AI Chips in Computers

PaulaJedi

PaulaJedi

Survivor
☀️ Zenith
AMD is launching AI chips now. What could go wrong? Granted, they aren't quite that powerful yet, but if we accidentally make them too smart, they'll start changing passwords and such, wouldn't they? Yes, we all say that AI only takes instructions from the programmer. If that's all it does, then it isn't AI.

You have reached the age of AI. I hope it's worth it.

www.oann.com

AMD launches new AI chips to take on leader Nvidia

AMD unveiled its latest artificial intelligence processors and detailed its plan to develop AI chips in a bid to challenge industry leader Nvidia.
www.oann.com www.oann.com
 
Technically, this is an AI "accelerator" chip and cannot run standalone AI applications by itself. The applications still need a full computing environment. A regular desktop computing environment could run the AI application just as easily, but it would be much slower. A phone computing environment could also do the same, except it would be painfully slow, probably to the point of not being useful or practical.

The accelerator chip offers common math functions that AI applications need in a massively paralleled environment to get the speed ups. As AI grows, it needs more advanced math functions, so Nvidia has been offering more advanced accelerator chips to meet demand.

ARM offers simple AI accelerator cores for mobile devices, but their use is very limited.

AMD typically caters more towards budget devices, so expect AI applications to grow even more over the next few years. The AMD VS Nvidia graphics wars have been going on a long time and have followed this pattern. Nvidia usually has the better product that is more capable, but they're expensive. AMD intends to cut into Nvidia's margins. I'm kinda surprised that AMD is this late into the game. Since AMD jumped in, Intel won't be far behind.

As far as Skynet being created, Pandora's Box has already been opened.
 

MODAT7 said:
Technically, this is an AI "accelerator" chip and cannot run standalone AI applications by itself. The applications still need a full computing environment. A regular desktop computing environment could run the AI application just as easily, but it would be much slower. A phone computing environment could also do the same, except it would be painfully slow, probably to the point of not being useful or practical.

The accelerator chip offers common math functions that AI applications need in a massively paralleled environment to get the speed ups. As AI grows, it needs more advanced math functions, so Nvidia has been offering more advanced accelerator chips to meet demand.

ARM offers simple AI accelerator cores for mobile devices, but their use is very limited.

AMD typically caters more towards budget devices, so expect AI applications to grow even more over the next few years. The AMD VS Nvidia graphics wars have been going on a long time and have followed this pattern. Nvidia usually has the better product that is more capable, but they're expensive. AMD intends to cut into Nvidia's margins. I'm kinda surprised that AMD is this late into the game. Since AMD jumped in, Intel won't be far behind.

As far as Skynet being created, Pandora's Box has already been opened.
Click to expand...

That still makes the AI smarter. At what point does AI become "too" smart?
 

A smart AI at risk to take over won't happen in your computer or mine. Those "AI chips" (big words here) are hardware facilitators and accelerators that make AI code run faster and more efficiently on your computer. It doesn't make it anything more than it is.

Right now, I don't consider the existing AI smart. They're extremely good at various things. Hacking things together, mixing ideas and concepts. Programming. Suggesting stuff. But in simple terms, all they do is pick existing stuff and then rehash it into something else. As I said in another thread, they do not actually create anything. They remix and mimic.

An AI that's able to actually upgrade itself. Not learning, but upgrading itself, and stretching the boundaries of its own faculties will certainly be our doom. Let alone an AI that would be able to upgrade its hardware? Game over, man!
 
Num7 said:
A smart AI at risk to take over won't happen in your computer or mine. Those "AI chips" (big words here) are hardware facilitators and accelerators that make AI code run faster and more efficiently on your computer. It doesn't make it anything more than it is.

Right now, I don't consider the existing AI smart. They're extremely good at various things. Hacking things together, mixing ideas and concepts. Programming. Suggesting stuff. But in simple terms, all they do is pick existing stuff and then rehash it into something else. As I said in another thread, they do not actually create anything. They remix and mimic.

An AI that's able to actually upgrade itself. Not learning, but upgrading itself, and stretching the boundaries of its own faculties will certainly be our doom. Let alone an AI that would be able to upgrade its hardware? Game over, man!
Click to expand...

Alright, but this is the beginning. We are at the infant stage. More will come.
 
As Num7 stated, the "singularity" point is when AI learns how to upgrade itself. The first upgrades are likely to be messy and not very good. When it figures out good programming practices, the next round of upgrades will be extremely dangerous.
 
MODAT7 said:
As Num7 stated, the "singularity" point is when AI learns how to upgrade itself. The first upgrades are likely to be messy and not very good. When it figures out good programming practices, the next round of upgrades will be extremely dangerous.
Click to expand...

Do you remember when, a few years ago, Facebook created 2 AI computers and they were chatting with eachother, and suddenly they developed their own language that they couldn't understand, so they turned the computers off? Now, they are denying it.

www.usatoday.com

Fact check: Facebook didn't pull the plug on two chatbots because they created a language

Facebook did have two AI-powered chatbots that learned to communicate more efficiently with each other. But Facebook didn't shut them down.
www.usatoday.com
 
Oh yeah, I think I remember. But I didn't know it devolved (or did it actually evolve?) into a weird code language.

I'll have to check it out, that looks fascinating.
 
If computers could reach singularity in isolation, then the human brain would have done so long ago. The problem is that intelligence is more about getting lots of reliable information to work with than the amount of computational power. A computer in isolation can not know if its conclusions are correct or not when based on human observations entered into the system.
 

Top