKS
— TESLARATI (@Teslarati) April 23, 2026
Better Maneuvering at Stop Signs
Many FSD users report some strange operations at stop signs, especially four-way intersections where there is a stop sign and a line on the road, and they’re not even with one another.
I experienced this quite frequently and found that FSD would actually double stop: once at the stop sign and again at the line.
This created some interesting scenarios for me and I had many cars honk at me when the second stop would happen. Other vehicles that had waved me on to proceed through the intersection would become frustrated at the second stop.
FSD seems to have worked through this particular maneuver:
🚨 Tesla FSD v14.3.2 with a singular stop at the correct spot
No double stopping anymore in my experience https://t.co/Wd0TaNjc1R pic.twitter.com/CdQPvJHaAM
— TESLARATI (@Teslarati) April 23, 2026
FSD should know to go to the more appropriate location (whichever provides better visibility), and proceed when it is the car’s turn to move. The double stop really ruined the flow of traffic at times and generally caused some frustration from other drivers.
News
Tesla plans to resolve its angriest bunch of owners: here’s how
Since the rollout of the AI4 chip in Tesla vehicles, owners with the last generation self-driving chip, known as Hardware 3, have been persistent in their quest for a solution to their issue: they were told their cars were capable of unsupervised Full Self-Driving. It turns out the cars are not.
Tesla has a plan to make Hardware 3 owners whole after CEO Elon Musk admitted that those with that self-driving chip in their cars will not have access to unsupervised Full Self-Driving.
The company’s strategy is so crazy that it is sort of hard to believe.
Since the rollout of the AI4 chip in Tesla vehicles, owners with the last generation self-driving chip, known as Hardware 3, have been persistent in their quest for a solution to their issue: they were told their cars were capable of unsupervised Full Self-Driving. It turns out the cars are not.
Tesla owners with HW3 finally get their answer: https://t.co/CSZTKKkWXx
— TESLARATI (@Teslarati) April 22, 2026
During the Tesla Q1 earnings call on Wednesday, Musk finally clarified what the company’s plans are for Hardware 3 owners, what they will be offered, and what Tesla will have to do internally to prepare for it.
The answer was somewhat mind-boggling.
Musk said:
“Unfortunately, Hardware 3 — I wish it were otherwise, but Hardware 3 simply does not have the capability to achieve unsupervised FSD. We did think at one point it would have that, but relative to Hardware 4, it has only 1/8 of the memory bandwidth of Hardware 4. And memory bandwidth is one of the key elements needed for unsupervised FSD.”
He continued, stating that HW3 owners would have the opportunity to trade their cars in at a discounted rate in order to get the AI4 chip:
“So for customers that have bought FSD, what we’re offering is essentially a trade-in — like a discounted trade-in for cars that have AI4 hardware, and we’ll also be offering the ability to upgrade the car, to replace the computer. And you also need to replace the cameras, unfortunately, to go to Hardware 4.”
Obviously, Tesla has a lot of people to work with and make this whole thing right. Musk was adamant that HW3 would be capable of FSD, and now that the company has finally admitted that it is not, there are some things that could come of this.
There has been open talk about some sort of class action lawsuit against Tesla. The promises that Tesla made previously could be considered a breach of contract or even false advertising, and that’s according to Grok, Musk’s own AI program.
Musk went on to say that Tesla would likely have to establish new microfactories to effectively and efficiently replace HW3 computers and cameras:
…So to do this efficiently, we’re going to have to set up, like kind of micro factories or small factories in major metropolitan areas in order to do it efficiently. Because if it’s done just at the service center, it is extremely slow to do so and inefficient. So we basically need like many production lines to make the change.”
This is going to be an extremely costly process, especially if Tesla has to buy real estate, properties, and equipment to complete this work. Additionally, there was no wording on pricing, but Musk never said it would be free. It will likely come with some kind of price tag, and HW3 owners, after being left hanging for so long, will have something to say about that.
Elon Musk
SpaceX just got pulled into the biggest Weapons Program in U.S. history
SpaceX joins the Golden Dome software group, deepening its role in America’s most expensive defense program.
SpaceX has joined a nine-company group developing the core operating software for the Golden Dome, America’s next-generation missile defense system. According to a Bloomberg report, SpaceX is focused on integrating satellite communications for military operations and is working alongside eight other defense and artificial intelligence companies, including Anduril Industries, Palantir Technologies, and Aalyria Technologies, to build software connecting missile defense capabilities.
The Golden Dome concept dates back to President Trump’s 2024 campaign, and on January 27, 2025, he signed an executive order directing the U.S. Armed Forces to construct the system before the end of his term. The system is planned to employ a constellation of thousands of satellites equipped with interceptors, with data centers in space providing automated control through an AI network.
FCC accepts SpaceX filing for 1 million orbital data center plan
Space Force Gen. Michael Guetlein, director of the Golden Dome initiative, has described the software layer as a “glue layer” that would enable officers to manage and control radars, sensors, and missile batteries across services. The consortium is aiming to test the platform this summer.
Trump selected a design in May 2025 with a $175 billion price tag, expected to be operational by the end of his term in 2029, though the Congressional Budget Office projected the cost could reach $831 billion over two decades.
The Golden Dome role is only the latest in a string of military wins for SpaceX. As Teslarati reported, the U.S. Space Force awarded SpaceX a $178.5 million task order on April 1, 2026 to launch missile tracking satellites for the Space Development Agency, covering two Falcon 9 launches beginning in Q3 2027. That came on top of more than $22 billion in government contracts held by SpaceX as of 2024, per CEO Gwynne Shotwell, spanning NASA resupply missions, classified intelligence satellites through its Starshield program, and military broadband.
The accumulation of defense contracts, now including a seat at the table on the most expensive weapons program in U.S. history, positions SpaceX as the dominant infrastructure provider for American national security in space. With a SpaceX IPO still on the horizon, each new contract adds weight to what is already one of the most consequential companies in aerospace history, raising real questions about how much of America’s defense architecture will depend on a single private operator before it ever trades publicly.
Tesla Summon got insanely good in FSD v14.3.2 — Navigation? Not so much
Tesla plans to resolve its angriest bunch of owners: here’s how
