What’s in the Mind of FSD? Yesterday, I had a strange thought. Driving home on the highway (v14.1.4), my car needed to take a crowded exit. The merge looked tight, and I felt that familiar feeling of anxiety— was FSD going to make it? I held back and let it handle things. It merged flawlessly. In fact, it was nearly optimal. And that’s when it hit me: how do humans do this? Obviously, our brains are doing heavy processing—tracking motion, estimating gaps, and predicting outcomes. But we also inject emotion and hyper-focus into the moment. That’s something entirely absent in the “mind” of FSD. Neural networks notwithstanding, machines think differently. It’s not in the mind of FSD to worry about missing an exit or to feel irritation toward another driver. It doesn’t hyper-focus on a single problem to the exclusion of everything else. Human “emotional thinking”, our evolved fight-or-flight mechanism, is a way to short-circuit the compute required to find a solution. It often helps, but not always. Picture this: you’re driving down the highway, and a car starts merging into your lane, risking a collision. Your brain locks in on that threat. But what else do you remember from that moment? Were you equally aware of what was happening in the opposite lane? Probably not. FSD, on the other hand, remains aware of everything. While it considers the potential risk of that merging car, it’s also equally accounting for every other object in its occupancy. That’s hard to fathom, but 36 times per second, FSD evaluates the entire scene, not just the emergent threat. (Still think you’re a better driver than a robot?) Ashok Elluswamy, Tesla VP of AI, has talked about how perception and planning— the two logical halves of the autonomy problem— aren’t as distinct as they seem. Human values shape how we resolve driving situations: some people slow down to avoid a merge, others speed up. Both can work; one reflects caution, the other a different risk-reward tradeoff. It’s fascinating to watch how we project our own cognitive logic onto these systems. This phenomenon will become more common as AI increasingly mimics the brain’s input-output behavior—if not its internal process. (If you think this is difficult with autonomous vehicles, just wait for Optimus to arrive on the scene. )

To me, a car is a car, but this 1947 Cadillac Series 62 Sedanette fastback is drool-worthy.

Tesla FSD Platooning can start at significant scale in 2026. It will be half the cost of Diesel Trucks per mile. 29 states have AV/platooning laws. 15+ explicitly allow no-driver trailing with lead driver present. FSD-Enabled Platooning (Mid-2026–2027). This enables 18–22 hours/day, costs hit 70–90 cents/mile versus $1.50 for diesel. This is more adaptable than physical multi-trailers, avoiding permits. Higher utilization (e.g., 1,000+ miles/day per platoon) could generate $2,000+ daily revenue per truck at $2.20/mile, vs. $1,100 for singles—driving fleet conversions. Three drivers in a convoy only needs one driver in the lead truck. The others could rest and take turns supervising the fleet. They can all go their own way for the final part of the delivery. 50,000 trucks with platooning software could generate $1.40 per mile on 200,000-300,000 miles per year of operation. The first batch of trucks could operate from Nevada, through Texas, Georgia and into florida. They could also operate in California. @elonmusk @herbertong @robotaxi @Tesla @teslaownersSV @RandyWKirk1 @SawyerMerritt @FutureJurvetson @stevenmarkryan @farzyness @davelee @heyitsyashu @TeslaBoomerMama