Meta AI - can we really trust AI???

[PaulSB]

I'm awake and bored. Recently I've noticed a coloured circle and some questions below Facebook posts. I decided to press a question that caught my eye:

"Who is Blackburn Rovers new manager?" I'm a season ticket holder. The answer given was John Eustace. In fact Eustace was appointed in February 2024 and left in January 2025.

My next spot was "Upcoming E Street Band dates." There are none. Meta AI lists the upcoming 2024 and 2025 dates. The tour finished on July 3rd.

I've never trusted anything that's AI generated. After spotting two more incorrect answers to questions I know the answers to I'm even more certain. I also checked other questions on local events I could verify. These were correct.

AI simply cannot be trusted to be accurate.

 

[Mad Doug Biker]

I think it's still 'learning'.
I've noticed some errors too and I couldn't see how to correct it (although in hindsight, I think it might have corrected itself had I said something about it in the comments... Which, could potentially be dangerous in certain circumstances if done by the 'wrong' person).

I was also vaguely bemused to see what it had to say about my Dad's dog (and the other dogs in the other posts), in this post (if you see the original post on the page anyway).
You can clearly see it trying to figure things out and 'learn' about the dogs, etc:

https://www.facebook.com/share/p/1BbEG45vTL/?mibextid=wwXIfr

 

[raleighnut]

Wasn't @Drago 's idea behind the 'true facts about Chris Rea' to throw a bunch of crap out there for AI to pick up on and propagate throughout the web......................I don't trust it as I don't trust 'driverless cars' and the crap I've read about 'driver aids' like lane assist etc.

Does anyone else here remember the collision avoidance clip where the guy stood in front of the car saying "it will stop before it hits me" only to be promptly run over.

 

[PaulSB]

I think it's still 'learning'.
I've noticed some errors too and I couldn't see how to correct it (although in hindsight, I think it might have corrected itself had I said something about it in the comments... Which, could potentially be dangerous in certain circumstances if done by the 'wrong' person).

I was also vaguely bemused to see what it had to say about my Dad's dog (and the other dogs in the other posts), in this post (if you see the original post on the page anyway).
You can clearly see it trying to figure things out and learn about the dogs, etc:

https://www.facebook.com/share/p/1BbEG45vTL/?mibextid=wwXIfr

It may well be "learning." As I understand it AI is far from intelligent but simply scrapes around the web for information and presents a summary of the content to the user. This isn't "intelligence," any of us could do this but more slowly.

I was simply bored, I wasn't looking to trick the AI. The issue for Meta is my view AI cannot be trusted is confirmed and will take a lot to change. No one at Meta will fact check Rovers. This is more Facebook misinformation generated by the company itself.

 

[PaulSB]

Wasn't @Drago 's idea behind the 'true facts about Chris Rea' to throw a bunch of crap out there for AI to pick up on and propagate throughout the web......................I don't trust it as I don't trust 'driverless cars' and the crap I've read about 'driver aids' like lane assist etc.

Does anyone else here remember the collision avoidance clip where the guy stood in front of the car saying "it will stop before it hits me" only to be promptly run over.

I don't know about Drago's thread as I don't follow it. The driverless car thing, in particular Tesla, is very scary. Published yesterday.

https://www.theguardian.com/technol...cars-keep-crashing?CMP=Share_AndroidApp_Other

 

[Mad Doug Biker]

It may well be "learning." As I understand it AI is far from intelligent but simply scrapes around the web for information and presents a summary of the content to the user. This isn't "intelligence," any of us could do this but more slowly.

I was simply bored, I wasn't looking to trick the AI. The issue for Meta is my view AI cannot be trusted is confirmed and will take a lot to change. No one at Meta will fact check Rovers. This is more Facebook misinformation generated by the company itself.

Yes, I never said it was 'intelligent', I meant that it is clearly still gathering, 'learning' about the world. I also pointed out that the info gleaned could potentially be false or dangerous if controlled by the wrong people.

