Skip to main content


The second time in a week that a South African government approved policy document has been caught having AI-fabricataed sources. :/

techcentral.co.za/schreiber-su…

#SouthAfrica #AI

in reply to yelling jackal

@len Only when someone decides to actually check the citations. And to be honest, how many people do that? Consider the average Wikipedia article: the vast majority of readers are just happy to see a little "[1]" next to an assertion. Very few would actually click on that [1] and go read the source article.
in reply to Graham Downs

@len in this case, Schreiber has engaged an independent law firm to review every single document since the chatgpt launch date. I expect that should catch quite a few more.
in reply to Allen but one of the good ones

@uastronomer @len Yeah, I'm also very interested to see what comes of that. As the article says, just how entrenched is this practice of using ChatGPT to generate documents and then releasing them without proper verification?

I suspect it's pretty deeply entrenched.

in reply to Graham Downs

@uastronomer yup, it's very unlikely that people read the sources, which is why I'm seeing it as a silver lining that it's taken months instead of years to uncover this (for the given documents, it's not unlikely there's some that /have/ gone undiscovered for years). Good to hear they're doing that review, too, because I also have that same suspicion.
in reply to yelling jackal

@len @uastronomer I was at a DHA office yest. Should I be worried that they are still using Windows 7 Enterprise edition 🫣
in reply to Chris

@flipsideza @len hehe, and to think my last visit i was worried about all the open usb ports 30cm from my face, next to the three different USB fingerprint scanners which the official had to use to sign in but none of the worked so she had to cycle between them over and over before she finally got in