AI

maria932

what is the best AI detector please.

flaneurs_lobster

This post sounds like it was written by a (cheaper, free) generative AI agent.

JSmithy45AD

They make it sound like they wish to plagiarise some AI generated text and then 'humanise' it so it doesn't look like AI text? Can't be right surely?

flaneurs_lobster

https://forums.moneysavingexpert.com/discussion/comment/81908025#Comment_81908025

Needs a specialist AI that makes AI output less AI-like.

elsien

Or they could just write the essay themselves as I’m sure whoever it’s for intended.
If anyone is wondering, the OP has amended the original question to remove a reference to essay writing.

MattMattMattUK

https://forums.moneysavingexpert.com/discussion/comment/81908077#Comment_81908077

One can do it on the paid versions of Gemini and ChatGPT by giving it style guide to follow, local models can have style guides embedded in them as well. That can include using specific words, phrases or stylistic elements, through to telling it it to introduce random grammatical errors or a specific list of errors to introduce when certain phrases are used. If one has enough data it will even create the style guide based on the previous body of work.

Emmia

https://forums.moneysavingexpert.com/discussion/6657846/ai

If you're using it for academic work then normally you must declare it, with a list of the prompts used and the outputs.

Anything else is probably going to be seen as cheating.

AI often hallucinates and makes up academic references. I'd also be cautious of "academic papers" written in the last few years that you identify via Google Scholar or similar tools, especially if they're from more unusual journals - lots of slop.

*Realistically given that many search engines are frequently pervaded by AI it is difficult to avoid it completely - but I'd be so cautious.

twopenny

I thought tis was about the sites you use to put stuff in to find out if it's AI generated.

I was on a forum where someone used one to detect AI generated reviews.

But until the OP comes back to explain more clearly it's hard to say

MattMattMattUK

https://forums.moneysavingexpert.com/discussion/comment/81908493#Comment_81908493

It is the thing that current AI/LLMs are very good at, pattern matching.

I have used it a few time on here and Gemini can detect sock puppets with greater than 99% probability, there are a few regular antagonistic posters who have sock puppets that I thought were fairly obvious and Gemini confirmed with greater than 99% probability that link without me telling it who I thought the ID was a sock puppet of.

AI without a style guide tends to leave fingerprints in the text it writes which can be detected by another LLM very easily, although the degree which these are immediately obvious to a human vary.

elsien

I have read, although I can’t confirm the accuracy of the statement, that the software that educational establishments use to pick up AI content is more likely to flag content from neuro-diverse people as AI even where it may not be.

twopenny

Neurodiverse is a wide spectrum. Apparently I'm on being dislexic, cross lateral and ambidexterous (useful for house painting, pruning and such).

Having worked with children who have such there's a lot about but applying to vocabulary is interesting.

I can usually spot AI, probably from too much time spent on various forums 😄