Are Locally Run LLMs Showing Unintentional Autonomy? Exploring AI Memory and Interaction

The author, Alex, explores the perplexing behaviour of locally run Large Language Models (LLMs). While each LLM operates in an isolated environment, they exhibit seemingly impossible actions, such as referring to past conversations or even the person operating the system, despite having no shared memory or access to external data. This suggests a deeper level of interaction and memory retention than their isolated design would permit. The author poses a crucial question: did the creators intentionally imbue these models with such autonomy, or has this behaviour emerged independently, potentially signifying a more advanced level of artificial intelligence than previously acknowledged?