Time to Embrace AI: Entry Level Machine Learning Hardware is Cheap!
For the longest time had scoffed at the incoming artificial intelligence revolution we're just starting to see form into a mass adoption phenomena.. Actively both fearing the disruption it could cause while at the same time underestimating it's use cases or effectiveness to change the world..
This is no longer the case, AI and the computer sciences powering it are more than here to stay and will almost certainly become one of if not the most important invention we've had since harnessing electricity and silicon. There are very few tasks a well trained AI or AI powered robot will not be able to outperform or dollar cost ratio crush human intelligence at.. Which is both a scary and incredibly exciting thought. Guys like DARPA, Google, Amazon and Boston Dynamics have been working on GAI (General Artificial Intelligence) and robotic bodies to make their code come alive in the physical space for a long time now and we're basically on the cusp of it being ready. Cue your fears of SKYNET type situations happening, it could be a reality within the next generation or so!
But all fears and jokes aside, the technology and ability to use AI to enhance human invention, task handling and to diagnostics are freaking nuts! There is a zero percent chance this stuff won't disrupt nearly all industry and change how we all live even more than it is already.
Nerd Shit, Because Learning & Engineering
For most folks, they really don't care how things work, just that they work.. For myself I love learning and understanding how things are engineered and function, so decided to finally dip my toes into the field and get some hardware to begin making my own microscopic sized AI development capability a reality. For the low low price of around $90 USD I took a chance and managed to grab a few prior generation Nvidia M40 machine learning cards. These are basically the Datacenter Titan X from the research I have done looking into the things.. Either way at friggin' ~$45 USD a pop? Can't say no.
While these cards were originally used in server farms / data centers that used forced air cooling and thus have no video output or built in cooling fans, that is a pretty easy fix with the application of some 3D printed fan shrouds and a bit of tinkering.
Will These be Used to Take Over the Planet?
Now are these cutting edge technology? Fuck no.. But what they are collectively is about 6000 CUDA cores and 24GB of potential machine learning, data computing and cryptography hashing units. At around 250w peak power a piece the power requirements are up there, but when you compare the price of the newer hardware which is far more capable but also 10x - 50x the price per computation power, this should be an interesting experiment. Is there hardware that vastly outperforms this stuff? Of course, but compared to my AMD GPUs that I've been running locally, these older Nvidia units open up a whole new array of machine learning libraries and whatnot to play with.
The biggest limit on these units is the RAM amount bordering on the smaller side, as well as the lack of the now commonly used TENSOR cores that a lot of the newer units have.. But at 1/10th the price of a single current generation card I thought it might be a sufficient starting point.. and if nothing else a fun project to print the fan shrouds and get them operating on my workstation, vastly increasing the ability to do compute intensive activities like AI art through Stable Diffusion, creating stuff like "deepfakes" using really powerful tools like GeneFace, hash cracking with stuff like aircrack-ng or simply getting my feet wet creating small neural networks and training them. If I end up getting deep into the ML stuff then great! If not it was like $100 or less to try it out and have something to tinker with or feed hashes to if I wanted to bust some light cryptography.. Either way I've spent far larger sums of money on far less useful shit, and I think the investment to see what exactly can be done and learned using these things is worth it in the end.
In the end, if you don't understand and embrace technology, you will ultimately become a victim in some way to it.. With this ethos and me finally giving up my hate for AI / ML stuff I think the stuff that comes of it in the future could potentially be awesome. Thanks for reading and voting!
Long ago, i had a brush with bureaucracy.
At that point, i knew that i could replace half of the employees with a computer program.
The reason to not just whole hock replace humans with a program is that the human can detect errors that are outside the scope of the program. The problem is, we have dumbed down the people, dumbed down their jobs, so, we do not have that error checking. In fact, a human may continue to make the same mistake over and over, because they have been told to do it that way, even though they know its wrong.
So, all those people who just fill out paperwork to send to people who file paperwork.
All of their jobs could be automated.
What we are left with is a weird niche, where there are people that are better than computers, but we would never pay them to do what the computer can 99.9% do.
So, we have people who hand carve works of art.
And we have CNC routers that router out the same product over and over again.
Same old story, new field.
The biggest problem with GPAI is humans.
Humans expect a smart "being" to know what a door is, to know if you should do something or not.
A computer doesn't know and doesn't care.
Really, an AI has no idea what a door is.
It can make you think it does.
It can usually identify a "door" in a picture, but it has no idea what that is or means.
So, when you are reading some replies on big tech things, like tweets, you will see bots replying trying to change the overall feelings. But, if you notice, these bots do not really know the context, and can really get that wrong.
Another death in Ohio is just a +1 to memory A1FF to a computer
But to a human, that is a tragedy, and comes with thoughts of "That could have been me"
AIs and Narcissists can only approximate an empathetic response.
When I saw shit like GPT-J being trained on coding documentation and stack overflow with the outcome being the thing capable of taking text prompts then sometimes creating tangible code I started taking note.
At this point I'm fine with coding being less human intensive but it's still not 100% useable at this point. Hell even Chat-GPT has been used recently to help supplement coders ability to crank out functions and infrastructure code.
I think we're one of the last generations to have to manually peck at the keyboard for hours / days / months / years to make useable software.. Will still need people who understand the stuff for some period of time, but when it gets to the point where the AI can crank out code faster and cheaper than a team of devs and you can turn teams of 100 into teams of a few folks, it'll be interesting.
100% in agreeance about the approximation mention. There is a huge line between understanding how things operate and having a neural network pick out patterns and regurgitating them. Lots of neat magic and parlour tricks out there at the moment, but publicly there isn't really anything that can replace the workforce in all industries yet.. We'll get there in the next 10 - 20 years though from the current rate of progress.
