Isabel Picornell: “We identified hundreds of articles containing a prompt that requested authentic articles be rewritten to adopt a “conservative stance against the Macron administration” in a likely effort to sway voters’ opinions ahead of the European Parliament and the snap elections”

“We identified hundreds of articles containing a prompt that requested authentic articles be rewritten to adopt a “conservative stance against the Macron administration” in a likely effort to sway voters’ opinions ahead of the European Parliament and the snap elections”

Our latest investigation uncovered a disinformation campaign targeting France's snap elections, using AI-generated content to manipulate voter opinions. The operation, CopyCop, used large language models (LLMs) to create fake news outlets with a conservative bias against President Macron. Read more:

Operation targets French snap elections using AI-generated contentdfrlab.org Operation creates inauthentic news websites and uses AI to rewrite articles to adopt a more 'conservative stance'

I wonder if rewriting articles constitutes plagiarism? It's not quite it, but yet ..

It feels like the very idea of plagiarism needs redefining to cover what LLMs are doing: hoovering-up huge quantities of human output and re-outputting it, usually without crediting authors or paying licence fees. There are some brilliant uses of AI (eg in medicine) but we are mostly seeing abuses!

I agree with you. At the IAFLL conference at Aston Uni two weeks ago, Rui Sousa Silva, University of Porto, spoke how plagiarism identification software was useless with AI for that reason.

This doesn't bother me personally, but content creators clearly feel threatened and undermined - and the "AI poison" movement is gaining traction... I must check where AI is up to with music ! Can it create a Bach fugue that fools other musicians ?

That's not what LLMs are doing, mostly. Indeed, an LLM can be built, even using "hoovered data" without the model itself ever seeing that data. And indeed more and more synthetic data is being used. "Plagiarism" does need redefining if you want to call creating derivative works "plagiarism."

Proper use of an LLM results in novel works, not copies of existing works.

Yes, but isn't this "novel" just in the sense of "mass recombination" ? You might argue this is what human authors do (decades of reading, then some writing) but at least they paid for the books they read! And their capacity to flood the market is limited. The LLM direction of travel worries me.

99% of our creative outputs is "mass recombination." That last 1% comes from our personal experiences, which can still be injected into the creation process, even with LLMs. You seem to misunderstand how these tools can be used. Example. bsky.app/profile/dgol...

I really should just do a video series, but I hate doing videos, it's not my medium, etc. Maybe one day I'll sit down and do one, to really show the entire process of working on a piece, but for now, here's a look into the process I use to craft a piece using LLM technologies.

The Making of an Articlemedium.com A partial journal on how I craft a piece from my mess of thoughts.

Especially interesting in light of the utterly OTT, unmediated praise for GAI emanating from Tony Blair this morning on the Today programme (from 8:10 AM).

Interesting. I'll have to find the recording.

Post