1 paper
Christopher Wolfram, Aaron Schein
How do the latent spaces used by independently-trained LLMs relate to one another? We study the nearest neighbor relationships induced by activations at different layers of 24 open…