Since LLM predicts what human writes, and most LLMs are trained on the internet…Yeah I am not surprised
Midnitte@beehaw.org 2 weeks ago
Here’s an article about it from Mar 5 for those who don’t want to listen/watch.
Interesting that it could perform better, but regardless enter meltdown status… I wonder if they tested Human+AI…
The fact it entered some conspiracy mode is… maybe that says something about us, lol
sculd@beehaw.org 2 weeks ago
LandedGentry@lemmy.zip 2 weeks ago
Midnitte@beehaw.org 2 weeks ago
I wonder what level of writing it performs if you were to remove all of his tweets from the database before building the model…