• Security incident: ISF was recently accessed by intruders. Please change your password, and change it anywhere else you used it. Read more

Merged Artificial Intelligence

JayUtah said:
Just a quick update on the Utah data center: O'Leary still wrongly maintains that all the opposition to him is foreign and paid-for, but the legislature and O'Leary are trading proposals to shrink the proposal to something less than apocalyptic size.
Click to expand...
He'll soon be happy he couldn't get it built at the scale he wants.
 
I think the future of AI is local models that get updated periodically like any other piece of software that you no longer truly "own". The biggest advantage of this is that it transfers the massive resource cost of running a model onto the consumer. Not only does this benefit the AI manufacturer, it also gives the customer more control over how much resources they want to invest in using their local AI.
 
theprestige said:
I think the future of AI is local models that get updated periodically like any other piece of software that you no longer truly "own". The biggest advantage of this is that it transfers the massive resource cost of running a model onto the consumer. Not only does this benefit the AI manufacturer, it also gives the customer more control over how much resources they want to invest in using their local AI.
Click to expand...
Proprietary local models won't work. Each time the US companies have started parceling the space, an open Chinese model sweeps along and upends everything, with open source hobbyists close behind.

That and companies are desperate for proprietary data. They all want to be the model that everyone uses because it has the edge, which it maintains from the data of everyone using it. That means getting it off your computer and onto their server. They're getting it by using AI for everything right now because it costs them pennies on the dollar. When the AI bubble pops that'll change, companies are going to have to charge what it actually costs which will limit the commercial space to a significant extent, but data centers will still be the dominant lifeform.
 
Beelzebuddy said:
Proprietary local models won't work. Each time the US companies have started parceling the space, an open Chinese model sweeps along and upends everything, with open source hobbyists close behind.

That and companies are desperate for proprietary data. They all want to be the model that everyone uses because it has the edge, which it maintains from the data of everyone using it. That means getting it off your computer and onto their server. They're getting it by using AI for everything right now because it costs them pennies on the dollar. When the AI bubble pops that'll change, companies are going to have to charge what it actually costs which will limit the commercial space to a significant extent, but data centers will still be the dominant lifeform.
Click to expand...
Companies can backhaul the data they want from your local model activity, just like they already backhaul the data they want from your browser and OS.

I'm telling you, local models with proprietary subscriptions are the future. Sure, there'll be some people using cheap Chinesium models, or open source hobby models, but the vast majority of users will just use whatever model subscription comes bundled with their OS or their browser. Saying proprietary local models won't work is like saying proprietary local operating systems won't work.
 
nVidia clearly pushes (also) for local models .. RTX Spark is all about that. And current models in 4GB range are getting quite decent.
Not for serious coding, for sure .. but good enough for basic text processing and natural language interaction with the computer.
 
Trump says he's meeting with the top 12-15 AI executives "very shortly" to discuss giving the public a stake in AI.

"If we do that, the public will become very rich. That's the kind of money we're talking about. I think they'll do that and I think it'll make it very popular."

Video in link
 
it makes so little sense. like, how does the public get rich? the ai companies get a bunch of taxpayer cash in exchange for shares, and the government just sits on the shares indefinitely? or leverages against i? are they going to collect dividends and then the public gets a check? or these guys just get billions, pumps their stock, they get to be the kings of america?

sounds more like bailout.
 
dirtywick said:
it makes so little sense. like, how does the public get rich? the ai companies get a bunch of taxpayer cash in exchange for shares, and the government just sits on the shares indefinitely? or leverages against i? are they going to collect dividends and then the public gets a check? or these guys just get billions, pumps their stock, they get to be the kings of america?

sounds more like bailout.
Click to expand...
It's a ponzi scheme TACO is foisting on the people of America.
 
dirtywick said:
it makes so little sense. like, how does the public get rich? the ai companies get a bunch of taxpayer cash in exchange for shares, and the government just sits on the shares indefinitely? or leverages against i? are they going to collect dividends and then the public gets a check? or these guys just get billions, pumps their stock, they get to be the kings of america?

sounds more like bailout.
Click to expand...
They stand to make as much money as they did from sanctions.

Meanwhile the real world would like a word: (longish read but worth it)
www.faros.ai

The AI Engineering Report 2026: The AI Acceleration Whiplash - Ten Takeaways

What two years of telemetry data from 22,000 developers reveals about AI's real impact on developer productivity, code quality, and business risk in 2026.
www.faros.ai www.faros.ai
 
I just found that "A new cognitive evaluation of artificial intelligence has unmasked a fundamental, systemic flaw running through Large Language Model (LLM) attention mechanisms."

Here is a link.

The flaw was discovered by administering a classic psychological test known as the Stroop task to LLMs.

The Stroop test consists of a series of color words that are displayed in colors that do not match the word, such as the word "RED" printed in the color BLUE.

The test requires the user to identify the color of the word, not the color it specifies.

This mismatch creates cognitive interference that must be overcome. This takes time that can be measured to check for cognitive decline.

Apparently a large range of LLMs have serious problems with this.

When presented with the list of color words, the LLMs get the first few right, but then their accuracy quickly declines.
 
steenkh said:
I didn't know LLMs could see colours.
Click to expand...
Multi modal ones are trained on images and videos as well as text so you can input an image and ask them about it. E.g.:

1781371396428.png
I can ask Gemini to describe that image:

The image is a cartoon-style illustration featuring two disgruntled-looking trolls sitting under a stone bridge, fishing in a shallow stream.​
​
Here is a breakdown of the scene:​
​
  • The Left Troll: He has green skin and messy orange hair. He is holding a fishing rod, and hooked to the end of his line is a blue bucket labeled "CAT LITTER".
  • The Right Troll: He has greyish-blue skin and dark hair, resting his head on his hand with a bored expression. His fishing line is hooked to an open laptop with the letters "HB" on the cover.
  • The Setting: They are sitting in a stream that contains some debris, including two tin cans and a black rubber ring. Floating in the water near the left troll is a newspaper with the headline "President Opens Soylent Green Facility". In the background, the stream leads out toward a landscape with pine trees under a partly cloudy sky.
  • The Caption: At the very bottom of the image, text reads, "He's taken all the fun out of it", playing on the concept of traditional mythical trolls versus modern internet trolls.

Can ask it specifics: What is the colour of the hair of the troll on the left?
The hair of the troll on the left is orange.​
​
And you ask it to make changes you describe: change the trolls to be scruffy human trolls

1781371687399.png


​
 
Last edited:
You must log in or register to reply here.

Share this page

ISF - Join now!

Every member here is approved by hand. No bots, no spam, just people who care about evidence and honest debate.

Membership is free!

Create your free account

Back
Top Bottom