Tesla FSD Is About to Start Learning How You Actually Drive
Two pieces of Tesla news dropped this week that are worth paying attention to. One is a software change that could genuinely make FSD less annoying to use. The other is a birthday party in Hollywood. Both are worth a few minutes of your time.
FSD Is Going to Stop Arguing With You
On July 18, Elon Musk announced that FSD will start remembering individual driver interventions and learning from them. The specific example he gave was HOV lanes. If you've been using FSD on a highway with a carpool lane and the car keeps trying to exit it against your preference, FSD will eventually learn that you want to stay in it. There's already a menu setting to enable HOV lane travel, but apparently it doesn't stick the way drivers expect in practice.
The stated goal here is notable: Tesla wants to get to a point where the only remaining takeovers are genuine safety interventions, not preference corrections. Right now, if you're overriding FSD because it parks in a spot you don't like or exits a lane you want to stay in, that's noise in the system. It makes the intervention data less useful and the driving experience more frustrating.
Parking Is the Real Problem
Musk identified parking as the most frequent reason drivers intervene with FSD. That tracks with what I've seen reported around v14.3.5 (as of July 13) and honestly with owner complaints in general. Parking is harder than highway driving in a lot of ways. The stakes of a bad decision are immediate and physical.
The planned improvements are specific: personalized preferences for where to park and whether to pull in or back into spots. Pulling into the correct driveway (your driveway, not a neighbor's) is apparently a relatively rare behavior that's also on the roadmap. And rare-but-important is exactly the kind of thing that separates a system you can actually trust from one you have to babysit.
This could mean FSD becomes more of a learned extension of how you drive rather than a generic driving style you have to constantly correct. One possibility is that drivers who use FSD consistently will end up with meaningfully different (and better-suited) behavior than drivers who use it occasionally. That would be a real differentiator.
Tesla Diner Turns One (and Gets Weird)
On a completely different note: the Tesla Diner in Hollywood opened a year ago on July 21, 2025, and is throwing a birthday event on July 21, 2026. If you're in the area and care about this sort of thing, there's a Cybercab on display and Tesla's Optimus Gen 3 will be in attendance. The first 50 customers who place an in-car order at midnight get a hat signed by Chief Designer Franz von Holzhausen.
The in-car ordering thing is the part that actually interests me. Being able to place a food order directly from the car's touchscreen before you arrive is a genuinely useful integration, not just a gimmick. It makes the Supercharger stop more efficient, which is the point.
Musk originally floated the diner idea back in 2018 as "an old school drive-in, roller skates and rock restaurant" and later described the concept as "Grease meets The Jetsons with Supercharging." The Hollywood location has apparently proven the concept enough that Tesla plans to expand it to major cities worldwide and along Supercharger corridors on long-distance routes. That second part (highway Supercharger locations) is the one that would actually change road trips in a meaningful way.
What to Watch
The FSD personalization update doesn't have a release date attached to it yet. But the framing around eliminating preference-based interventions is the clearest articulation I've seen of what Tesla is actually aiming for with FSD at this stage. Not perfection, but a system that learns what "correct" means for each driver specifically. Whether that plays out as described is a different question. But it's the right problem to be solving.
Source: Teslarati