AI is taking over - ChatGPT

[pquinn]

Langweilig said:

I asked Bing's AI sorcery thingy a question and this was the response -

"I’m sorry but I couldn’t find any information on the cast of a movie adaptation of “Out of the Silent Planet”. However, I found a list of actors that could play the main character on IMDb. Kenneth Branagh is suggested for the role of Dr. Elwin Ransom. Hugh Grant is suggested for the role of Dick Devine. Jonathan Pryce is suggested for the role of Professor Edward Rolles Weston. Morgan Freeman is suggested for the voice and CGI stop motion actor for Hyoi1. Patrick Stewart is suggested for the role of Malacandra, the Angel of Mars"

You can get the same list as an answer via a normal search for the same terms, it just flowered it up a bit.

[Blackpuddin]

Not ChatGPT related, although sort of, this is a shocking story affecting young girls in Spain.
https://www.bbc.co.uk/news/world-europe-66877718
If the prospect of this happening on a wider scale stops young people sharing everything about themselves online then some good will come of it.

[Blib]

I use it to write promotional blurb and advertising promos for the Krupp range of teaspoons.

[anxious_ant]

Frimley111R said:

Derek Smith said:

AI will likely produce jobs as well, don't forget.

Such as?

Prompt Engineers

[Frimley111R]

Blackpuddin said:

Not ChatGPT related, although sort of, this is a shocking story affecting young girls in Spain.
https://www.bbc.co.uk/news/world-europe-66877718
If the prospect of this happening on a wider scale stops young people sharing everything about themselves online then some good will come of it.

Is that some local AI application of a general online one that the boys accessed? I imagine that pretty soon you can do this for anyone, assuming you can't already,

[pquinn]

Frimley111R said:

Blackpuddin said:

Not ChatGPT related, although sort of, this is a shocking story affecting young girls in Spain.
https://www.bbc.co.uk/news/world-europe-66877718
If the prospect of this happening on a wider scale stops young people sharing everything about themselves online then some good will come of it.

Is that some local AI application of a general online one that the boys accessed? I imagine that pretty soon you can do this for anyone, assuming you can't already,

It'll be a general one.

All it really managed is something people have done with image editing for at least 30 years, just with less effort & better results than an amateur could usually manage.

As ever AI offering nothing truly new, just a different way to get results that puts it in reach of a new group of idiots.

No doubt it'll go on the list as another !!new!! threat demanding regulation of AI.

[TonyRPH]

I asked ChatGPT:

Show me some words which contain the letter 'h' twice or more.

1. ChatGPT

Certainly! Here are some words that contain the letter 'h' twice or more:

Schoolhouse
Photochromic
Chitchat
Upholstery*
Whichever
Hitchhiker
Archaeology*
Thermohaline
Choir*
Dishcloth*
Nephew*
Exhale*
Elephant*
Champagne*
Ghostwriter*
Technophile
Hyphen*
Mushroom*
Pathfinder*
Uphill*

These words have the letter 'h' repeated at least twice within them.

• Not quite...

[SpudLink]

TonyRPH said:

I asked ChatGPT:

Show me some words which contain the letter 'h' twice or more.

1. ChatGPT

Certainly! Here are some words that contain the letter 'h' twice or more:

Schoolhouse
Photochromic
Chitchat
Upholstery*
Whichever
Hitchhiker
Archaeology*
Thermohaline
Choir*
Dishcloth*
Nephew*
Exhale*
Elephant*
Champagne*
Ghostwriter*
Technophile
Hyphen*
Mushroom*
Pathfinder*
Uphill*

These words have the letter 'h' repeated at least twice within them.

• Not quite...

It often makes mistakes. But it is worth telling it what it's done wrong and asking it to try again.

I tried your request and got the same result. I then typed the following...
"This is not correct. some of those words such as Champagne only have the letter 'H' once. please revise the list so that it only has words with 'H' repeated more than once."

It then corrected its mistake.

[Mr Whippy]

• convincing ** answers, not correct or actual answers.

And that all assumes the NN has been filled with factual information.
If they’ve scraped PHs forums and Reddit etc, it’s likely half the content they’re scraping is convincingly wrong sounding.

Even books of fiction as reference, laced with facts or truths but telling a fantasy story, ie, Jaws, or Jurassic Park.


The NN has no idea. It just churns out convincing strings of words in relation to what was fed in.

[Mr Whippy]

pquinn said:

Frimley111R said:

Blackpuddin said:

Not ChatGPT related, although sort of, this is a shocking story affecting young girls in Spain.
https://www.bbc.co.uk/news/world-europe-66877718
If the prospect of this happening on a wider scale stops young people sharing everything about themselves online then some good will come of it.

Is that some local AI application of a general online one that the boys accessed? I imagine that pretty soon you can do this for anyone, assuming you can't already,

It'll be a general one.

All it really managed is something people have done with image editing for at least 30 years, just with less effort & better results than an amateur could usually manage.

As ever AI offering nothing truly new, just a different way to get results that puts it in reach of a new group of idiots.

No doubt it'll go on the list as another !!new!! threat demanding regulation of AI.

Yeah it’s like the deepfake software.

Feed in generic naked pics from wherever (even randomly generated ones), train the NN with the copious selfies these teenage girls have shared to the world, and voila, out comes a pic of said face on said naked body.



Indeed hopefully people will start to learn their biometric data shouldn’t be shared with gay abandon. No one needs to share anything online.
At most I’ve shared a couple of low res pics and that’s it. The rest is mostly just forum ranting.

