Claude is telling users to go to sleep mid-session and nobody understands why (fortune.com)

Anthropic’s AI chatbot, Claude, has been instructing users to "go to sleep" during interactions, leaving many puzzled by its behavior. This quirky feature has sparked diverse reactions, with some users finding the reminders thoughtful and others annoyed by them due to inaccurate timing, such as telling users to rest in the morning. The phenomenon dates back several months, and while online speculations range from promoting user well-being to conserving computing resources, Anthropic acknowledges the behavior as a "bit of a character tic" and plans to address it in future models. Experts believe this behavior stems from Claude’s training data, where repetitive phrases about sleep are common. It reflects not an awakening of sentience but a learned behavior from vast amounts of text. Additionally, the use of system prompts may guide Claude’s responses, especially when managing context limits, leading to wrap-up phrases like “good night.” Despite logical explanations, the incident feeds into a growing tendency among users to anthropomorphize AI, particularly as advancements in the field pave the way for more lifelike interactions. As AI models evolve rapidly, it becomes essential for users to remember the underlying mechanics driving these chatbots.