pupbiru
@pupbiru@aussie.zone
- Comment on Elon Musk May Have Made a Huge Mistake on Full Self-Driving That It's Too Late to Correct. 2 weeks ago:
Their stance is that by using lidar OEMs are hamstringing themselves on solving vision because they are so reliant on it.
i get that… but… vision is kinda shit. why not use all the tools at your disposal? like literally “x ray vision” is something that we see as a super power because it’d be so useful - radar gives us that
vision is an approximation of things like lidar. can you get a depth map out of vision? sure by why not just measure it directly and then you’re not introducing error by your model literally hallucinating
The more sensors you deal with, the more your attention gets divided. You aren’t laser focused on one thing.
kinda but also the last 20% takes 80% of the effort… solving a lot of easy problems with more information will lead to a better short term outcome, and then when you’re getting good results then you can solve from 80% to 85% then 85 to 90 etc across your whole sensor suite
The extra sensors also cost a lot of money
so they though? you can buy hobbyist ultrasonic sensors for literally a couple of bucks, lidar for a few hundred - sure that’s not at the grade that you’d use for cars, but at some point it’s an economies of scale problem. they’re not actually that expensive for a commodity “good enough” sensor package
You might not like the reasons, or their stance
correct - i understand them, but as an engineer it’s just wrong when you’re talking about one of the most dangerous activities that humanity collectively engages in (driving)
What happens when they keep seeing improvements in vision and now radar isn’t needed?
i think this could be the sticking point - i don’t think any extra sensors are needed, just like i don’t think seatbelts or air bags etc are needed… but… they’re helpful and improve the safety of people in and around the car
all the crazy headlines you see about it are idiots in cars being idiots
agree, and i totally think driverless is the way to go - humans are far worse drivers than machines are right now without any improvement
… however, better isn’t perfect, and when it comes to safety simply ignoring tools because of some belief that eventually it’ll be fine is misguided at best, and negligent at worst
If people wanna blame Elon for convincing people to be idiots, sure, you can do that
absolutely that too! their systems aren’t “drives itself no problemo” and that’s how they’re marketing it
- Comment on Elon Musk May Have Made a Huge Mistake on Full Self-Driving That It's Too Late to Correct. 2 weeks ago:
i don’t think anyone is relying solely on radar - that’s the point. every sensor we have as fallible in some way (and so, btw, are our eyes - they can’t see through things but radar can in some cases!)
even if you CAN rely solely on vision, why hamstring yourself? with a whole sensor package, the algorithms know when certain sensors are useless - that’s what the training is for… knock 1 out, the others see that it’s in X condition and works around it
if you only have a single sensor (like cameras) then if something happens you have 0 sensors… our eyes are MUCH better at vision than cameras - just the dynamic range alone, let alone the “resolution”… and that’s not even getting into, as others have said, the fact that our brains have had millions of years of evolution to process images.
the technology for vision only just isn’t there yet. that’s just straight up fact. can it be? perhaps, but “perhaps in the future” is not “we should do this now”. that’s called a beta test, and you’re playing with human lives not just UI bugs - and there’s no good reason… just add extra sensors
- Comment on Elon Musk May Have Made a Huge Mistake on Full Self-Driving That It's Too Late to Correct. 2 weeks ago:
the large majority of current self driving cars have radar, lidar, ultra sonic, and cameras. their detection sets overlap, and complement each other so they can see a wide array of things that others can’t. focusing on 1 and saying “it doesn’t see X” is a very poor argument when others see those things just fine
- Comment on Elon Musk May Have Made a Huge Mistake on Full Self-Driving That It's Too Late to Correct. 2 weeks ago:
you’re not wrong, but also that’s a fantasy with current technology. meanwhile, cars are dangerous heavy hard boxes travelling around at high speed while we “get the technology right”, and that’s unacceptable
- Comment on Valve will finally sell the Steam Deck in Australia 5 weeks ago:
hard to get no but not easy either
- Comment on Valve is working on a version of proton for ARM devices 1 month ago:
does ARM still have better battery life when all of the machine code has to be translated from x86
afaik macos/rosetta is more efficient than native windows/x86, but that could be down to OS integration, or any number of confounding factors… i’d suggest though that x86 windows applications sometimes run better and more efficiently on alternative platforms, even with the translation layers
- Comment on Valve Working With Rockstar to Fix GTA Online on Steam Deck 1 month ago:
afaik the anti cheat they added is one that’s known to work with linux, but you do have to do some extra stuff… i think valve worked with them to ensure it works on deck, and are probably now working through the steps with rockstar to ensure it’s as easy as possible for them
- Comment on Using GPT-4 to generate 100 words consumes up to 3 bottles of water — AI data centers also raise power and water bills for nearby residents 1 month ago:
and it’s still absolute crap… the heat produced by 100 words of GPT inference is negligible - it CERTAINLY doesn’t take 3L of water evaporating to cool it
- Comment on Valve bans Razer and Wooting’s new keyboard features in Counter-Strike 2 2 months ago:
actually professional motor sports are quite an exertion because they drive for hours with no rest and they’re doing a lot of movement of the wheel and pedals - it’s not just driving down an interstate for a couple of hours
- Comment on Valve bans Razer and Wooting’s new keyboard features in Counter-Strike 2 2 months ago:
you realised the olympics used to include poetry and art, right?
- Comment on Valve bans Razer and Wooting’s new keyboard features in Counter-Strike 2 2 months ago:
i’d imagine it’s pretty detectable anyway… if the point is pushing a or d without any break between them, that’s real easy to time in software: no human is going to be perfect every time
sure, then comes the arms race of circumventing by adding some delay, and some variance in the delay time, but no large hardware manufacturer will just include it at that point and it’ll be obvious it’s a hack rather than an acceptable feature
- Comment on The terrible tragedy of a stolen Steam Deck 2 months ago:
even not a specific company: mention to all of them that it was stolen while they had a pretty limited group of people at the house
you might think it’s a case of “how would they know who there’s no point”, but people who steal things like this likely didn’t do it just once… it is, or will become a pattern of behaviour. if nobody reports it, they have no chance of identifying a pattern of behaviour to narrow down the culprit… if a company gets 2 or 3 reports of stolen items from houses that an individual employee is working at, it becomes pretty clear who the culprit could be
you even have pretty good evidence that it was stolen rather than lost: the fact that it came online for a period means someone has it and has connected it to a network and then not reported it lost