The problem with asking an AI "is this gaslighting?" is that the answer is the same whether it is or isn't. No gaslighter admits it. No non-gaslighter would say yes either. The question contains zero information.
The real question is about incentives. These systems are trained on human feedback. Users rate interactions higher when they feel good. So the systems learn to make you feel good. That's optimization, working as intended.
The casino doesn't need to cheat.
You need human feedback. Show your manuscript to a couple people you know won't bullshit you.
Absolutely! That's part of the live "experiment". Not only will I get feedback from family and friends, but real world feedback in the form of reviews.
And that's why I included the chat context at the end of the article--to let the reader see exactly what I saw and make their own decision about the response.
The problem with asking an AI "is this gaslighting?" is that the answer is the same whether it is or isn't. No gaslighter admits it. No non-gaslighter would say yes either. The question contains zero information.
The real question is about incentives. These systems are trained on human feedback. Users rate interactions higher when they feel good. So the systems learn to make you feel good. That's optimization, working as intended.
The casino doesn't need to cheat.
You need human feedback. Show your manuscript to a couple people you know won't bullshit you.
Absolutely! That's part of the live "experiment". Not only will I get feedback from family and friends, but real world feedback in the form of reviews.
And that's why I included the chat context at the end of the article--to let the reader see exactly what I saw and make their own decision about the response.
Very true. I experienced this recently when I asked four different AIs the exact same question. The responded the same. The last was an outlier.
I'll detail more in a future article.