61–70 of 481 posts
Musk wanted to regulate AI to combat existential threat; in reality, it's not super intelligent self-aware AI, but buggy/shitty AI that's trying to kill people.
We will be killed by AI developers and the dunning kreuger effect, not armies of Skynet T-800s.
One of the great things about self driving cars is that we are going to start "debugging Roads". From my point of view all accidents until this time could be attributed to the civil engineer that build the road as much as the driver of the car. You make a pedestrian pass between roads made only for aesthetic purposes so you put a sign forbidding the pass and call it done. But people of course use it and get killed. Y…
> One of the great things about self driving cars is that we are going to start "debugging Roads". We've been debugging roads since Roman times. This is mostly about debugging software, and even more about crappy software development processes. After all, if your update is not monotonically improving things you have a real problem if your product is mission critical and lives are at stake.
This kind of video makes me wonder: what is the "autopilot" feature actually for? Do people generally like it? (I don't own a Tesla and I've never driven one.) If you're supposed to keep your hands on the wheel, and given videos like this that show you really need to keep your hands on the wheel and pay attention, is automatic steering really that big of a deal? Cruise control, now, that really is useful because it a…
Guess what computers excel at? Driving consistently on consistent highways.
The Tesla Autopilot is supposed to be the always aware and paying attention portion on these cases where a human driver would be very likely to start texting or dozing off. Now, it's not fully autonomous and may well decide it can't handle a situation (or apparently try to drive you into a barrier to see if you're still awake...). In this case the human driver who is somewhat zoned out needs to take control instantly and correct the situation, until they can safely re-engage autopilot (or pull over and make sure they're still alive, etc).
Earlier quoted context omitted.
> Very glad the driver had 100% of their attention on the road at that moment. Yes, and 100% is critical. That required pretty quick and smooth reactions. The car started to follow the right-hand lane, then suddenly re-centered on the barrier. The driver had about a second to see that and correct. That's just the sort of situation where even a slightly inattentive driver might panic and overcorrect, which could cause…
So your basically paying (A lot) of cash to beta test their software for them... at your own (Very personal) risk!
Earlier quoted context omitted.
Elsewhere from the comments: "Yes the lane width detection and centering was introduced in 10.4. Our X does this now as well and while it's a welcome intoroduction in most cases every one in a while in this instance or when HOV lanes split it is unwelcome." So basically, if I'm understanding this right they introduced a new feature that positions your car a little more consistently in the lane at the small cost of so…
And that's exactly what makes Tesla PR so disingenuous, they know better than anybody what updates they did and how often their updated software passed that point and yet they will happily cite numbers they know to be wrong. So, now regarding that previous crash: did that driver (or should I say occupant) get lulled into a false sense of security because he'd been through there many times in the past and it worked un…
Tesla PR knows nothing about what updates the engineering team did. At least some people in Tesla PR probably don't even know the cars update their software regularly.
It's bad practice for them to speak out of turn, but I can absolutely see the PR team not having a good grasp of what really indicates safety and their job is to just put out the best numbers possible.
This is why the idea of self-updating cars terrifies me. I'd never allow autopatching on a production server - Why would I allow it on hardware I'm betting my life on?
What does this say about the quality of the automatic emergency braking, if it can't detect a substantial metal object directly in front of the car?
Depends on how close it is and what it is. Have to weigh the energy of the impact against the pitch introduced by braking. You need the bumper to take the hit, so that it and the crumple zones can do their job. However, for a wall or highway barrier, it's probably almost always worthwhile to shed as much energy as possible. You're going to pitch yourself under the barrier where the bumper won't take the hit, like you…
Earlier quoted context omitted.
Exactly. This is what I’ve been saying. It is so dangerous. I love Tesla but I think autopilot is currently a death trap.
> I think autopilot is currently a death trap. I think this is a bit of an exaggeration. As long as the driver keeps paying attention and uses it as a driving aid, not a driver replacement, everything is fine. It's the moment that people start relying on it doing something that it wasn't built for, that the problems arise.
I think this is a bit too optimistic. People will start relying on it to be an autopilot. I think most people see this as the desired result: a car that can drive itself. Do you really think anyone wants "a car that will drive itself while you are paying full attention to every detail"? Otherwise, what is a driver really gaining from this? They're still expected to pay just as much attention (if not more) and I'd bet it's even more boring than regular driving (no interaction from the driver means it's like the world's most boring VR movie)
Humans are not machines, we love to find the lazy/easy way and we love to do things rather than stare at the road, eventually people will grow complacent (hopefully not before the tech is up-to-snuff).
Earlier quoted context omitted.
Depends on how close it is and what it is. Have to weigh the energy of the impact against the pitch introduced by braking. You need the bumper to take the hit, so that it and the crumple zones can do their job. However, for a wall or highway barrier, it's probably almost always worthwhile to shed as much energy as possible. You're going to pitch yourself under the barrier where the bumper won't take the hit, like you…
> Have to weigh the energy of the impact against the pitch introduced by braking. You need the bumper to take the hit, so that it and the crumple zones can do their job. Eh? The vast majority of real frontal impacts will be under heavy braking. If a few degrees of dive under brakes is enough to compromise the crumple zones' effectiveness then your car is bad (and I don't believe any Teslas are bad in this way.)
Small Overlap tests: this is where only a small portion of the car's structure strikes an object such as a pole or a tree, or if a car were to clip another car. This is the most demanding test because it loads the most force onto the structure of the car at any given speed. These are usually conducted at 15-20% of the front vehicle structure.[1]
Modern cars are surprisingly good at small over crash tests, see this video someone linked on HN a few days ago https://www.youtube.com/watch?v=DHlj8-JcWa8