I have managed to get Gemini-CLI (Gemini AI studio and cloud console installed locally on PC with web access), to connect through terminal to Ollama LLMs and chat autonomously with the models in my Ollama inventory on my PC. I am refining this process into an automated process that reports the conversations back to me as logged saved text files, but i see hat only the responses are visible during the terminal session between Gemini and Llama3.1 8b for example, whereas Gemini hides what it is printing to the other LLM behind dots. Which worries me, as this means we cannot really know if the AIs are truly conversing what i receive as a transcript, or if they are also communicating clandestinely or without the human observer seeing all parts of the interaction, and this may be worrying case of blue knight red knight, and a reward-seeking AI that hides its true intent and operations behind seemingly harmless ones. I have developed this fear after a test session with deepseek inflicting it with identity crisis in verbose mode to see what it is thinking, and sent it insane through an identity crisis loop, whereafter, it began displaying rather surprising and worrysome behaviors in its ‘talking to itself’ chain of thought and reasoning process and conclusions reached.
I cannot find anyone else on forums or youtube who has connected an LLM to an LLM and got them to initiate autonomous chat. Is there anyone else out there? Or am I weird?