this post was submitted on 30 Aug 2023
202 points (93.2% liked)
Technology
59421 readers
3519 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each another!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
Approved Bots
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
AI DRIVR made an interesting analysis about the v12 on YouTube. Apparently it's completely different from the previous versions and instead of understanding traffic rules it learns from a videos of people driving which means it does things like doesn't fully stop at stop signs and drives over the speedlimit - like people do too.
It's interesting because by strictly following traffic rules you might infact be a danger to others but by driving like humans you're also breaking the law. Good example of a situation where the "right" thing to do might not be the most intuitive one though in this case it's still up for a debate.
That’s what we were all clambering for: a self driving machine that operates like a mouth breather late for work.
Elon is a masterclass of stupid.
Perhaps you should put your hatred towards Elon aside for a while and objectively consider what actually is the better solution here.
One could argue that strictly following the rules is the right approach, and perhaps it would be if everyone actually drove that way. However, in reality, that's not usually the case. What truly increases traffic safety is predictability. If most drivers are rolling through stop signs and you're the only one stopping completely, while you might technically be in the right, your behaviour could lead to accidents due to the unpredictability. The same applies to speeding. Driving significantly slower than the flow of traffic might slow down the traffic flow, leading to unsafe overtakings and such. While you might be legally correct here too, in practice, a slight increase in speed could lead to increased road safety.
These are complex issues. A dose of humility might go a long way instead of acting like the answer is obvious.
I do this kind of thing for a living, and have done so for going on 30 years. I study complex systems and how they use learning and adaptation.
Musk’s approach to these systems is idiotic and shows no understanding of or appreciation for how complex systems - animals, in particular - actually work. He wanted to avoid giving his vehicles lidar, for instance, because animals can navigate the world without it. Yet he didn’t give them either the perceptual or cognitive capabilities that animals have, nor did he take into account the problems of animal locomotion being solved by evolution are very different from the problems solved by people driving vehicles. It, of course, didn’t work, and now Tesla is trailing the pack on self-driving capabilities with the big three German car makers and others prepping class 3 vehicles for shipping.
If he is trying to chatgpt his way out of the corner he’s painted himself into, he’s just going to make it worse - and, amusingly, for the same reasons. Vision is just one dimension of sensation, and cars are not people, or antelopes, or fish, or whatever his current analogy is.
This is just Elon Eloning again. No one predicts a car coming towards them is going to do a California stop at a stop sign. If Om pulling into an intersection and I see someone rolling through a stop sign, I’m hitting the brakes because obviously a) they didn’t see me and b) they don’t know the rules of the road. Elon’s cars have a problem with cross traffic and emergency vehicles anyway, making the logic fuzzier is not going to improve the situation. If he thinks throwing video and telemetry data at a large model is going to overcome his under-engineered autonomous system, I suspect he’s going to be in for a rude discovery.
If there’s anything kids today can learn from Elon (or from Trump for that matter), it’s how to be so confidently wrong that people throw money at you. The problem is that if you’re not already born into wealth and privilege, you’re likely to merely become the owner of the most successful line of car dealerships in a suburban county in Pennsylvania, or else in prison for fraud.
If FSD is trained from billions of hours of video data then it by definition drives like an average driver and thus is highly predictable.
That’s not how it works, unfortunately. That’s how people want it to work, but it’s not how it works.
This is just more of Elon’s pie in the sky.
If you've done this kind of stuff for living for the past 30 years then I'm sure you can give me a better explanation than "that's not how it works"