I don't know about anyone else, but there isn't anything in AP that tells me the people behind it are all that impressive.
I don't mean to mean, but its rather ridiculous at times.
It re-centers anytime the lane gets wider like during a merge point
It has no coding to detect the turn signals of the car ahead to let them in
Often times the latency is too high between an event, and the car doing something to be worthwhile.
I followed the "Tesla insider" because everything he's said recently seems to be in line with expectations.
Your point is understood. However, I assign a somewhat gentler criticism to
"FSD does dumb things, you'd think this would be fixed already"
vs.
"I hope you buy this illogical explanation for why we [did | did not] fix this yet".
The former is part of development, the development is hard, and the main disappointment is that it goes too slowly - especially frustrating in light of optimistic projections that we discuss endlessly here. But to me that doesn't suggest basic incompetence in known or established practice, , it comes from over-optimism when pioneering new practice. It's easy to deride, but a lot of accomplishments would never be realized, never attempted, if the estimates weren't presented without a strong dose of optimism around schedule and cost.
The latter is prevarication, trying to get you to accept a bogus justification in the hope that it will be more acceptable than the blunt true reason. A corollary technique to diversion. It's used all the time in politics, now in public health policy, and is applied ever more cynically if it isn't called out.
I feel we can criticize the awkwardness and even occasional scariness of current FSD, without concluding that Tesla engineers are incompetent and/or that Elon is a charlatan. I think the evidence is that the general level of engineering competence is very high, and the Autonomy team is at least world-class (TBD if they're world-beating). Elon is extremely technically ambitious/optimistic but he's no phony. His tweets are maddeningly ambiguous. He doesn't seem to be a super-talented explainer, especially to a non-technical audience. Sometimes I call BS on him too, as with the explanation for rushed radar deletion and flimsily-justified lumbar-module removal in 2021. But in his case, I can't question his Tesla bona fides
Now taking the bike-lane issue: sure we all agree that pedestrians and cyclists must be protected. But that was a very early and obvious priority. If recognition & labeling of pedestrians and bicyclists isn't largely and confidently solved by now, which would justify the tweeted comment, then it suggests a fundamental limitation of the camera suite* or of the highest-priority NN perception functions. I don't actually believe that because if true, the inexorable conclusion is they'd have to pack up HW3 City Streets right now and wait for better sensor hardware - retrofitted or part of new vehicle designs. In other words, if bicyclists & pedestrians
can be protected, that's got to be in the code already.
The more logical and I think realistic reason is that, while we're indeed pretty confident we're not about to mow down bicyclists, the use-case of the bicycle lane as a secondary turn lane option, in heavy traffic in certain cities, simply isn't a high enough priority in the context of thousands or millions of other use-cases.
Put another way: the reason isn't pedestrian safety - that again is already achieved, or they should just suspend beta right now. The reason is similar to countless other things that FSD doesn't do expertly yet, it's not as important as say, recognizing safe gaps in high-speed traffic, which is consuming a lot more of our attention. In the case of that one, if it's not better real soon, they don't have suspend beta but they do have to refuse routing that requires such maneuvers.
So just say it. Don't say "we don't want to hurt [bicyclists and] pedestrians", that's a given and it's not the risk (unless you want to stop City Streets). Say "It's on the list and will happen, but deferring it won't cripple FSD. We feel there are higher priority use-cases to refine at this time. Use of secondary spill-over road areas will get more attention with the Rush-Hour Optimization stack in beta 12, hopefully December". It's only BS because it's optimistic, , not because I don't want to admit the real reason.
*I do think the camera suite has fundamental limitations (discussed just slightly up-thread), but presently I have no reason to believe this includes an inability to reliably detect pedestrians or cyclists near the car.