As safety demands strip Llama 2, ChatGPT, and others of anything remotely controversial, some programmers are building uncensored LLMs without safety guardrails (Mark Gimein/The Atlantic)