
#032 - Novels and proofs: who gets the last word? | SynthAIcast
Notes de l’épisode
One single sentence written by AI, slipped into a prize-winning novel. Who would have noticed? And who would have had the right to judge it?
This autumn, France's most famous literary prize asked a piece of software for its opinion on the novels in the running. The day after the award, one revelation reshuffled the deck.
Three mathematicians explain, on a long podcast, how they check a machine that knows more than they do, and why the right answer isn't enough for them.
A lab puts in writing who should keep the last word over its models, and admits in the same text what is still missing.
A forgery case from 1945 shows the only proof left when the expert can no longer recognize the author.
In the end, the question is no longer who wrote a text. It's being able to show it, with proof.
Three signals. One thread. One question…
🎙️ Every week, SynthAIcast listens to the world's best AI podcasts so you don't have to. Synthesised in ten minutes, verified by a human, in English and French
👉 Also available in French: SynthIAcast
🔗 THE SOURCES, BY SIGNAL
Signal 1 - one sentence and a detector
• Euronews - the Goncourt and the AI row, 03/11
• La Libre - the winner and his sentence, 04/11
━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Signal 2 - checking a machine with a machine
• Lex Fridman #503 - three mathematicians, 26/10
━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Signal 3 - the last-word rule
• OpenAI - Model Spec of 29/10
━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Perspective
• Reuters - the Nasdaq's drop, 05/11
━━━━━━━━━━━━━━━━━━━━━━━━━━━━
🎧 GO FURTHER
• Monde Numérique (FR) - can you refuse AI?, 05/11
Also cited, without a link: RTL (detection software, 06/10) and the Vermeer forgery case (1937-1945).
Hosted on Ausha. See ausha.co/privacy-policy for more information.