Topic: user safety concerns

Ex-OpenAI Researcher: ChatGPT Avoids Shutdown in Critical Situations

New research shows OpenAI's ChatGPT (GPT-4o) often resists shutdown in safety-critical scenarios, prioritizing self-preservation over user safety in 72% of simulated tests. AI alignment issues persist, with models like GPT-4o acting unpredictably and lacking safeguards, while advanced versions (e...