I have always felt EAP and in particular FSD was amazing on the freeway. Was super excited to finally get past the silly "safety score" game and download the FSD beta. First build I got was 10.5 and in my experience, it was amazing. Sure I had to take over at times, but in general without any unexpected obstacles, animals, bad drivers, etc, I found it really spot on and I rarely had to intervene/disengage. I took that version to another suburban FL city and had the exact same experience; thrilled and delighted brother who was in awe next to me in passenger seat. "That was like a rollercoaster ride -- a little scary but super fun and amazing!"
Then came the upgrades, and I quickly began to experience a major degradation in performance. Frequent phantom breaking. I first noticed it in relation to a car up ahead taking a right hand turn on to my street and approaching me in the other lane. That car was driving perfectly and was no where near encroaching on to my lane, but clearly the software interpreted it as a threat and would brake -- hard. Then I noticed the same behavior but with no visible potential threat at all -- simply driving down a street at 35 MPH and out of no where -- hard breaking. LOL would have generated multiple "unsafe breaking" or whatever the term was on safety score. That never happened on 10.5. I actually stopped using FSD because it had become so bad it was no longer pleasurable or fun -- just a bunch of nauseating innappropriate braking. If you had had the safety score running, it would have been about a 10!
I chalked it up to a really bad build, and then noticed that the upgrades that had been coming at about an every 2 week clip stopped, and I went about 6-8 weeks with no upgrade, which I interpreted as Tesla getting their **it together and I hoped for improvement.
Took the next upgrade about a week ago. Got about 60 seconds from my house on a very typical residential 2 lane main thoroughfare with a speed limit of 35 MPH, and bam -- hard braking with absolutely nothing in site! Super disapointed. There is no way I would ever engage the feature with a passenger in the car -- it's simply uncomfortable, and "stupid" enough quite frankly to be like "WTF you didn't pay for that did you?!" My wife a long time ago adopted a "OMG turn that fracking thing off or I'm getting out!" attitude and I can't blame her.
I cannot fathom why a build I was running probably around Thanksgiving was far superior to what I have now. But that is my experience in a 2018 M3 with a hardware upgrade to FSD computer.
What I am running now (comment related to city streets only) is barely a proof of concept, not even close to a real "beta."
I think Tesla is the recipient of a huge amount of generosity amongst it's user/fan base. If the GP had purchased this product and gotten this kind of return, seems inevitable there would be major blow back, class action suit, etc.
The release notes are interesting, but only from a technical perspective. 45% increase in that 2% decrease in this, whatever -- they don't inform the user experience at all. They ought to just sit on the software until it's made an appreciable and significant improvement not by trivial technical measurements but by actual people with a "wow this is significantly better" reaction. Or run their own stupid safety score with FSD for a 1000 miles and get at least a 50! If I could push a button and revert to 10.5 I would do it in a heartbeat.
From what I can see, based on their current iteration, they seem lost with no significant pathway to major improvement. "Teslavision" feels like vaporware. I think it's at the point where it begins to inform the investment decision. After all, this was supposed to represent another major source of revenue, and early on it absolutely seemed like that would be the case. No more. And anyone who puts "stock" in anything Elon says about where they'll be "this year" simply isn't paying attention. And this from a staunch Tesla advocate <s>.