[LINK] OpenAI just admitted it can't identify AI-generated text.

[Jan Whitaker]

On 29/07/2023 11:24 am, Roger Clarke wrote:
>
> Overlaid over that is the cluster of issues that underlie 
> misinformation, such as selective quotation and incompleteness / 
> acontextuality;  and bias and discrimination arising from dominance of 
> some value-sets over others. 


And one of those is when those word sets were written. If you think 
about what is accessible, much of it is from copyright free things like 
writings before 1950 (allowing for the 72 year limit in some places). It 
was a totally different value set back then. I wonder if you could put a 
parameter that limits the source data to information published since 
2000 to at least make it work in this millennium. And there are also 
cultural differences, whether the data source was fact checked or 
speculation, what information domain, too many variables/categories to 
contemplate. It's like those old phrase generators on magna steroids.

Language is so ambiguous, it's a wonder we can communicate anything with 
any clarity.

Jan