The "AI is fancy autocomplete" vs. "Imminent RSI/AGI" argument.

As AI models get increasingly more capable, including in their ability to conduct frontier research on how to improve themselves, the current public discourse (at least in reddit?), in my view, currently centers around two camps: 1- The people who believe current AI models are still fancy autocomplete - this means they believe these models never developed emergent capabilities or any kind of self-awareness, and will therefore never (or at least will take a long while or another breakthrough) to develop the ability to recursively self-improve, or reach AGI, whatever comes first. Their main argument is that model improvement has largely come through reduced hallucinations, incremental improvements in (manipulated?) benchmarks, better RAG/context management tools, and better harnessing. Those people believe AI labs and media channels are inflating the capability of their AI models as a way of bringing new customers - utilizing fear mongering as a means of "coercing" people into choosing their solutions ("our AI tools are so good they're scary!"). 2- The people who believe current AI models, despite using largely the same techniques than GPT-3.5, have exploded in capabilities, includin