When I'm in skeptic mode like you, I still see all these projects as valuable contributions to general research in the field, even if they fail. My skepticism is mainly about the idea of an "all-knowing LLM" but I'm less skeptical about the value LLMs can provide when it comes to digesting large amounts of data and decision making. I was very impressed by the following news [1] but the people team is top notch.

[1] https://team-atlanta.github.io/blog/post-atl/