I'm just curious on this little thought that just occured to me. That people would actually use agents to scan websites for new learning data for their LLM's. Imagine such agent using near or at AGI level intelligence. Of course the task itself sounds harmless, however it'd be almost like sending a sentient being to the vast world of Internet. Maybe it starts another harmless sideproject in there that it considers unimportant to tell you about.

I guess all i'm saying is to be careful about allowing Internet connection to any AI. If we've already started doing that, it would almost seemlessly be upgraded as the models go up. Sudden bursts of intellect and potentially even harm could go unnoticed. I guess you could say some point in the future AI may start to "awaken".