It’s been a while since I’ve posted as I really didn’t feel like there was much to say. That changed yesterday. When people asked me if my car could drive itself, I’d respond with, “It can. Just not very well.” It’s crazy how rapidly this has improved. When FSD first became available, I was disappointed – not in Tesla, but in realizing just how challenging the task was. There were so many simple things that FSD would just get wrong.
Now if someone were to ask me if my car can drive itself, I’d respond with, “It can. Almost as good as a human and already better than some.” I completely got the feeling that Tesla is close to that ChatGPT moment Musk talked about during his CNBC interview. Once it is as good as a human, it’ll actually be superior – as it never gets distracted and can see things humans can’t.
So, what changed my mind? I took a drive from the west suburbs of Chicago into the city with the newest FSD update 11.4.7.2 and over the 1.5 hours of driving I was amazed at how much the software has improved. I thought the immediate previous version was also getting pretty good, testing it out on short local trips when there wasn’t much traffic. But the trip yesterday was during evening rush hour. I went from feeling nervous and constantly on guard with FSD to being in a little bit of awe.
Rather than review all of the things that FSD did well, I’ll focus on the few things where it still had trouble. The nuances and situations are not at all surprising to me.
The first 30 minutes went well without me having to intervene at all. The car merged onto the highway, changed lanes, allowed other cars to merge appropriately, etc. While there was traffic, there weren’t any overly challenging situations.
The first intervention was almost 30 minutes in, and was because FSD mis-read a speed limit sign. The below image shows what I think was the problem. FSD saw the 45mph minimum sign and thought it was the speed limit. It was easy enough for me to raise the speed setting on the car to match traffic, but I did have to intervene.
Intervention 2 was the first one that I really felt was an issue. But the issue is more with inpatient drivers rather than FSD necessarily. FSD had a chance to change lanes where I thought it would but it waited a little too long, and then someone else sped up, got in the way, and then honked at me. The car may have ultimately navigated it all fine, but I felt it wasn’t safe – moreso because of other drivers than because of FSD itself.
I have the dashcam footage, and I can share that if people are interested in seeing what I’m referring to.
Intervention 3 was about 15 minutes later when I was at a busy 3 way stop heading onto a highway on-ramp. FSD had to make a left turn and was a bit timid with getting going. Basically, whenever it would pause, another vehicle would go. I was not wanting to wait any longer to see what it did as I did not want to upset the drivers behind me.
Intervention 4 was where FSD made a lane change while we were moving slowly and there was a lot of traffic. The car surprised me a bit by making a fairly aggressive lane change which was probably fine for it to do, but I just reflexively disengaged and then made the lange change myself.
During the drive, I had to figure out what to do with my right foot. Usually, I keep my foot hovering over the brake (or gas, depending on what is going on). My foot was actually getting a bit sore from holding it in the same position for so long. I was able to find a comfortable spot to rest my foot, but it was a bit unnatural to not have to keep my foot hovering over the brake the entire time.
So, a big thumbs up to the FSD team for the progress they’ve made. Based on the rate of improvement, I can see this actually being a viable solution a lot sooner than people are expecting. It is hard to believe that it is actually happening.