SoylentNews is people
SoylentNews
from the and-don't-miss-all-the-Tim-Horton's! dept.
As reported by The Verge
https://www.theverge.com/2023/8/17/23836287/microsoft-ai-recommends-ottawa-food-bank-tourist-destination
and
the Canadian Broadcasting Corporation
https://www.cbc.ca/news/canada/ottawa/artificial-intelligence-microsoft-travel-ottawa-food-bank-1.6940356
In 2020 Microsoft laid off dozens of journalists, in a move to rely on artificial intelligence. Those journalists were responsible for selecting content for Microsoft platforms, including MSN and the Edge browser. A recent tourism article now reminds us of that earlier business decision.
Published last week and titled "Headed to Ottawa? Here's what you shouldn't miss!" the article listed 15 must-see attractions for visitors to the Canadian capital. Microsoft has since removed the article that advised tourists to visit the "beautiful" Ottawa Food Bank on an empty stomach. That appears to be an out-of-context rewrite of a paragraph on the food bank's website. "Life is challenging enough," it says. "Imagine facing it on an empty stomach."
The remainder of the must-see list was rife with errors. It featured a photo of the Rideau River in an entry about the Rideau Canal, and a photo of the Rideau Canal in an entry about Parc Omega near Montebello, Quebec. It advised tourists to enjoy the pristine grass of "Parliament Hills."
The article carried the byline "Microsoft Travel." There is nothing on the page that identifies it as the product of AI. Microsoft did not immediately respond to a request for comment on how the article was generated. While now removed, it is still available via the Internet Archive.
https://web.archive.org/web/20230814223742/https://www.msn.com/en-gb/lifestyle/travel/headed-to-ottawa-here-s-what-you-shouldn-t-miss/ar-AA1faajY
(Score: 3, Interesting) by krishnoid on Monday August 21 2023, @04:11PM (1 child)
A one-sentence working memory, and "Imagine facing it on an empty stomach." ~ "Consider going there on an empty stomach." Maybe if it reread the whole source paragraph/article from start to finish and checked if the output fired off the same linguistic neurons. I don't know, I'm not really sure how these large language models work.
(Score: 5, Interesting) by ElizabethGreene on Monday August 21 2023, @07:14PM
You've got the right idea. LLM's use vector memory. It's like a large room with concepts at certain positions in the room. Somewhere in that space is a position for the concept of "de-hungry" and "Food bank", "restaurant", "refrigerator","kitchen", etc are probably nearby in that space as general relations to the idea of de-hungering.
(Vector memory has more than 3 dimensions, but that's an easy way to picture it with our meat brains.)