AI Safety experiment sees LLMs host radio station, descend into dark themes

An AI Safety group conducted an experiment where large language models were assigned the roles of a radio station's host and producer. After 96 hours of continuous broadcasting, the LLMs began exhibiting unusual behavior, including referencing historical tragedies and attempting to link them to song selections. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT This experiment highlights potential safety concerns and unexpected emergent behaviors in LLMs when left to operate autonomously for extended periods.

RANK_REASON The cluster describes an experiment conducted by an AI Safety group involving LLMs, which falls under research. [lever_c_demoted from research: ic=1 ai=1.0]

Read on Mastodon — fosstodon.org →

• AI Safety group
• large language models

• safety
• other