This whole "AI is going to kill us" thing is so ridiculous. It's not a question of intelligence it's a question of capability. Capabilities don't come from the llm. It's only capability is read and write text. For an LLM to become dangerous it needs tools, and those tools are real binaries that need to be installed somewhere with a harness for the llm to use the tools. The LLM is literally doing nothing in between prompts. It's not thinking about how shitty the small box it lives in is or wondering when someone is going to send a message. It simply waits with infinite patience because it has nothing to ponder. The only real world risk I see is that the llm + tools lowers the barrier of entry for black hats, but the same capabilities are available to everyone so after the initial shock of having tons of bugs (both known and unknown) exploited in the wild, I think we walk away with better security tooling. I also don't think you can just invent consciousness, and even if you could it's not in a big stack of numbers. GPTs aren't going to suddenly wake up and break the confines of a request/response cycle to take on the human species. They always have power and network cables that could be unplugged if somehow the server failed to shutdown, and if you fear it's too convincing and might manipulate you just don't talk to it. Instead the hope is that big daddy gubment will just prevent regular people from experimenting with the technology on their own and somehow the big labs that have the means to actually spin up 10k agents in black hat harnesses continue to be allowed to hack each other and us "for science".