I often enjoy discourses where articulated, well-aimed, and thorough arguments are exchanged between the debating participants. But I also can't help but marvel at the spartan unambiguity of points like yours, especially when fired at point-blank range.
The two last scenarios are wrong. You won't get electricity at "fair prices". Because when one of AI/EV fails, the other will grow to consume all low-price electricity available.
LLMs are not suitable for science and engineering. The AI I was using the other day got LeChatelier's principle exactly backwards. The wording is sometimes too subtle and precision matters. I could imagine a 2nd AI layer that is trained to double-check and correct errors like this. If the wording and syntax are stiff, like in a programming language, ot seems to work much better. I had it generate a Powershell script today, it worked right away.
The fact that a newspaper/book/catalogue/whatever is printed does not infer human curation of the content. I think that, in the future, the deciding detail will be the date of publication (2022, the year ChatGPT came out). Since pretty much all printed material includes this information for copyright reasons, media printed before 2022 will become valued vintages.
If what you say is true, then it's going to be much worse this time around. Because now, in contrast to the 90ies, there is also SpaceX. NASA is going to be blamed for doing a crappy job, while SpaceX will end up filling the national pride gap. Then it's only a matter of time before NASA is deemed redundant and obsolete.
30 GB of scratch is a lot for normal bureucrats. I assume here that South Korea does not have 28000 workers doing numerical simulations and heavy-duty film editing. This was asking for trouble since nobody would bother cleaning up that space: archive old stuff on backed-up space, delete junk etc. I would think that the Korean government lost about 100 TB worth of actually useful data and the rest was obsolete junk anyway.