Perhaps the key sentence: > Large language models have no concept of “truth” — they just know how to best complete a sentence in a way that’s statistically pro

Perhaps the key sentence: > Large language models have no concept of “truth” — they just know how to best complete a sentence in a way that’s statistically probable these many-parameters model do what it says on the tin. They are not like people, which, having acquired a certain skill, are very likely to be able to adapt its application to a different social/technical scenario, by adding constraints and assumptions not inherent to the application of the skill itself.