|
Think we could have had it by 1980,
given enough funding and someone with the gift of hindsight. Not sure as a hobbyist thing, given the still nasecent Internet and the nature of funding. But I say yes, technologically speaking, based on the story of how they got the pre-GPS Etak Navigator to work, which still blows my mind. https://www.fastcompany.com/3047828/who-needs-gps-the-forgot... Though that has more obvious advantages as a product that consumers want which translates to an more possibility for funding. There's a clear need for a device that gives directions, vs a 3d printer. While I'm very happy with my Bambu 3d printer, it didn't scratch an actual need in my life. Looking for similar opportunities today, we have access to relatively cheap compute power. A 3d printer in the 80's would not have been all that cheap as a consumer device. Today though, we have 3d printers and relatively cheap (for now) parts to drive them - motors/gears/belts/UCs. While LLMs have taken up all of the air in the room, they're not there only ML technique out there, and while GPUs aren't cheap, they're fair cheap considering the cost of a Cray, back when they made them. Looking those things, I wonder the number of parameters you'd need for a system where the AI model is continually being updated as new information comes in. What size pet robot would be feasible, backed by a 4090? 5090? 10090 or whatever it's called in six years? Could you make a robot planaeria that had basic sensors and responded to limited stimuli? Robot drosophila? Could you make a robot snake with cameras for eyes? Of course, where I'm leading is a robot dog or cat with an AI model that learns from input given to it by its operator. Not practical enough to get much funding given the state of the industry; the market for a $20,000 robot dog is too small. The technology is only somewhat there, or about to be, and I'm sure it'll be a "why'd that take so long" after the algorithms for it are invented/discovered which makes it seem obvious in hindsight. I'm sure there are similar opportunities (the army rejected a robot pack mule for being too loud, but how fun would that be?) Or a food replicator - there are some 3d printers that do chocolate but nothing with an AMS. or how about a specific type of home robot food machine. How about a machine I put flour and water, a block of cheese, tomato sauce, and a pepperoni into, and out pops a ready to eat pizza or calzone? What else could you make an advanced device for egg ingredients go in, and it cooks a finished product? Thing is, with the Internet and especially YouTube, I'm sure there's someone trying to build one (I saw the vending machine thats basically that, last time I was in Vegas, but it's way too big and expensive for a consumer product.) Or a robot arm that plays chess or something else where the arm only needs to manipulate the things in front of it and doesn't need a moving chassis, so you can play chess against someone remotely, without being on a computer. Robot arms have existed since forever but it's only recently that we've gotten enough compute to make controlling them easy for an end user. I keep going back to robots because parts are 3d printable or source able via Amazon, and since 3d printing's really come into being lately, that's would be the thing to take advantage of, in addition to relatively cheap compute. Looking just a compute there's a load of stuff that involve helping users automate things on their computer, but that's so obvious a use case that many well funded players are after it. The problem with distributed computing is that the interconnect speed between nodes on the Internet is just too low to make it useful, so since Folding@home, there hasn't been one that garnered the same popularity. But we have this global scale network of computers. After the appropriate algorithms and a use case for that is found, it'll be forehead slappingly obvious in hindsight. Maybe as a better AI for civilization games, which afaik still just give advantages to computer civilizations in order for them to be more difficult. If each computer controller race had their own computer to run on, how would that play? |
Technically, not speed low, but latency is huge.
For about speed, if you lucky, could build 100Gbit fiber to nearest exchange and this is comparable to speed of desktop RAM (around 40Gbytes, so just 320Gbit). But RAM latency is somewhere around 10ns and on fiber internet practically achievable around 10us to nearest DC (1000 times more).
With 3G/4G/5G latency around 1ms achievable.
Anyway, main question is will to do something. - If person don't want to do, he will not, even if have all need.