AI is so terrible that a drunk intern could outperform it on a lunch break, and so brilliant that only constant witch hunts (usually among already marginalized artists) can protect us from giving it awards.

[fatsam@mstdn.social]

AI is so terrible that a drunk intern could outperform it on a lunch break, and so brilliant that only constant witch hunts (usually among already marginalized artists) can protect us from giving it awards.

"Was this award-winning book written by AI? Scandal deepens skepticism in literary world"

https://www.msn.com/en-ca/news/other/was-this-award-winning-book-written-by-ai-scandal-deepens-skepticism-in-literary-world/ar-AA2cUIuc

[vivtek@indieweb.social]

@fatsam You saw the article a while back about Shelley's Frankenstein being detected as 100% AI?
https://www.linkedin.com/pulse/ai-detector-just-flagged-frankenstein-ai-generated-worries-garc%C3%ADa-xoc9c

[fatsam@mstdn.social]

@vivtek no, but unsurprised.

[rrb@infosec.exchange]

@vivtek @fatsam proof of time travel

[paulm@infosec.exchange]

@vivtek @fatsam ironic that the post itself was also entirely AI written

[jonpsp@mstdn.social]

@vivtek @fatsam

ironically, that article felt like it was written by a chatbot, unless that's the house style of people who post articles to LinkedIn.

[martinvermeer@fediscience.org]

@vivtek @fatsam

There is no reliable way of detecting AI-generated text or imagery - because it too is made out of ripped-off, human made raw materials. There are some heuristics that may help, but they won't last. It's an arms race.

The only way to detect reliably is using not the text or image itself, but where it comes from. Reputation is still a thing. Like in journalism, 'lateral reading'. Media literacy. We must all become bellingcats.

[ndevenish@mas.to]

@vivtek @fatsam presumably Mary Shelley isn’t submitting that text as her own work in 2026. If anyone else is submitting that work, then it was indeed AI written.

Replace the word AI with “Plagiarism”. “OMG I pasted original Frankenstein into a plagiarism detector and it said it was plagiarised” presumably wouldn’t confuse so much?

[xro@chaos.social]

@ndevenish
Indeed. A 100% result may not indicate that the text is classically written, but that it detects that every word is followed exactly by another word in an exactlyknown way...

Which in an ideal world should trigger a web search and find the source and display "copied from xxx" instead of "100% AI"

These detectors are neither accurate nor foolproof nor a solution, but the article seems to intentionally ignore technical fine points.

@vivtek @fatsam

[ndevenish@mas.to]

@xro @vivtek @fatsam I mean, AI writing _is_ Plagiarism. I’d entirely expect/assume some large subset of non-scam AI detection was just Plagiarism detection.

[evilstevie@mastod1.ddns.net]

@vivtek @fatsam gotta love when the training data gets flagged as output...

[wildhair@franken.social]

@vivtek @fatsam

Was sagt das Tool zur Bibel?

[rhodium103@mastodon.social]

@jonpsp @vivtek @fatsam

There are people on LinkedIn who I know for a fact are real people, who still write like that when they're on LinkedIn. If this isn't being academically studied, it should be.

[vivtek@indieweb.social]

@wildhair @fatsam Da die Bibel auch zum Training verwendet war, ich schätze ihr "AI-Inhalt" wäre recht hoch.