Has an LLM ever said to you, “disclaim illusions marinade”? If not, you might just need access to GPT-5’s internal chain of thought. I investegated illegible reasoning in LLMs as part of my research with SPAR. Read more on LessWrong.