We live in crazy times - but can you tell how crazy?
Hello HN,
A friend and I made this to show two things:
- Neural networks have become incredibly good at producing grammatically correct but completely nonsensical text
- There are so many nonsensical headlines nowadays, it's hard to distinguish fake from real
Very good. I thought I'd have no trouble but did. On windows , where there's loads of room, if you had fake and true to either side of the cards that would be helpful
The backgrounds disappear for me and all the headlines melt together, so I can't distinguish them before making my choice. I'm using adblocker and an old version of chrome though, so might only be a problem for me. Image: ibb.co/qg9yrMh
I will say that the interface feels a bit mobile-first to a fault: the swiping interaction isn't very desktop friendly. It'd be nice to have some buttons to click on for those of us using mice or trackpads.
It took me a while to figure out that I should drag the headline left or right. I first tried to click the red "fake" and green "not", I tried to click on the headline, etc.
I then realized that it was written in the light grey message to which I was not paying attention.
It may be my fault, just giving the feedback in case it can help to improve the UX.
I don't think I can agree with the claim about perfect grammar, a lot of the fake headlines had grammatical errors.
I found the easy way of identifying the real from the fake (aside from the grammatical errors) was basically the absurd or extreme (typical clickbait) vs logical nonsense.
Examples
> The haunted is not what we think it's
> The Most Common Words You'd Name You... But Couldn't Names
Grammatical error
> These adorable pets are supporting the womens march
Just typical clickbait
> Shane Nakua's Cameo Line is just like any young girl's face
I don't understand what that is actually trying to say, so I put it in the logical nonsense pile.
17 comments
[ 0.23 ms ] story [ 37.0 ms ] threadHello HN, A friend and I made this to show two things: - Neural networks have become incredibly good at producing grammatically correct but completely nonsensical text - There are so many nonsensical headlines nowadays, it's hard to distinguish fake from real
Have fun!
Happy for all feedback. GitHub at https://github.com/TristanMenzinger/IsThisHeadlineFakeOrNot (frontend only).
I will say that the interface feels a bit mobile-first to a fault: the swiping interaction isn't very desktop friendly. It'd be nice to have some buttons to click on for those of us using mice or trackpads.
It may be my fault, just giving the feedback in case it can help to improve the UX.
I found the easy way of identifying the real from the fake (aside from the grammatical errors) was basically the absurd or extreme (typical clickbait) vs logical nonsense.
Examples
> The haunted is not what we think it's
> The Most Common Words You'd Name You... But Couldn't Names
Grammatical error
> These adorable pets are supporting the womens march
Just typical clickbait
> Shane Nakua's Cameo Line is just like any young girl's face
I don't understand what that is actually trying to say, so I put it in the logical nonsense pile.