That's exactly my understanding as well. This is, essentially, the LLM hallucinating user messages nested inside its outputs. FWIWI I've seen Gemini do this frequently (especially on long agent loops).
That's exactly my understanding as well. This is, essentially, the LLM hallucinating user messages nested inside its outputs. FWIWI I've seen Gemini do this frequently (especially on long agent loops).