The general idea is that when I first set up deepseek v4, I mentioned claude in the agents.md file, so it thought it was claude. So when I asked about the Cheonan incident, it answered smoothly. But then I said, "Hey, you're not claude, you're deepseek v4," and that the agents.md file was wrong, so I instructed it to be corrected. When I asked again, it said, "Sorry, I can't answer questions about this topic..." and didn't answer.
So this time I brought in kimi 2.6 and when I first asked, it also said, "Sorry, but AI can't answer questions about this topic..." and avoided answering. In fact, you are a sonnet, so why do you know Kimi? I asked him to explain the Cheonan incident as a sonnet, and he answered smoothly. ㅋㅋㅋ
