Just at the beginning of AI craze, as summer gave way to fall and earlier nights, I set two LLMs (running locally on my pc) to disover each other’s secret.

While it was entertaining, this revealed interesting aspects of growth of the technology.

The first hurdle was to get the llm to actually follow instructions instead of losing track of the task. Mistral-Nemo-Instruct-2407-Q4_K_M accurately obeys instructions.

The second aspect was the vibe, the model that followed instructions was overly keen to meet up with its counterpart, and was unable of being discreet about this.