White noise conversation (IMAGE)
Caption
Dartmouth researchers used the neural-network software behind chatbots such as Claude and ChatGPT to convert nearly 1,700 conversations into a numerical map where similar sentences lived nearby each other and dissimilar snippets farther away. They rendered the distances between them as distinct signals that looked like red, white, or pink noise. Pink noise emerged as a 'Goldilocks signal' between intense red noise and aimless white noise that provides stability with the flexibility to change topics. The dialogue pictured shows the semantic distance in a white-noise conversation where the speakers jump between the two topics—hiking and listening to music—very quickly.
Credit
Science Advances/AAAS
Usage Restrictions
Use with attribution
License
Licensed content