One issue I see with AI text prediction services is that over time normal humans’ output could trend stylistically towards being LLM generated and increase the false positive rate significantly. The human version of model collapse if you will.
That is to say, right now, an “average human” likely has 20, 40, 60 etc years of natural language mimicry to go on. However, as time goes on the ratio of “LLM-input” to “non-LLM-input” increases.
I would be interested to know whether this is being monitored, for example I would expect it to be most visible first in language learning communities e.g. new learners of a language beginning to sound more “LLM-y” over time.
esotericn
Karma: 0
The “halo defence” is not, I think, an attempt to actually state that factually the outcome is less likely or has not happened.
It is, rather, that X is not important enough to warrant focus/punishment because of the positive effects of Y.
It is socially costly (basically, negative EV) to be publically seen to claim that X is somehow not a problem, hence the deflection.
To use an absurd example—imagine that my neighbour is a unique person in the world with some sort of superhuman ability to cure cancer or do incredible research that no-one else can do with really high expected value for society. He also may or may not have murdered a few people.
It is less costly socially to argue, even if everyone knows that this isn’t really true, that he probably didn’t do the murders, because he’s such a good guy in other areas, as opposed to arguing the utilitarian “well yeah, he’s a murderer, but he’s saved 5000 people from cancer so like, whatever”.