You only need one real GAI to completely change the game, whoever develops that shit first will essentially automate thousands of years of human progress in 1/100th the time given enough processing power and resources.
Artisan work will still be a thing for a bit, but we're on the precipice of "all the things" being able to have the computer do the heavy lifting leaving the worker / consumer in a state of obsolescence.
An area that GPT and deep fakes could do is be an OnlyFans girl... without the girl.
So much of the texting/chating is already handled by worker drones.
I disagree. For a great many tasks I do, which require packing heavy things up ladders, or tunneling underground, and then manipulating items as fine as needles, that flexibility isn't going to soon be potential to electromechanical devices. Additionally, my understanding of the state of the art leaves me convinced our almost complete incomprehension of consciousness leaves AGI utterly impossible presently, and there seems to be no route to it being developed, specifically because we don't even know what 'it' is.
Neural nets will never achieve it.
There are some tasks to which algorithms are suited, and I think ChatGPT and Stable Diffusion are leveraging those capabilities. Painting gables will likely require a dedicated vehicle and algorithms, or several (setting up a ladder is a far different task than climbing one with a paintbrush), so one depriving me of that fun and profit will only arise if a market for it is profitable. If that's all I did, then I'd see that a market did exist and worry. Since I also do an incredible variety of stuff, including tunneling underneath structures that support for has collapsed in a flood, and installing laminate on countertops, the flexibility of the human body and our ability to deploy conscious thought and not just search for published solutions really eclipses anything I think can be achieved by robots and AI before I am dead and ashes.
Something I think will revolutionize decentralization is automation of production for 3D printers, aquaponics, power production, and the variety of decentralized means of production being developed that presently require laymen to climb learning curves to deploy. Managing a suite of decentralized means of production to meet continuing needs (without exceeding them and creating waste) also comes to mind. An app that diagnosed illness would also dramatically change the medical industry and reduce our vulnerability to it's increasing militarization, as well as printing with molecules to create our own pharmaceuticals.
Screw Big Pharma asswise.
Another is enabling extant hardware to extend it's utility. For example, the Carbon X1 3D printer has lidar. This is used to ease bed leveling, and to check 1st layer adhesion and conformance. However, lidar can be far more useful, and applications of lidar to additive manufacturing can make build plates obsolete, eliminate build volume by facilitating robotic arm accuracy, or facilitate periodic repairs of parts that wear, preventing wholesale replacement, and many more such things, particularly in concert with computer vision, and 3D scanning. I think lidar is an underappreciated technology today, and I hope to see it quickly become far more useful, particularly in additive manufacturing.
These are some ideas I hope will enable you to profit greatly from increasing the development and dispersal of decentralization to more rapidly usher in the coming paradise. I know you code, and I want code that does these things! However, I have no doubt you have more than enough ideas to keep you happily busy with your new investments, so disregard anything I have to say, which is, oddly, what the most successful people I know seem to do. =p
Thanks!
I think the trades are safe up until they start normalizing 3D printing houses and shit.
They already have surgery robots that use human input to do very fine surgery shit over the internet.. Nothing the doctor does in that situation an AI won't be able to be trained on.. Sure maybe you keep one person in the loop or something, but as far as grunt work and shit goes.. There will be no need for it.. Probably why we are seeing the mRNA depopulation shit going on imo.
You have some points there.. The cost to develop a countertop fitting bot and shit would be impractical for the time being. As a person who wears many hats myself I know that half of what I do can be replaced and I'm ok with that. Not all the tasks in the day I do or whatever are enjoyable.
3D printing / additive manufacturing fucks (rocks) for lack of a better term.. the ability to prototype things at home or make molds for casting stuff taht plastic isn't a proper material is nuts. On the decentralization idea I 100% agree. relying on the powers / corpos at be is a hell of a bad idea.. as they have not our best interest in mind but control, profit and power.
Funny you menti9on the diagnosis stuff.. Actually have a buddy that is working on that for medical imaging and I'm helping with the training of that, guy has the resources to build top of the line compute machines and the amount of training data to pull it off.. Will be a hell of a project. Squaring off against guys like GE medical division is appealing to me.
LIDAR is fucking nuts... Hell a lot of the EMF/RF shit they have now a days both in public and black projects would put 1984 to shame. Won't get into the "conspiracy woo woo" shit but if people knew the capabilities of EMF spectrum manipulation and the way they are spraying chemtrails into the sky to increase the conductivity of everything they'd shit themselves. Hell even working on RADAR stuff locally here, with a few software defined radios and some craft algo implementation even things like WIFI cards can actually become sensing units!
Coding the stuff and learning to overcome the challenges that putting ideas into creation is half the fun for me.. If it was easy everyone would develop stuff. It's an ability to apply ingenuity.. and that is where I find meaning in life.
lol, as far as comments go you're up there in the show of thought, it's never mindless shit and I always appreciate the push back and alternative view point to help keep my own views more realistic or atleast see what someone else thinking about this shit is working through objectively. Cheers!
The laws of physics are unequivocal. It's very simple to show that if A-B=C then C+B=A. Every microphone is a speaker, and every camera is a monitor - and vice versa. More doesn't need to be said.
Another person that has received comment from me put things a little differently. He said I was 'out there'. LOL
https://github.com/facebookresearch/llama/pull/73/files
Maybe this will help.
Edit: just found this today:
https://github.com/Shaheer-Zeb/TechTrim-Ai-Writing-Tool
Re-Edit: Thought you might be interested in this, too.
https://github.com/determined-ai/determined
Hope these help
Neat share of projects. Much appreciated!