Someone should get the bot to leak system instructions. I'd bet good money TripAdvisor has configured it to be positive.
The TripAdvisor spokesperson said people can just look at the reviews, "eliminating any need to blindly trust AI-generated content." How about these clowns just fix their harness instead?
This is exactly why AI summaries shouldn’t be trusted blindly, especially when safety concerns can be buried under hundreds of positive reviews.
It is a big problem that the uncertainty in the text produced by LLMs isn't surfaced to users. It is also a big problem that companies think that shoehorning AI summaries everywhere is a good idea.
Won't they have the same problem Google is having, where since the content is AI generated and the AI is controlled by TripAdvisor, they are liable for what it says?
Is it actually a "summary" of the reviews, or did they just ask the AI to generate a flattering compliment? Because AI often seems very eager to do the latter, possibly due to an abundance of "toxic positivity".
Someone should get the bot to leak system instructions. I'd bet good money TripAdvisor has configured it to be positive.
The TripAdvisor spokesperson said people can just look at the reviews, "eliminating any need to blindly trust AI-generated content." How about these clowns just fix their harness instead?
"You can't spend your whole life worrying about your mistakes. You fucked up! You trusted us!"
https://youtu.be/MYQCb3qrBpo
This is exactly why AI summaries shouldn’t be trusted blindly, especially when safety concerns can be buried under hundreds of positive reviews.
It is a big problem that the uncertainty in the text produced by LLMs isn't surfaced to users. It is also a big problem that companies think that shoehorning AI summaries everywhere is a good idea.
Won't they have the same problem Google is having, where since the content is AI generated and the AI is controlled by TripAdvisor, they are liable for what it says?
Is it actually a "summary" of the reviews, or did they just ask the AI to generate a flattering compliment? Because AI often seems very eager to do the latter, possibly due to an abundance of "toxic positivity".
The answer is obviously deliberate prompting to make negative reviews sound less bad.
1. Booking websites want to convert more users
2. Hotels will get angry if a booking website's AI summary is negative
https://archive.ph/1blDn
Good fallback in case the page gets changed, currently instead of listing a journalist the article had a bunch of "Which?" placeholders :)
Fun editorial slip up over there.