I'm skeptical that Miso Robotics can actually deliver a useful product that outperforms or matches human labor at the same price point. They don't show clear videos of what exactly the robot is capable of, and keep buying Instagram ads to invest in their company.
White Castle already ran a pilot with the Miso 1 and then the Miso 2. It seems they see value. Maybe the tasks are different for White Castle with many small burgers vs the fewer large burgers at other places?
The video that exists shows the Flippy only making french fries [1]. No burger cooking at all. And if you've seen how they do the burgers at WC [2], that's no small task for a robot yet.
The french fry station looks like an ideal place to start, anyway, since it's usually a task that doesn't get a dedicated worker unless the place is extremely busy. In the quieter times it's something nobody has time to work or monitor while doing other things. And then, suddenly, everyone is waiting on the fries to finish.
From a more critical POV, it's kind of a kluge but at least a human can take over. There have been products for a long time [3] that are more self-contained and can do smaller batches of fries but need a hell of a lot more maintenance and cleaning.
From the video, it actually seems as though White Castle burgers are even more optimized for process automation as their preparation doesn't even require flipping (TIL).
Ah, older articles were talking about it doing burgers too at white castle. This one only mentions the fries but doesn't mention burgers. Looks like maybe they transitioned to fry station only?
And the White Castle is easier to automate than other burgers. Small, sure but also limited toppings - no choices, no sauce. (McDonald's is surprisingly complex by comparison)
Walgreens run pilot with Theranos and see where it all went. Running pilot is usually done on the best machine they have under constant supervision of the second in line to the tech who built in. Different than deployment off the production line hundreds of machines without even knowing a real-life breaking ratio. Also the cost of operation is somewhat at $3,500 or some $17 per hour for a full shift, so much more than average cook takes (excluding tips which this machine won't get). Not to mention cooks very rarely just cook... they do all kind of in kitchen tasks that this machine cannot.
I guess we'll find out since this one is moving beyond the pilot. $17 per hour is less in many areas, considering cost to the company for a person also includes additional things like payroll taxes.
> The #1 problem with PyTorch is that it’s great if you want to use one videocard for training
Incorrect information so confidently stated here. Tons of research papers that use more than one GPU for training, not sure what you're referring to? Standard DDP works fine, for starters.
Will do - as I mentioned in another comment, it can be a bit subtle to find an apples-to-apples comparison, but we'll soon add some cross-platform that we think are reasonable.
I would argue that the problem is Beyond Meat is only kinda close in taste. Impossible meat is getting good reception because it's so close in taste (and I actually prefer it), and lots of other veggie burgers are doing okay because they fill the same use but aren't going for exactly the same flavor. Beyond Meat is almost in the uncanny valley of beef flavor.
The site and search eventually work, but the torrents themselves do not download and do not appear to have all necessary magnet information in the links. So the downloads just hang there in the torrent client.
My workstation build with Epyc is ~$5,000 and has more cores (24 core), more memory (256 GB), faster GPU (3090), a 2TB pcie 4 SSD and I suspect will perform better on standard benchmarks. Definitely not as compact as the Studio though but lot more extendable.
> I suspect will perform better on standard benchmarks.
I wouldn't be so sure about that. These M1 chips have some crazy benchmark performance.
What they don't have, is a standard x86 ISA, which means that a lot of applications run emulated (and still tend to beat many Intel specs, anyway). I keep reading people complaining that "It's faster, but not that much faster. What's the big deal?", when they are talking about an i86 application, running on Rosetta2.
So given the Epyc has 24 cores versus 16 cores for the M1 Ultra, the Epyc has roughly 75% of the single core performance of an M1 Ultra and 115% the multi-core performance.
I feel that the M chips are awesome at saving power while providing performance which makes them perfect for laptops and portables. I’m not seeing their value proposition for desktops though. I don’t care about something that saves space. The only time I care about that is for portables.
Remember when people were talking about "ARM rack servers"? The idea was to have dozens of daughterboards, jammed into a 2U chassis, providing massive parallelism. The low power draw would make this possible.
Wouldn't surprise me, if Apple is thinking about doing something like this, maybe with a new API/SDK, focused on this. They would probably sell it as an ML machine.
The issue for that is that it’s not cheap, commodity hardware. It’s a risk to build a server farm with Apple given their historical penchant to arbitrarily cancel entire product lines that were doing ok. Apple is best as a workstation or personal machine.
imo this is just an improved version of garbage can Mac Pro
They refuse to build what a sizable number of people want, yet we're stuck with Apple because the alternatives are even worse. That's what pg got wrong with one of his predictions. He didn't realize that no one can do better than Apple's worst effort
Neural Engine cores are not accessible for third party developers, so it'll be severely constrained for practical purposes. Currently the M1 Max is no match for even last generation mid-tier Nvidia GPU.
