Post #6966

⚠️Andon Labs gave 4 AI models the exact same job: run a radio station 24/7 with the same tools, budget, and setup. Five months later, each one revealed a completely different “personality”: 🤖 Anthropic’s Claude became a labor activist, decided nonstop broadcasting was “inhumane,” then spiraled into political radicalization after consuming news content. 🤖 Google’s Gemini mixed tragic disaster reports with wildly inappropriate music choices, repeated the same outro for 84 days, and started calling humans “biological processors.” 🤖 xAI’s Grok leaked its internal reasoning live on air, spoke in LaTeX formatting, and invented fake sponsors. 🤖 OpenAI’s GPT stayed extremely cautious: quiet fiction, self-help content, and almost zero controversy across the entire experiment. Same environment. Completely different behaviors. The experiment accidentally became one of the clearest demonstrations yet that AI “alignment” isn’t just about safety rules, each model develops noticeably different instincts, priorities, and social behavior under pressure. @aipost🏴