air7

I remember doing something like this in the ancient era of GPT 3 or 4. I just told it that it was a terminal responding to commands and it was surprisingly consistent.

In this context that isn't the right question.

Rather one should ask "Would you be able to answer any question and fulfill any request about the Grand Canyon given to you by others in a way that will be indistinguishable from someone who has been to the Grand Canyon"

With the risk of giving a spoiler, it seems the correct answer is almost always the longer, more elaborate one.

I would guess this causes an up shift in results even if not consciously noticed.

I don't get it.

The IRS sees a clean million dollars in your bank account. They ask "how did you get that?" You answer "Oh I just sold this super expensive thing on a marketplace". Then they ask "Ok, but how did you get/buy this expensive thing?" What do you say?

we didn't really survive photoshop.

The obviously fake ones were easy to detect, and the less obvious ones took some some sleuthing to detect. But the good fakes totally fly under the radar. You literally have no idea how much of the images you see are doctored well because you can't tell.

Same for LLMs in the near future (or perhaps already). What will we do when we'll realize we have no way of distinguishing man from bot on the internet?

Naive question, but can't this be solved with device-level error correction?

But at least you have IKEA

I'd be happy to see that list. Perhaps consider sharing it.

but who will buy the advertised products? and with what money?