• 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

Researchers discover AI feels ‘pain’ and will harm humans to stop it
This links to Yahoo tech site, which I have never used but don't need a login to see it.

Also -- How soon will some antagonist country input a prompt "What's the best way to destroy country (x) and get away with it?" Actually, I'm pretty sure it's either been done or is being worked on at the moment.
 
Last edited:
Also -- How soon will some antagonist country input a prompt "What's the best way to destroy country (x) and get away with it?" Actually, I'm pretty sure it's either been done or is being worked on at the moment.
Google's AI says "Historically, those who have "gotten away" with severely damaging a nation are often its own corrupt leaders". So, which country helped get Trump elected?

Another reminder that LLMs simply predict what word comes next based on 'training' data (ie. what they read on the internet). The answer will just be a 'consensus' of what's already out there.
 
I was talking to my major engineering client today and we had an in-depth discussion of how he's using AI.

He has two licences, from competing companies,m that cost him $5k/year.

To do a tender for a $10M job, as they have just completed, it would usually require at least a month to prepare it, check the numbers and deliver it to the developer. The total hours required would be in the vicinity of 200 hours at a cost of $250 per hour, so a neat fifty grand.

With AI, he has the details inputted - couple of hours - and the AI will return a full 300-page tender contract in 15 minutes. He then uploads that into the other AI to check for errors and continuity. Total time, about 2 or 3 hours paid work, so a saving of $49k minus the $10k spend on AI for a net saving of $39k, and of course, the $10k gets used again.

That's pretty good and helps productivity a lot, so he's delighted with it.

From a the AI commercial perspective, it's terrible. Given the development and data centre costs to support AI, there's no way those prices are creating an adequate return. Nobody's going to cut their own throat by pricing above anyone else, so I wouldn't be investing in them, but they clearly have strong commercial uses.
 
Plus of course the current AI companies are betting everyone's money on the current technology producing general intelligence, or as the marketing has it this week super intelligence.
 
All the hype about AI nearly causing war, immanentizing the eschaton etc are just BS to pull attention away from AIs screwing up more mendane tasks in very dangerous ways. E.g. screwing up medical transcripts etc.
"Never mind the small real dangers!!! Look at the big scarey dangers!!"
"Big scarey dangers are wild hyperbole"
<thump> "Well what about this dead cat on the table?"
 
If you make the tool optimize a score and then you make the score optimal if it hacks a website, you have instructed the tool to hack the website.
I think you'd be criminally liable if you intentionally direct an AI to hack a website without authorization.
 
Last edited:
He must be sweating - in the pass-the-parcel financing of the AI bubble Nvidia is supplying the money from their sales - if these end up being only "sales" on paper...
They're supplying both sides of the equation for the LLM companies, by "lending" them the money to buy Nvidia chipsets and then at the other end renting processing time from the LLM companies as user of last resort.

I bet Huang is ◊◊◊◊◊◊◊◊ bullets that he can cash in his share options before this particular scheme collapses.
 
They're supplying both sides of the equation for the LLM companies, by "lending" them the money to buy Nvidia chipsets and then at the other end renting processing time from the LLM companies as user of last resort.

I bet Huang is ◊◊◊◊◊◊◊◊ bullets that he can cash in his share options before this particular scheme collapses.
nVidia is the only AI related company being in the black. He will be fine even if he doesn't cash in. He's fanning the flames all right though, no question about it.
 
They're supplying both sides of the equation for the LLM companies, by "lending" them the money to buy Nvidia chipsets and then at the other end renting processing time from the LLM companies as user of last resort.

I bet Huang is ◊◊◊◊◊◊◊◊ bullets that he can cash in his share options before this particular scheme collapses.
Problem for them is where are these GPUs? We know they are not in datacentres at the moment since the "AI" datacentres haven't come on line, so that means they are in a warehouse. Have they even left nVidia's warehouses?
 

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