chat GPT and other AI tools

[Jdsk]

I started the AL CHAT to show the limitation of the process and the disjointed reasoning.
...

It's an interesting subject. You would probably have had a better reception if you'd said that was why you're posting. And kept it all in one thread.

Jonathan

[axel_knutt]

Four threads were started using AI generated posts, all of which have been removed. I think some rules will be needed regarding use of AI generated material in posts, but for the time being please do not use AI to generate posts, especially not for a new thread.

I think that AI 'opinions' are of legitimate interest, and all the more valid a topic for discussion if they're unreliable, as long as they're not passed off as human, and banning them altogether is likley to increase the risk of just that.

[re_cycler]

But is quoting from AI any different to quoting from the media, for both you need to decide if it's credible.

[sizbut]

But is quoting from AI any different to quoting from the media, for both you need to decide if it's credible.

Very. AI like ChatGPT doesn't come up with stuff spontaneously, they are seeded with parameters by the requester. Someone can ask it to generate a 1000 word essay on how you murdered your whole family and it will do precisely that (you didn't I hope). If you're going to quote ChatGPT, you're going to have to quote the parameters of what it was asked to do and by who.

[re_cycler]

But is quoting from AI any different to quoting from the media, for both you need to decide if it's credible.

Very. AI like ChatGPT doesn't come up with stuff spontaneously, they are seeded with parameters by the requester. Someone can ask it to generate a 1000 word essay on how you murdered your whole family and it will do precisely that (you didn't I hope). If you're going to quote ChatGPT, you're going to have to quote the parameters of what it was asked to do and by who.

I would agree that quoting the whole conversation allows the reader to make a better judgement, in the same way as giving the research paper or journal that a quote comes from. But how do we know that when a post has information copied from the media, that the original article wasn't generated by the reporter researching using ChatGPT.

[Biospace]

I started the AL CHAT to show the limitation of the process and the disjointed reasoning.

I find it laughable the removal of these posts.

That is a very understandable reaction; it sounds as if moderators are having ongoing problems with people trying to create or use fake accounts using AI to fool them, so they're fearful of any use of it, clearly even if it is someone well known to them like you.

A little transparency and clear indications of the motivations and intentions of those who moderate may help explain their actions.

[gaz]

The general explanation is that the moderators etc., have always had a section of the forum which is only visible to them, which is why it's normal to be unaware of its existence.

The secret forum board can be found here.

[Paulatic]

I shall be taking the hump today and might return tomorrow
During the reorganisation of posts it has not escaped my notice that a post of mine which recognised why Philip had made a post including AI has been deleted. Also I no longer see a post by JDSK using the word moderator in a signature position. This doesn’t feel like fair editing to me.
I get the impression there has been a lot of back seat moderation of late. I might be right I might be wrong but that’s my impression.

[Vorpal]

I shall be taking the hump today and might return tomorrow
During the reorganisation of posts it has not escaped my notice that a post of mine which recognised why Philip had made a post including AI has been deleted. Also I no longer see a post by JDSK using the word moderator in a signature position. This doesn’t feel like fair editing to me.
I get the impression there has been a lot of back seat moderation of late. I might be right I might be wrong but that’s my impression.

I removed the thread because it didn't really provide any new information, and thought that this thread was the place to discuss chat GPT & other AI tools. I sent Phillip Benstead a pm, explaining that.

[mattheus]

I started the AL CHAT to show the limitation of the process and the disjointed reasoning.

It's a very important and/or interesting topic, worth having a thread here.

But it wasn't helpful to start multiple threads on the same topic, and not giving any explanation.

[Vorpal]

There are various issues with these tools. Besides those mentioned by slowster, there remain ethical issue with the use of text & language, as these tools have no way to distinguish between copyright material and material in the public domain. Therefore, there is a risk that someone posting text from chat GPT or similar is unknowingly plagiarising. The risk of this increases with specialist subjects (e.g. risk compensation)

The output is correlation based; it's basically a complex version of autocorrect, so it isn't verified, and because the reader / user doesn't know the source, it cannot be verified. In the worst case, such tools will lend additional weight (increased statistical correlation of words) to biased or completely incorrect information, just because it is 'popular'. That is something the forum should not be contributing to.

Lastly, there are security issues with it. Most do not yet impact the forum, but the potential is there. Demonstrations have shown that AIs can be manipulated to produce malicious content.

Until we are able to take measures to protect the forum from these issues, we cannot allow it.

[mjr]

these tools have no way to distinguish between copywritten material and material in the public domain.

Copyright not copywritten or copyrighted, please. Copywriters are people who write copy. Copywritten material is what they have written. Almost everywhere now, copyright is an innate (imaginary) property of most creative work: an article has copyright, it is not copyrighted. Eventually, its copyright expires and it becomes public domain. And not "in the public domain" which simply means it is known to the public. If you think all this is a flaming mess of language, you're correct.

Therefore, there is a risk that someone posting text from chat GPT or similar is unknowingly plagiarising.

I think almost everyone knows now that GPT is simply plagiarising other people's hard work. Sometimes with permission (which is why big social media demand a broad licence to your writing) and we're fairly sure that quite often without permission simply by copying websites into its training data.

[...] In the worst case, such tools will lend additional weight (increased statistical correlation of words) to biased or completely incorrect information, just because it is 'popular'. That is something the forum should not be contributing to.

Agreed.

Lastly, there are security issues with it. Most do not yet impact the forum, but the potential is there. Demonstrations have shown that AIs can be manipulated to produce malicious content.

I think the helmet post contained enough human-harming malicious content to demonstrate that!

[re_cycler]

But does Chatgpt plagiarise, from what I understand the created text is just the most likely answer constructed from it's database. Not a copy of existing text.
Predictive text on steroids.

Edit - I accept an idea maybe be plagiarised, but not the body of text.

[Jdsk]

But does Chatgpt plagiarise, from what I understand the created text is just the most likely answer constructed from it's database. Not a copy of existing text.
Predictive text on steroids.

Edit - I accept an idea maybe be plagiarised, but not the body of text.

It's output isn't a character for character or word for word match.

Using it without appropriate declaration is classified as plagiarism by all of the universities whose policy I've seen. But each sector is going to have work this out for itself.

Jonathan

[re_cycler]

It's output isn't a character for character or word for word match.

Using it without appropriate declaration is classified as plagiarism by all of the universities whose policy I've seen. But each sector is going to have work this out for itself.

Jonathan
[/quote]

Thanks for the correction, I shall add it to my personal LLM

I did notice that where Bing search now gives AI generated replies it includes citations and the web links to view the original source. I would have inckuded an example but thought maybe that wouldn't be appreciated