MysterG
Member
- Messages
- 1,293
- Location
- Wiltshire, UK
It sounds a bit like Schrodingers Table, both in and out of the van ....
- Thread starter indy4x
- Start date
It sounds a bit like Schrodingers Table, both in and out of the van ....
It probably is, and its also the case that putting in a robots.txt to not index the site if one is present is being ignored by AI companies who flout the agreed conventions on this sort of thing, so you can't really stop it.
Also, how many times have you seen a thread, where someone comes in and corrects something wrong after 5-6 posts agreeing with it. When someone asks that AI, its going to weight the response with the majority of responses favourable in its database, and when it makes up its nonsense from the probability tables following its calculated decision tree, it will spew out the wrong answers.
It has no intelligence, everything is just numerics in a btree database and it plots a course of most probability through the tree. And that's not even considering the training data being poisoned or skewed.
I did a lot of work researching AI when the hype was building, and we found some stuff to attack it and get it to leak credentials and talk about banned subjects in inappropriate ways, and also that it was completely unreliable, churning out nonsense, sometimes that nonsense would look plausible to all but someone who knew that subject extremely well & we concluded that it would result in a generation of coders blindly remaking every error posted on stack overflow because they never invested effort to learn in depth, and from that aspect it was dangerous even.
But suddenly we got told to stop all research and put a lid on all findings and to leave AI to the ai team only even in our personal media postings, because the CEO had decided it was the new direction for the company to jump on the hype train. And that's where we are, the hype train is still rolling, people are still making huge sums from people investing in what they're told is the next big thing etc.
Also, how many times have you seen a thread, where someone comes in and corrects something wrong after 5-6 posts agreeing with it. When someone asks that AI, its going to weight the response with the majority of responses favourable in its database, and when it makes up its nonsense from the probability tables following its calculated decision tree, it will spew out the wrong answers.
It has no intelligence, everything is just numerics in a btree database and it plots a course of most probability through the tree. And that's not even considering the training data being poisoned or skewed.
I did a lot of work researching AI when the hype was building, and we found some stuff to attack it and get it to leak credentials and talk about banned subjects in inappropriate ways, and also that it was completely unreliable, churning out nonsense, sometimes that nonsense would look plausible to all but someone who knew that subject extremely well & we concluded that it would result in a generation of coders blindly remaking every error posted on stack overflow because they never invested effort to learn in depth, and from that aspect it was dangerous even.
But suddenly we got told to stop all research and put a lid on all findings and to leave AI to the ai team only even in our personal media postings, because the CEO had decided it was the new direction for the company to jump on the hype train. And that's where we are, the hype train is still rolling, people are still making huge sums from people investing in what they're told is the next big thing etc.
OJCar
Member
- Messages
- 2,331
- Location
- Lothian & Westmorland
Is that how he got the table out ...
Got the cats to flick it off the edge.
ajlelectronics
Member
- Messages
- 10,794
- Location
- Gloucester, England
Just like applying the word "Digital" to everything. Funny that they forgot to market "digital light switches" which would be an accurate description.Somewhat ironically there is now a story in which Mumsnet are looking at using AI to help moderate their site
Mumsnet: Parenting site targeted with child sexual abuse images
Company founder Justine Roberts tells the BBC the "horrific incident" has been reported to the police.www.bbc.co.uk
I do wish people would stop applying the AI label to almost everything using an algorithm these days
Until half the population of the world understands the difference between machine learning and AI it's not really possible to have much of a conversation about it.
redlig
Member
- Messages
- 4,158
- Location
- wiltshire uk
Rodney !
Bullet2012
Member
- Messages
- 5,365
- Location
- on average about 200 miles from the ISS.
isn't that called the off switch ?
winchman
Member
- Messages
- 6,197
- Location
- Merseyside
It was touch and go but as we had a buyer for the van and he didn't want the table lots of Bodgery later a Mini digger with no bucket and faulty starter switch and an assortment of slings from the car boot sale it came out easily
The latter.
I think the best example is a company that made video processing chips (GPUs) renaming them as AI processing chips. Did wonders for the stock price.
Jack
What I find most interesting about AI are the numbers relating to the hardware that it runs on.
The first computers I ever worked on had actual transistors in them, and then I mostly dealt with small-scale integrated circuits, or SSI, such as Texas Instruments 74 Series TTL chips, containing a dozen gates and maybe 100 components.
So now when I read the statistics about the biggest AI machines ( from Cerebras Systems Inc.) I am totally astounded. They use wafer-scale integration with 8-inch square "chips", and the numbers speak of 4 trillion transistors, 1 million cores, 20 billion Megabytes-per-second memory bandwidth, and 500 kW power consumption....
The first computers I ever worked on had actual transistors in them, and then I mostly dealt with small-scale integrated circuits, or SSI, such as Texas Instruments 74 Series TTL chips, containing a dozen gates and maybe 100 components.
So now when I read the statistics about the biggest AI machines ( from Cerebras Systems Inc.) I am totally astounded. They use wafer-scale integration with 8-inch square "chips", and the numbers speak of 4 trillion transistors, 1 million cores, 20 billion Megabytes-per-second memory bandwidth, and 500 kW power consumption....
Last edited:
Training the chat gpt4 took some ware around 15million kwh
no wonder microsoft is buying three mile island
Stu