That article about the cars below... 😱
I was unaware, but then, I am not into cars.

 

[gavroche]

I don't know about Drago's thread as I don't follow it. The driverless car thing, in particular Tesla, is very scary. Published yesterday.

https://www.theguardian.com/technol...cars-keep-crashing?CMP=Share_AndroidApp_Other

Our grand daughter took a driverless taxi when she was in San Francisco. She said it was weird but they got to their destination safely.

 

[Drago]

These large language models are not AI and should not be touted as such.

All they do is mimic intelligence, and not very well at that.

They're also proving to be a vast burden on the world's electricity resources. I will never use them if at all possible. They're just wrong on so many levels.

 

[Drago]

I don't know about Drago's thread as I don't follow it. The driverless car thing, in particular Tesla, is very scary. Published yesterday.

https://www.theguardian.com/technol...cars-keep-crashing?CMP=Share_AndroidApp_Other

Yes, I read that yesterday.

Its not so much the product itself that is scary, although occasionally that can be bad enough. Its Teslas unwillingness to be totally open and transparent, and the lack of will from the judiciary to hold them to account for their lack of cooperation, that is scary.

 

[markemark]

Yes, I never said it was 'intelligent', I meant that it is clearly still gathering, 'learning' about the world. I also pointed out that the info gleaned could potentially be false or dangerous if controlled by the wrong people.

That article about the cars below... 😱
I was unaware, but then, I am not into cars.

It is dreadful at facts and opinions. Its ‘learning’ is making itself more stupid. The dataset it’s using is inaccurate a d can be manipulated. It’s also partly ai generated so it’s eating itself.

However give it tasks and processes and it’s incredible. People aren’t using it to gather facts. Theyre using it to generate reports from data. To create code. To generate images. To summarise text. To translate. It’s not perfect but as a first draft or for ploughing through simple tasks it’s a huge time saver.

“Write around 200 words marketing info about our new widget. The spec can be found at…”. The write my responses for the following
People who want to buy it
People who query the price
People who query the quality
People who query the warranty

The data for these responses can be found at…

This is the sort of thing it can do in 2 mins that would take someone hours. It needs a read through but it can save a huge amount of time if you are selling hundreds of different widgets.

 

[fossyant]

Yes, I read that yesterday.

Its not so much the product itself that is scary, although occasionally that can be bad enough. Its Teslas unwillingness to be totally open and transparent, and the lack of will from the judiciary to hold them to account for their lack of cooperation, that is scary.

A pretty damning report that.

 

[cyberknight]

Ask Sarah Connor...

 

[Dogtrousers]

The important thing is the rate at which it's improving. Current capabilities were a dream only a few years ago. So yeah, it gets things wrong and confidently states complete fabrications as truth. But if it keeps improving at its present rate it won't be this bad for much longer.

And that will just exacerbate the various problems described above like energy use, concentration of power, transparency, privacy (and more)

What should be happening is that it should be open and auditable, able to fully substantiate every choice it makes with evidence and reasoning (as an intelligent person could) but I doubt any effort will be made in that direction.

 

[Pblakeney]

Yes, I never said it was 'intelligent', I meant that it is clearly still gathering, 'learning' about the world. I also pointed out that the info gleaned could potentially be false or dangerous if controlled by the wrong people.

This.
It is as simple as - Garbage in, garbage out. All evidence shows that it cannot be trusted.

 

[Electric_Andy]

I do wonder how AI knows when it's wrong i.e to learn, you need to know when you're wrong. Is there any sort of "report" or "you're wrong" button? We have been forced to use Google for everything at work, and are encouraged to use Gemini. I had a go at doing stuff and it does seem very good at writing code. I also asked it to write an email and it did, but it sounded less grammatically and contextually correct than what I could have written. But this is all playing out as I expected; the world now seems more generalised, diluted, dumbed down, with an emphasis on productivity rather than creativity