1,735 karma · joined May 9, 2023
> I insist on a wired headset
usb audio generally works great, and because it's powered, you often get better audio quality (with microphone arrays for anc)
and even neat features like system volume integration with buttons and ANC
i have a yealink uh48 but you can get almost-as-good functionality for a lot less (or pay more for even better!)
folks you can't make this stuff up
but you said "FSD" which stands for FULL Self Driving. You didn't say FSDS, or FSD Beta, or "Tesla's Level 2" self driving.
It is NOT a FULL self driving solution yet. They are lying and you are repeating their lies.
In addition, a certain CEO of a certain car company has made very aggressive sales pitches about how LIDAR isn't necessary and even going so far as to sell unfinished self driving solutions as being completely capable of doing so.
So hypothetically, if that car company were to adopt LIDAR, they would have been selling their solution under false claims in a sales capacity.
I think the natural solution is a 360 lidar as a source of truth, followed by optical for everything else
If there is a stretch of surface road with a complex intersection and aggressive drivers, you are there to step in. The Waymo Driver instead will try to avoid that area.
Also, Tesla's hardware is not nearly complex or powerful enough to approach the low level of intervention or incident rate that Waymo has - but HW5 will be a massive step forward for them.
That's going to be the make or break for me, personally. If Tesla can't make their solution work well on HW5, then they have serious structural issues in their program.
An Ojai costs less than a mid tier luxury vehicle. Is that not cheap enough?
And the price is only coming down.
Of course not!
So, you may be surprised to find that Waymo's sensor package doesn't cost $200k. Whoops!
To me, true self-driving starts at level 3. Otherwise, it's supervised.
And Tesla selling their product as "Full Self Driving (Supervised)" is like selling "Fully Cooked Meal (Frozen)". No! You call it a "frozen meal"!
What is in the article is explicitly walking around. Yes, you can do it with controller input.
> No offense but did you look at what they exported from Fusion? Or the video demo of what they built? The only things that are not blocky are the models downloaded from IKEA. Nothing about this is impressive - AVP can render much better. Even the Meta Quest 3 can render much better.
No offense, but mentioning the MQ3 only proves my point. It's amazing how far we've come. Headset mounted compute VR from even 15 years ago would have struggled with this.
> Yes, but this article makes use of none of it because they are not doing AR.
I was just mentioning it. Quote me where I said they were doing that, thanks.
> Yeah there was, you mounted some markers on your headset and attached the mocap camera on marker set.
You did VR? How did you get the round trip latency below 100ms? That was a serious issue until relatively late in the game. Maybe they cracked it by 2011, but even in 2006, it was a serious issue (mocap rigs are not designed for utra low latency feedback)
I didn't say it was.
> Is this a parody? The chip is a laptop-grade APU tethered to a power bank, and you're bragging about pixel-shading a <1000 tri mesh? Your mind will be blown when you discover what the Nintendo Switch can do with 10w.
OP was talking in the timescale of "since forever", and so am I. Your mind will be blown when you discover what happens when you export that mesh without optimization or LOD maps to a head mounted computer from 2004.
> What am I missing?
So much.
> This is an absolutely standard feature of any VR headset that has ever been released in the history of VR headsets
The article is describing "room scale positioning". The first consumer available headset to offer this was released in 2016.
It may feel like it's been around forever if you're not familiar with the space except as a consoomer, but exploring a 3d space at room scale has only recently come into the realm of possibility for consumer grade headsets. For much of the 80s and 90s (and 00s), it was head tracking only. Real time positioning and moving around required positioning hardware that was a magnitude more expensive - and pretty much always had very specific space limitations. Sure, it existed to some degree, but it was such inexpensive addition that using an expensive and complicated treadmill was often considered the economical choice.
In fact, I don't think there was even a "professional" solution that offered room scale positioning before the vive. Previously, they all had specific limitations like a 4 x 4 foot area or were massive tech demos that existed in a quantity of one in a university basement somewhere, running a very tailored demo, and they certainly weren't feasible in the 90s.
> Make a 3D world, walk through it.
First of all, how do you map the room? Do you set up lighthouses? Low latency positioning? To do what is in the article, you need a large space and the ability to map into it. Only recently, has "inside out" tracking become feasible. Previous generations didn't have it, and it's almost a guaranteed requirement for any large scale deployment of VR - for example, exploring multiple rooms.
Second, how do you create the "3d world"? It may seem simple, but VR headsets require more than double the graphical computation, and that means that any geometry and graphics you inert of the world needs to be within the compute budget of the headset solution. The fact that the person in the article was able to just export a USDC directly out of fusion and have it render without hitting the compute budget limit is amazing and really speaks to the graphical power of the AVP.
Lastly, stitching, the 3-D solution and compositing, and the real world is a very difficult problem. Again, it may feel like we've had the technology, but really we haven't had the high speed camera solutions in small packages, and the compute power to process all of that in anything approaching a consumer solution. To do what AVP is doing with the USDZ files, composing them into the real world, and even applying lighting information, previously would have required a fleet of large and expensive machines. And that doesn't even include transparency and applying real-world attributes of lighting and reflection onto the USDZ file, a a task that's hard to parallelize.
garlic/onions: aliums that thrive in most climates
rice: grass that can be drowned (to fuck off pests)
eggs: per day, one chicken produces enough binding enzymes to thicken and modify a meal for 5-8 people
veg/seed oils: cooking, flavor, etc
and since that's in the caspian sea area, they also have a ready supply of tomatoes and eggplant from centuries of trading with asia
the burning answer is really, by the time you've chosen a modern merchant soc, and put a screen on it, you're really more or less at the same BOM cost as a very low end android device (and frankly, the soc bsp you chose is likely to be android/aosp to begin with)
so you're basically competing head to head with every shitphone maker out there, and they have economies of scale. you do not.
> Why can't somebody take simple smartphone hardware
and so... now you're competing with simple smartphones. you're going to charge more for a fancy case for something that requires to be carried separate from your phone and offers zero additional advantage?
there's a huge market for external mobile DSPs for this reason. this is an example of an entry level one https://www.microcenter.com/product/705193/moondrop-freedsp-...
btw, im sure someone like shanling is making exactly what you're thinking of, but they often have limited production runs and still cost within the same range as a basic android phone (usually a bit more)
No, but a small university team with dark time on the state school system's cluster can build a basic, functional LLM.
It will be a few years (or more) until true "open source" LLMs are available and high quality, for sure. But it's not as impenetrable as it seems, IMO.
He spent an entire weekend before having the wisdom to pause, and let someone else contribute their time to finding a counter.
Just because they're exclusive about their clients doesn't mean they're not real. Their impact and effectiveness is a separate topic though. I don't think they're still actively operating or taking new clients, at least.
btw: "Oh, so you've done x?" What a snarky and confrontational way to ask someone something. Especially when it's asserting a well documented company and person is "not real".
It still won't be as big of an issue for night sky photography, for reasons mentioned in the parent comment.
However, there are many other severe issues, of course. But for hobbyist and professional photography? Mitigable.
I can see a problem with 100x the current deployment, but the "other severe issues" kick in long before that, anyways.
Starlink gets a lot of attention because it's new, and because it's run by a polarizing guy who sucks.
Meanwhile, the airplane and streetlight situation has been destroying night sky photography for decades, to the point where photographers consult charts and have communities tracking the best times of years to minimize the hours of driving you have to do.
But let's "both are real issues" it, sure.
Also, shipping winbox on all platforms has been a huge QoL improvement
what differentiates you? you could tell me "type xyz into chatgpt" or even share a structured document link from their site. but when you're copy/pasting, then you're basically useless in the equation.