Anthropic NLAs translate LLM activations to human-readable text for safety #1 Post by sebastianperezr » Sat, May 09, 2026, 2:54 PM UTC Anthropic NLAs translate LLM activations to human-readable text for safetypresciente.com