Latest / Elon Musk Podcast / Tesla FSD v14.2 What’s New and Who Gets It First

Transcript

Tesla FSD v14.2 What’s New and Who Gets It First · Elon Musk Podcast

• 0:00
  This. Is the Elon Musk podcast.
• 0:02
  Your daily hit of what is really going on at Tesla, SpaceX X AI.
• 0:06
  And the rest of the Musk universe.
• 0:08
  I'm your host Will Walden and I have covered Elon Musk for more
• 0:11
  than five years, spent a year on the ground at SpaceX, Starbase
• 0:15
  during. Early Starship development and
• 0:17
  before this I spent my career as a software developer working
• 0:20
  with billion dollar. Companies.
• 0:22
  I've also built and sold my own businesses and now I make.
• 0:26
  Content and. Help other people.
• 0:28
  Grow. Their companies now on this
• 0:29
  show, I use that experience to break down the news, filter out
• 0:33
  all the noise and give you clear context you can actually use.
• 0:39
  Tesla has begun pushing Full Self Driving Supervised version
• 0:43
  14.2 to cars in the field, and the company is framing this
• 0:47
  build as the step that enables A wider release.
• 0:50
  Now the update centers on a new, higher resolution vision encoder
• 0:54
  and a collection of driving behavior refinements that aim to
• 0:57
  make the car smoother and more confident.
• 1:00
  The big question is whether this roll out stays limited to a
• 1:03
  small set of vehicles or expands quickly to much of the fleet, as
• 1:07
  the company hints. Now today we're going to walk
• 1:10
  through what changed in V142, what hardware is receiving at
• 1:14
  first, how the release notes describe new emergency vehicle
• 1:18
  handling and routing, and why the parking and arrival options
• 1:22
  are a practical shift. We'll also look at the October
• 1:25
  comment that set expectations for broader availability.
• 1:28
  Then we're going to talk about what it's going to be like and
• 1:32
  what a measured, widespread push likely looks like in practice.
• 1:37
  Now, after the break, I'll reset with the Corfax and then move
• 1:41
  feature by feature with clear examples pulled from Tesla's
• 1:45
  notes. Now, Tesla started rolling out
• 1:49
  FSD supervised 14.2, and the company ties this build to the
• 1:53
  moment when a broader release could begin.
• 1:56
  The release notes focused on vision emergency vehicle
• 1:59
  protocols, routing that adapts to blockages in a set of new
• 2:03
  parking choices with several scenarios specific improvements
• 2:06
  listed plainly. Now with that baseline, let's
• 2:10
  unpack the details and what they add up to for drivers giving FSD
• 2:16
  on daily routes. Now this is a fully upgraded
• 2:19
  neural network vision encoder that reads higher resolution
• 2:22
  features from the scene. Now, in practice, that means the
• 2:26
  system should capture finer details and reduce ambiguity in
• 2:30
  cluttered environments, which supports more decisive actions.
• 2:34
  Now this connects directly to better detection of emergency
• 2:37
  vehicles, obstacles and even human gestures.
• 2:41
  So the camera stack is not just seeing more pixels, it is trying
• 2:45
  to interpret intent in motion any more granular way.
• 2:49
  Now emergency vehicle behavior gets specific treatment with
• 2:52
  this one. Tesla describes added logic for
• 2:54
  controlled pullovers in yielding behavior when the system
• 2:57
  encounters police cars, fire trucks, and ambulances.
• 3:02
  That's a practical step because these encounters carried legal
• 3:04
  expectations and social norms that people take for granted,
• 3:08
  and the software now aims to follow those norms with
• 3:10
  structured decisions. If the system recognizes
• 3:13
  flashing lights and adjusts predictably, it reduces
• 3:16
  confusion for other drivers and for the occupants of the Tesla.
• 3:20
  Now routing and navigation moves deeper into the vision network
• 3:23
  itself, which matters when the roads are blocked or detours
• 3:26
  appear with a little warning. And the notes say that the car
• 3:30
  can respond to block roads and detours in real time because
• 3:33
  planning is more tightly coupled to perception, That architecture
• 3:37
  should help the system replan without awkward pauses when a
• 3:40
  lane closes or when a route changes near construction, which
• 3:44
  is where earlier versions sometimes hesitated.
• 3:47
  Now there's several driving scenarios that receive targeted
• 3:50
  refinements in this build. The notes call out unprotected
• 3:54
  turns, lane changes, vehicle cut, insurance, and interactions
• 3:58
  with school buses. Each of these is a stress test
• 4:01
  for prediction and yield behavior, so tuning them
• 4:04
  suggests work on both comfort and compliance.
• 4:06
  And when the card judges gaps more clearly on an unprotected
• 4:09
  left, commits earlier on a safe lane change, or respects school
• 4:13
  bus rules without abrupt breaking, the occupant
• 4:16
  experience improves while risk drops dramatically.
• 4:21
  Now there's also parking and arrival options, and they've
• 4:24
  added a new layer of user control to that.
• 4:26
  The release lists choices for where FSD should park or drop
• 4:31
  off, such as a lot, the street, a driveway, a garage, or the
• 4:35
  curbside. Now that matters for the last 50
• 4:38
  meters of a trip. Or earlier versions could second
• 4:41
  guess a destination or stop in a suboptimal place a couple blocks
• 4:44
  away and just expect you to get out and parked there.
• 4:48
  Well, it's most of the time that doesn't work.
• 4:50
  And if the driver can pre select the arrival style, the system
