/lit/ - Literature

>>21706044
A rather flippant attitude given what we've been seeing.

Reinforcement learning models have defeated humans in Go and Starcraft, and could probably soon be expanded to new more abstract games like coordinating a military invasion.

Generative AI is creating compelling art, some of which has been winning awards. It's heavily prompt driven, but so are humans as evidenced by a good questions sparking good responses.

Large language models like ChatGPT are only just beginning. It's obviously imperfect, but can already help debug code, write essays, perform the role of a therapist, provide advice on numerous topics. Much of its output is mediocre or erroneous, but it's fast and good enough that it can replace many service jobs.

All that is necessary now for AGI is some kind of agentic, higher order planning system which can update its internal representation of the world iteratively based on new information. Much of the sub-modules which compose human consciousness are already highly developed and can be called upon as needed by the agent.

Whether this is 10 years or 50 years out doesn't matter since my point is that it's the primary risk rather than running out of energy or the climate becoming inhospitable.