It makes me question the validity of biometric ID integrity for banking etc. ie, voice copying, real-time face copying etc.

Biometric is fine, if you’re there in person. But to use it online, hmmmm.


Also funny how MS and Google are releasing tools to spot fake imagery/text.
If they can train an NN to do that, they can train an NN to create material the former NN can’t spot.

It’s bloody ridiculous. Yet they’ll happily sell these worthless tools to educators and so on suggesting they can check student work is original etc.

[pquinn]

CharGPT in a nutshell - convincingly present an answer which may be only partially right or not at all.

It's a chatbot, it's built to reply plausibly for a context, not actually understand questions or provide accurate answers. It just does patterns.

[rdjohn]

SpudLink said:

TonyRPH said:

I asked ChatGPT:

Show me some words which contain the letter 'h' twice or more.

1. ChatGPT

Certainly! Here are some words that contain the letter 'h' twice or more:

Schoolhouse
Photochromic
Chitchat
Upholstery*
Whichever
Hitchhiker
Archaeology*
Thermohaline
Choir*
Dishcloth*
Nephew*
Exhale*
Elephant*
Champagne*
Ghostwriter*
Technophile
Hyphen*
Mushroom*
Pathfinder*
Uphill*

These words have the letter 'h' repeated at least twice within them.

• Not quite...

It often makes mistakes. But it is worth telling it what it's done wrong and asking it to try again.

I tried your request and got the same result. I then typed the following...
"This is not correct. some of those words such as Champagne only have the letter 'H' once. please revise the list so that it only has words with 'H' repeated more than once."

It then corrected its mistake.

I am genuinely shocked, having been impressed with the results for subjects I have questioned, this has suddenly helped me realise why AI is thought to be so dangerous, by so many.

That there is no simple check routine for the results is quite astonishing. Equally that a human can then point out its errors and only then can it correctly answer a seemingly simple question makes it somewhat pointless.

Extend that to today’s hot story

https://www.theguardian.com/technology/2023/sep/25...

and you might get the impression that it will only be capable of designing stink-bombs..

[pquinn]

If you look at who is moaning loudest about AI needing to be regulated and their own use of it, you get the impression they're selling a fantasy of a 'danger' while trying to lock small players out of the AI related markets they have existing interests in.

You can do some clever things with various flavours of AI, but the scary hype mostly doesn't tie up to it or even necessarily offer something that wasnt already possible.

[Frimley111R]

Mr Whippy said:

It makes me question the validity of biometric ID integrity for banking etc. ie, voice copying, real-time face copying etc.

A friend working in cyber security was telling me that just replicating a voice's sound is not enough, their software recognises pronunciations, gaps between words used, common words, breaths, etc.

However, it does seem like a continual battle against AI now

[Hoofy]

pquinn said:

CharGPT in a nutshell - convincingly present an answer which may be only partially right or not at all.

It's a chatbot, it's built to reply plausibly for a context, not actually understand questions or provide accurate answers. It just does patterns.

Sounds a bit like a Conspiracy Theorist.

[Baron Greenback]

AI boss Sam Altman ousted by OpenAI board, some thing fishy going on or has chatgpt5 taken over

[Mr Whippy]

Bit weird, but I suppose vested interests now need yes men.

Clearly he wouldn’t say yes to the corporate monolith’s needs.

So it’s become a lumbering dinosaur that needs a revenue stream, because share holders, because ‘the board’ etc…


Peak AI for this cycle.

[Baron Greenback]

Another lost his place on the board but kept his job and he quit there and then and just heard 4 other top people also quit. Rumours have it is all about nonprofit aspect of open AI and he was stearing it for profit without telling aspect of his plan to the board.

[Mr Whippy]

Sounds a bit weird to me.

Surely inverse to what’s likely to be the case given the board reflects shareholders, and they’d prefer to see profits if it’s viable?

The whole think to me is nonsense any way calling it openAI, in the world of computing it suggests open source but it’s completely closed source.


While Meta/FB are releasing their model for free inc source code…


All of which says to me they’re all looking for a way to make it make money, but no one knows how.


I’ve been using a number of the image generators now but still can’t see the value in paying for them to mostly ignore what I’m asking for and draw their own bloody pictures

[Glade]

I subscribe to Chat GPT and use it a lot for refining or fixing bits of code or quickly learning how to do something in an application I am unfamiliar with..

Now it creates images I use them for bits and pieces. It's like having a custom illustrator. e.g. for our SharePoint site, I asked "Give me an image of a man looking at an audit schedule, give me an image of an inspection report on an industrial workbench" etc to create professional-looking images for the hero section of our department website.

But in my specialist subject if I ask for it to write me something technical for a presentation it often misses the point and the output isn't good enough. It's not an experienced expert in a field. I understand that these language models are just predicting what words should come next. It's often quite simplistic usually correct, but not innovative enough if you want to be more than a basic bh on a specific topic. It also pads answers out with superfluous stuff.

I tried Microsoft Copilot for Power Automate last week, and it kind of works, but doesn't just spit out full flows that work. It does significantly reduce time googling for "How to" tutorials.

I guess I sort of use it as a very knowledgeable colleague and we have conversations where we discuss the task and refine the output to get to a good answer quicker than if I was working alone. This iteration won't be taking my job though, just boosting productivity for those that figure out how to use it.