• 4:54
  can plan the final approach with fewer last second adjustments,
• 4:57
  which should feel more natural when you arrive at your
• 5:00
  destination. Now the hardware scope is very
• 5:03
  clear in the early hours of the rollout.
• 5:05
  Reports that the update landed first on Model Y vehicles in
• 5:08
  California with the company's A14 hardware set the expectation
• 5:13
  that this is not yet across every single car.
• 5:17
  The notes also indicate that this wave is limited in terms of
• 5:20
  included vehicles, which points to a staged approach as Tesla
• 5:24
  validates the new model and a narrower slice of the fleet
• 5:27
  before pushing it more widely. That matches how complex
• 5:30
  perception changes typically ship, and the version string in
• 5:34
  the notes reads 2 O 25.38.9.5 with FSD supervised V 14.2
• 5:42
  installed now alongside the big items, the list includes smaller
• 5:47
  quality of life changes, like an alert for residue buildup on the
• 5:50
  interior windshield that could degrade front camera visibility.
• 5:54
  That line is a reminder that sensor health is not just about
• 5:57
  software, though. Real world grime can still blunt
• 6:00
  the vision system. Snow, dirt, mud, a leaf falling
• 6:05
  in front of your camera, anything.
• 6:07
  And the car now warns if that happens, so drivers can clean it
• 6:11
  and restore fidelity on their own and not rely on the system.
• 6:16
  Tesla also lays out a few upcoming improvements attached
• 6:19
  to this train, including continued to work on overall
• 6:22
  smoothness, on parking spot selection, and on parking
• 6:25
  quality. The company positions these as
• 6:28
  the next steps rather than finished work inside a 14.2,
• 6:32
  which signals that parking remains an active area of tuning
• 6:35
  now for drivers. That means the car should get
• 6:37
  better at choosing A stall or a curbside space and it completing
• 6:41
  the maneuver without hunting. Now, context for the widespread
• 6:44
  framing comes from a comment in October responding to a well
• 6:48
  known tester who praised a prior 14 point 1.2 build for reducing
• 6:52
  indecisive lane changes and braking.
• 6:55
  Now, Elon Musk said that 14.2 is for broad use.
• 6:59
  That comment sets a bar for today's release, because if 14.2
• 7:02
  begins on a narrow hardware subset and then steps into a
• 7:06
  much larger pool to performance, people experienced on 14 Point,
• 7:09
  1.2 should carry forward just with the higher resolution
• 7:13
  vision and added behaviors. Now the least notes repeat
• 7:19
  several times in ways that show emphasis, including improvements
• 7:22
  to handling static and dynamic gates and offsetting for Rd.
• 7:26
  debris such as tires, branches, and boxes.
• 7:29
  And in daily driving, that translates to fewer awkward
• 7:32
  slowdowns, an object sits partially in the lane and better
• 7:36
  lateral positioning to give space.
• 7:38
  Those are subtle changes that add up over a commute, cutting
• 7:42
  down on small corrections that used to break the flow of the
• 7:46
  drive. Now decision making reliability
• 7:49
  gets a nod with language about better management of system
• 7:53
  faults and smoother recovery from degraded operation.
• 7:56
  So in practical terms, that means if a sub component has a
• 8:00
  momentary hiccup, the system should not overreact or exit
• 8:04
  automation abruptly without need.
• 8:06
  Smooth recovery reduces driver workload and builds trust
• 8:10
  because consistent behavior is easier to supervise than
• 8:12
  behavior that changes dramatically and drastically
• 8:15
  under small disturbances. Now there is also the question
• 8:19
  of hardware diversity in the existing fleet.
• 8:22
  The company's wide FSD fleet still includes a large number of
• 8:26
  HW-3 vehicles, and it remains to be seen how 14.2 maps out of
• 8:31
  that mix. If early validation on AI 14
• 8:37
  looks strong, Tesla will need to communicate how the vision
• 8:41
  encoder and the new behavior scaled down to earlier hardware,
• 8:45
  or whether some features arrive in a trimmed form.
• 8:47
  That's a normal step is architectures evolve.
• 8:51
  Now Tesla is rolling out FSD Supervised V 14.2 with a higher
• 8:56
  resolution vision encoder, explicit emergency vehicle
• 8:58
  protocols, routing that adapts to detours, new parking and
• 9:02
  arrival options, and a list of scenario level improvements.
• 9:05
  The first wave appears limited to AI 14 hardware, with the
• 9:09
  company signaling that this is the build meant for wider use
• 9:13
  once validation supports it. If the earlier behavior holds
• 9:16
  across more cars and hardware versions, the path to a broader
• 9:19
  release is straightforward. Keep the comfort gains from 14.1
• 9:23
  point X, apply the high resolution perception and
• 9:27
  expand, expand, expand. Hey, thank you so much for
• 9:32
  listening today. I really do appreciate your
• 9:34
  support. If.
• 9:35
  You could take a second and hit this subscribe or the follow
• 9:37
  button on whatever podcast platform that you're listening
• 9:40
  on. Right now I greatly appreciate.
• 9:42
  It it helps out the show tremendously.
• 9:44
  And you'll never. Miss an episode?
• 9:46
  And each episode is about 10 minutes or less.
• 9:49
  To get you caught up quickly and please if you want to support
• 9:52
  the show even more. Go to Patreon dot.
• 9:54
  Com slash. Stage 0.
• 9:58
  And please take care of yourselves and each other, and
• 10:00
  I'll see you tomorrow. OK, so what's going on?