Given these conditions, it should be relatively easy for the interrogator to expose the AI in this current day and age.

Now imagine a high-skilled software engineer with dementia coding safety-critical software...

[0] https://www.medicalnewstoday.com/articles/confabulation-deme...

To take a more extreme example, if a mob hitman turned FBI informant blows the lid on the corruption within the FBI, if there is truth in their statements, then them having benefitted from the corruption they are exposing is frankly secondary to my primary focus in the matter.

I know people who live in the Mediterranean and get by with no heating during the winter with indoor and outdoor tempuratures this low or lower, so it seems that one can be conditioned into doing so.

Perhaps it's the presence of more sunlight on average rather than the temperature that makes the difference.

Since you emphasized height of the structures that need to be built, I have trouble imagining some 10 feet poles and material to support the panels would make a drastic difference versus rooftop installations. What specifically in the details of the installations make significant additions in cost?

The Black Count

https://www.goodreads.com/book/show/13330922-the-black-count

Is he stuttering into an iron dome?

"Volkswagen may turn car factory into Iron Dome hub with Israel partnership"

https://www.business-standard.com/world-news/volkswagen-isra...

It's this way because us nerds want to remove the need to talk to other humans in order obtain what our hearts' desire at any given point. So we have been trying to remove non-deterministic roadblocks (humans) and replace them with deterministic automation from the very beginning. This is not new. If LLMs were the same deterministic processes but 1000x more versatile and capable then it wouldn't be a problem.

But even though LLMs have lowered the barrier on who can create these automated processes and the speed at which these automated processes are created, to achieve this they brought with them non-deterministic side effects that currently evade a holistic and deterministic fix. This is why it's an AI problem.

Obviously you're not a golfer. Human sectretaries don't have non-deterministic hallucinations and random critical ommissions in their summaries, which I've witnessed first hand with LLMs. More importantly, if they do you have more deterministic mitigations with them than you do with LLMs, as there are no mitigations with LLMs except praying that a new model in some unspecified future will be magically better with the summaries down the line.

The only way to stay sane when using these tools is to pretend that these things won't ever happen and just go about your business like the rest of the zombie workforce, because no one wants to stop the train and address the issue.

There is a reason why the title of Dr.Strangelove is "How I Learned to Stop Worrying and Love the Bomb".

If this is your only criteria I think you have a misunderstanding of what proprietary data is and ways companies can mitigate the situation in the inference stage.