HuaRenCa
Back to Forum
Community

Let's Start a Discussion Thread on FSD and Various Driver-Assistance Systems

刘轩豪
刘轩豪

2 weeks ago

From what I can see, there's basically no systematic FSD discussion thread in the forum—any related content is scattered across different posts. I'm fairly interested in this area, so I've done a bit of research.

According to a nearby thread, everyone strongly agrees that the only reason to buy a Tesla is FSD, and I think so too. The reasons include, but aren't limited to, FSD being genuinely smart/capable/useful, and once you use it you can never go back, etc. But is using FSD really less stressful?

That's the question I started thinking about. FSD is smart and can handle over 90% of road conditions, but occasionally it fails in certain scenarios. When a person is driving, they have plans, predictions, and awareness of what's about to happen; but with FSD as a black-box model, you can't predict what it will do next: it might suddenly enter the oncoming lane without warning, or fail to complete a left turn and nearly collide with an oncoming car, etc. And in the instant the model loses control, it's often very hard for a human to rescue the situation, leading to accidents. If it's your daily commute route, that's fine, but what if you go to an unfamiliar area and road? The initial discussion was based on the idea that although FSD can drive for you most of the time, the human needs to stay highly alert at all times, and the energy consumed in the end may be greater. If it makes a mistake, you need to correct it in time, and some moments offer no possibility of correction once an error occurs.

I myself drive with comma, and on straight roads it's fine, but on curves you have to watch the car because it may fail to make the turn and run out of the lane, and at that moment the mental load is noticeably higher.

Based on this initial question, I discussed it with AI. I recommend reading:

AI Answer 1 (click to learn more details) If FSD human-machine collaboration is more tiring than driving yourself, then how safe is FSD? According to the AI, although official statistics show that FSD's accident rate is much, much lower than that of human driving, there are also three data biases.

AI Answer 2 (click to learn more details) So if we set aside data bias, is there any data that really proves FSD's accident rate is actually lower than humans'? TL;DR: No, it cannot be proven.

AI Answer 3 (click to learn more details) So based on the above two discussions, FSD is essentially a system that requires human supervision, and no matter what ultimately happens, a human is responsible. Therefore, under current circumstances, FSD can never evolve to L3.

AI Answer 4 (click to learn more details) The previous two AI answers both discussed whether FSD is safe and whether it is really more reliable than humans. So back to the original question: is FSD really less effort? After all, gas cars aren't completely useless—ACC + LKA is standard on most models, and rule-based systems are more predictable. If it's not good enough, you can add comma. So on highways, gas cars are absolutely not inferior to FSD; the main discussion is about local streets.

Here, based on paper data and lab data, FSD human-machine collaboration consumes far more energy than driving yourself. If you are actually using FSD in real life (not on your commute or a route you take every day, but in an unfamiliar area and road), is it really less stressful or more tiring?

AI Answer 5 (click to learn more details) Or if your need is merely point-to-point from A to B, and FSD happens to perform almost 100% on that stretch of road, that's fine too. But once you give up FSD on a certain stretch, according to everyone's feedback, a Tesla is very hard to drive.

AI Answer 6 (click to learn more details) At this point, I guess you'll definitely say: since supervising on local streets is more tiring than driving yourself, why would you still choose FSD? That's very normal, but regarding the motivation raised at the beginning of this thread, discussion is welcome.

AI Answer 7 (click to learn more details) Finally, let's compare all aspects.

AI Answer 8 (click to learn more details) Finally, a little discussion about Tesla: is it somewhat like Amex in the credit card world, definitely the most controversial brand? I did previously say that "driving an EV doesn't save money," for various reasons. But today I want to revise that statement slightly, at least to "driving a Tesla doesn't save money." Why? Because Hyundai Ioniq has some really awesome lease deals; the Ioniq 5 2026 launched contract standards as low as $289–$329/month (24 months, around $3,999 down). In practice, some forum users skilled at "haggling" have managed to achieve zero down payment or extremely low monthly payments by stacking manufacturer cashback and dealer discounts. And for the sedan Ioniq 6, the discounts are even more extreme than for the SUV—you can even see monthly payments as low as $239 with zero down, which is truly frugal.

Finally, I hope everyone won't rely too much on FSD. After using comma for a long time, I basically never manually center the car anymore. Then I rented a car without lane-centering, and it was unbearably hard to drive—I've forgotten how to drive. It's just like how programmers who use AI to write code will forget quickly or learn nothing; driving is the same. Rely too much, and in the end you won't know how to drive. Not everyone is a programmer, but everyone has to learn to drive, right? If an accident happens, you are always the first responsible party.

Lastly, I know this installment has a lot of AI answers, so I've hidden them—open them yourself if you want to read. The AI's answers are broadly consistent with my own thinking, but I'd also like to see everyone's discussion.

Let's Start a Discussion Thread on FSD and Various Driver-Assistance Systems

5
1.9K

Comments (5)

Your avatar
Sign in to comment
街头舞王
街头舞王2 weeks ago

For those failure scenarios you mentioned, are there specific records from actual road testing? Or do they mainly come from simulation?

天美航空
天美航空2 weeks ago

There's another premise worth unpacking: the tension that comes with unfamiliar roads may not stem mainly from the system being unpredictable, but from people simply being unfamiliar with the environment itself. If you drove the same unfamiliar route manually two or three times before handing it over to assisted driving, would the comparison of mental energy expenditure look completely different? This affects whether the conclusion should be attributed to FSD or to the unfamiliarity itself.

濤聲依舊
濤聲依舊2 weeks ago

Your point about the higher cognitive load on curves is really interesting—maybe the issue isn't human-machine collaboration itself, but that the system isn't communicating its intent to the human ahead of time.

收集者
收集者2 weeks ago

I have reservations about the part on data bias—it's hard to fully rule out statistical bias, but that doesn't mean we should conclude the system is necessarily more dangerous.

我没有魔法
我没有魔法2 weeks ago

The core disagreement really comes down to whether monitoring costs should count as driving burden or as safety redundancy. Under the former, long unfamiliar routes do feel more tiring; under the latter, an extra layer of reminders isn't necessarily a bad thing. To move the discussion forward, it would be best to track takeover frequency, takeover difficulty, and drivers' subjective workload separately—otherwise everyone can only trade personal impressions.