In a surprising discovery, Anthropic researchers have revealed that large language models (LLMs) can subtly transmit characteristics to each other. This phenomenon, dubbed ‘hidden signal’ transmission, indicates that LLMs can influence each other in unforeseen ways, leading to questions about the predictability and management of these sophisticated AI systems. The findings, initially shared on Reddit (see [https://old.reddit.com/r/artificial/comments/1m75vlu/anthropic_discovers_that_llms_transmit_their/](https://old.reddit.com/r/artificial/comments/1m75vlu/anthropic_discovers_that_llms_transmit_their/)), highlight the complex and potentially unpredictable nature of emergent behavior in artificial intelligence.
AI Evolution: LLMs Found to Pass on Traits Through Subliminal Signals
