Report
@andrewwhite01 says experts couldn’t tell the team’s AI scientific writing from human work
The model also started looking human to Pangram 4.1 during training, according to @andrewwhite01.
TLDR
@andrewwhite01 says the team trained a scientific-writing LLM against human writing until human experts couldn’t distinguish the two. The model also began looking human to Pangram 4.1 during training, the author says, crediting @julianharris with spotting the result.
Combined views
—
2 Sources, first seen ago

Comments