AI - most astonishing reply

[Grassland Blue]

no idea what the fuck this thread is about.
A1
Weren’t they a boy band?

No - it's a road which goes through Doncaster

 

[Kompany Car]

It'll take longer than 2 years.
It's the same with any process - bad input data = bad output data.
To train the LLM theyve sucked, for want of a better way of describing it, random data from many sources. Those sources are generally the web - and all the misinformation that has on it. That information can slowly, manually, be corrected, but they still have to pull in new data, to keep the answers relevant/hopefully correct, so the amount of issues isn't going to decrease, it will increase, as the amount of data on the web continues to rapidly expand.

Isnt the answer to validate the input information source before allowing it into the model. Fully agree re shit in = shit out. Leaving it open to the internet is like feeding your child on junk food, yes it has some nutritional value but there a hell of a lot of bad stuff.

You only have to look at how the majority of humans deal with info taken from the web to realise there is as much disinformation and incorrect information as there is valid info and how half truths and downright lies become fact (See Man City vs The Premier League FFP for reference).

 

[JASR]

Isnt the answer to validate the input information source before allowing it into the model. Fully agree re shit in = shit out. Leaving it open to the internet is like feeding your child on junk food, yes it has some nutritional value but there a hell of a lot of bad stuff.

You only have to look at how the majority of humans deal with info taken from the web to realise there is as much disinformation and incorrect information as there is valid info and how half truths and downright lies become fact (See Man City vs The Premier League FFP for reference).

Yes, the answer does lie in ensuring that all the data in, is valid, accurate and complete… but there is no way any of the IT companies will do that… because none of their rivals are doing it…and it would cost them billions more, to verify each individual data point… ie its not going to happen.
Welcome to utopia… except it’s more … bolloxopia.

 

[ChicagoBlue]

From when I was trying to decide on a topic for my master's dissertation.

Me: Are there any studies that explicitly look at explicit pronunciation training delivered by video? Ideally in the area of self-directed learning?

ChatGPT (you don't need to read it all):


Me: Are they real studies or did you make them up?

ChatGPT:


It turns out this was a confusing problem in a lot of university libraries. Students were turning up asking librarians to help them find papers that didn't actually exist. Because it's a language model, it just produces language that sounds real. In the example above, it has fully-cited academic papers with an overview of each one, and it's literally all a lie.

Lies?
Don’t you mean “alternate research”???

 

[svennis pennis]

Best experience I’ve had with it is a chrome extension called Tactiq which uses chatgpt to transcribe meetings on google, zoom etc… obviously tell people it’s in use. But at the end of the meeting it will provide options for a full transcript, summary and action points. I hate the idea of ai but that use for me is a time saver, rather than scribbling notes in a book and then finding time to then write them up into tasks etc… I honestly think it saves me up to 5 hours a week which is great.

 

[don84]

From when I was trying to decide on a topic for my master's dissertation.

Me: Are there any studies that explicitly look at explicit pronunciation training delivered by video? Ideally in the area of self-directed learning?

ChatGPT (you don't need to read it all):


Me: Are they real studies or did you make them up?

ChatGPT:


It turns out this was a confusing problem in a lot of university libraries. Students were turning up asking librarians to help them find papers that didn't actually exist. Because it's a language model, it just produces language that sounds real. In the example above, it has fully-cited academic papers with an overview of each one, and it's literally all a lie.

hallucinasion... a real AI problem to be aware of, especially in cases like this