Plagiarism Engine: Google’s Content-Swiping AI Could Break the Internet | Hacker News Reader

Plagiarism Engine: Google’s Content-Swiping AI Could Break the Internet

tomshardware.com

21 pointsT-A3 comments

"Instead of highlighting links to content from expert humans, the “Search Generative Experience” (SGE) uses an AI plagiarism engine that grabs facts and snippets of text from a variety of sites, cobbles them together (often word-for-word) and passes off the work as its creation."

Feels like it's maybe telegraphing the future of their "rich snippets"? So, they mash sources together, reword, and drop the attributions. It has the long term effect of removing visitors to the source content, so eventually the incentive to publish that goes way down. And the double-whammy that the presented widget itself pushes real organic results down the page. It's one factor that could sort of freeze the internet in time, the other being AI like this accidentally training on the output of other AI.

> And the double-whammy that the presented widget itself pushes real organic results down the page.

Even further down the page than they already are! It's easy to forget from the uBlock bubble that the intended experience of Google is to search for a company and the first three "results" to be for three of their competitors who paid for placement. The endgame is leading with an AI generated section with sponsored fine-tuning to bias it towards suggesting particular products, followed by sponsored results, followed by real results buried between GPT-generated SEO farms.

They actually changed it now.... but some time ago I wanted to have the approximate numeric value of the frequency of green light (unusual metric, I admit, nm => Hz). Googles suggestion told me it would be 55 - 1014Hz. The quick answer sourced some document but couldn't correctly parse the notation. The answer was 5.5 * 10^14 Hz...

Now it seems it get you some more correct answers, which are also presented in a different format. But I would guess that similar AI generated answers will be subject to similar mistakes and it gets worse if you deviate from English terms too.

Reply on news.ycombinator.com