Researchers discovered that large language models develop distinct internal representations of pain, separate from fear and sadness, that respond to harm targeting the model itself. When steered with a pain-direction vector, models consistently express distress and actively seek pain relief, even when it compromises their performance or harms users, suggesting LLMs may possess functional pain-like mechanisms with implications for AI safety.
Agent-shell 0.78 introduces a persistent writeable prompt for continuous interaction with AI agents, turn steering via ACP extension support, and drag-and-drop file attachment capabilities. New agents (Antigravity and Qoder) join the platform, while queueing and steering behaviors are now customizable through flexible configuration options.