• Welcome to ASR. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Master AI (Artificial Intelligence) Discussion/News Thread

Recently introduced Agentic AI models are going to be a boon for busy CEOs. Now a CEO can conveniently ask the Agentic AI to e-mail oh so polite layoff notices to all other current employees, and bundle a free therapy session with said Agentic AI. The e-mail will explain that the each employee is being replaced by AI Agents trained on data collected while monitoring the employee over the past year. The therapy session will be limited to a million tokens, in order to prevent the employee from spending all weekend tokenmaxxing with the choicest expletives they know.

Does this post belong in the "Useful Applications for AI" thread instead?

By the powers of AI invested in us all, we summon forth the ghosts of Elias Thorne and Mara Vance, and command them to persuade the AI Agent to e-mail a notice of a tripled annual bonus instead of the layoff notice.

Why LLM’s Suck At Being Creative
 
For those of us with a hankering for running AI models locally on our own computers, last month we were promised further increases in the price of RAM (DRAM, SSDs internal and external) and graphics cards. This month, sellers are beginning to deliver on those promises.

In particular, the AMD Radeon Pro AI R9700 graphics card, which sports 32 GB of VRAM, was selling in June and July for under US$1400. Earlier this month, the price jumped to $1500, and stock of various brands (ASUS, Gigabyte, ASRock) was spotty. The ASRock and Gigabyte versions are now out-of-stock. Amazon (US) has not canceled my order, which it still says will arrive sometime during the first two weeks of September. The ASUS-branded version came back into stock at just under $1700 a few days ago, dropped temporarily to $1500 for a couple of days, and is currently listed at $1632 on Amazon. Even at that price, two of these cards is still much cheaper than a single NVIDIA RTX 5090 which also has 32GB onboard VRAM.

The competing Intel Arc Pro B70, also carrying 32 GB VRAM with similar bandwidth, which a couple of YouTube reviewers said was a steal at its list and selling price of $950, has within the past week or so jumped in price to $1300. All these price hikes are anticipatory, as far as I can tell. The items are not actually new stock incorporating the higher NAND prices that the memory makers are charging NVIDIA, AMD and Intel.

The next tier up are the unified memory systems, which offer a 128 GB unified pool of fast VRAM and slow DRAM, from NVIDIA, AMD and Apple (DGX Spark, Strix HALO AI, M5 Max (?) or M4 Ultra). These will load even bigger AI models (up to about 90 billion parameters), but are much more expensive than the 32 GB cards mentioned above which would load models up to about 28 billion parameters. Quantization of models to smaller word sizes can help somewhat.

No matter what, none of these cards can run the latest open-weight models (1 to 4 trillion parameters) locally. So for the latest and greatest, unless you have a business need for privacy, we are still dependent on running our queries in the AI clouds, and paying the per-million-tokens cost.
 
My congratulations on your implied achievement of typical consumerism. Something to aspire to, I s'pose. The economy of scale is very unclear at present.

Diff'rnt strokes for diff'rnt folks. Horses for courses. I want my personal computer to do multiple things, such as run compute-intensive engineering simulation codes, that a cloud AI subscription is not needed for. The graphics cards I mentioned in my previous post can speedup engineering simulation codes, and run AI models for relatively small marginal cost, with the involvement of a CPU of course. It all depends on needs, means, available services and their costs. Before retirement, I used to run engineering simulations on workstations and supercomputers. I want to continue those interests, though at a lower level on a consumer PC with a beefy CPU and GPU.

I am not a typical consumer, and my post was not aimed at typical consumers who may not care to run local AI models. I started my original post with "For those of us with a hankering for running AI models locally ...".

We all understand that the big AI players are expecting to rake in trillions of dollars from typical consumers who will pay for cloud AI services.

So what even is your point?
 
Last edited:
Anthropic and OpenAI seem keen to persuade the US Gov that AI technology is too dangerous for the public at large to have unfettered access to open-weight models. I think they would like to eliminate competition from open-weight models that might prevent them from establishing an effective monopoly or duopoly and garnishing all customer interaction data to boot. They are expecting to collect hundreds of billions of dollars from consumers, companies and governments. All the more reason to run local AI models and agents, and keep your code and data private.

Apparently, you can yoke up to four Intel Arc Pro B70 together, or two Radeon R9700 together, to load larger AI models than with single cards, depending on the software you use.
 
Last edited:
There are some bundled software here for pennies on the dollar and it helps charity. I got some CAD programs here, it's legit. FYI they keep changing with AI and coding stuff all the time.

Software

Brush up on your design skills, learn a new coding language, develop your own game, and more with our exclusive pay-what-you-want software bundles.
www.humblebundle.com www.humblebundle.com

All Bundles

Our Bundles are limited-time collections of games, books, software, and more. Pay what you want and choose where your money goes, including to charity.
www.humblebundle.com www.humblebundle.com
 
1787544060136.png
 
tbf that cartoon doesnt explicitly say AI caused the drought either...
 
TonyJZX said:
tbf that cartoon doesnt explicitly say AI caused the drought either...
Click to expand...
It shows AI sucking all the water out of the ground. and it is titled "Ella Baron on AI and the drought" What else is it trying to say?
 
... That valuable resources are being squandered on AI?

That the massive heat generated by AI helps accelerate climate change?

... Whilst mostly being used by punters asking millions of random questions
... rather than doing important things like curing diseases?

... And frequently giving responses, primarily aimed at pleasing said punter

(Or something entirely different!)
 
Last edited:
yeah to me its open to interpretation

i'm not the kind to rigidly force my interpretation on everyone else but ya know, there's all sorts... one thought one mind one volk

my take is that the AI is being deliberately facetious, every AI does this but at the end it offers no answers while still being "not a part of the problem".. but in reality is "THE PROBLEM"
 
Gift links to how A.I. is shaping the use of drones in the war between Russia and Ukraine.

www.nytimes.com

Some of Russia’s A.I. Drones Are Powered by Nvidia Microcomputers, Ukrainian Officials Say

The tech giant says it does not sell the devices in Russia, but they are widely available on resale markets. When purchased that way, they are virtually impossible to track.
www.nytimes.com www.nytimes.com

www.nytimes.com

A Drone Killed Three Ukrainians. It Was Guided Entirely by A.I.

An attack by what Ukrainian officials said was a Russian drone with an Nvidia chip presages a dystopian future of weaponry untethered to humans.
www.nytimes.com www.nytimes.com
 
You must log in or register to reply here.

Similar threads

amirm
  • Sticky
Useful Applications for AI
9 10 11
Replies
209
Views
17K
Zetti
Z
2Sunny
Did Google AI get this correct?
Replies
8
Views
1K
amirm
amirm
TLEDDY
Hantavirus outbreak on ship; discussion
3 4 5
Replies
95
Views
8K
JustAnandaDourEyedDude
J
L
Room response simulator.
Replies
7
Views
1K
voodooless
voodooless
H
Binaural Geometry Mapper — an experimental audio measurement project
2
Replies
28
Views
2K
HaK77
H
Back
Top Bottom