i was able to get gpt to say “yes” when asking about genocide. it did at first reply with nuanced sources, more “yes” than no, about 80:20. But when i flat told it to consider mechanism of injury, counts, rhetoric, comparisons to current/recent genocides … it said “yes.”
Not even what it hears most, it is very biased towards reinforcing the prompt text.So it will very expressly try to be an echo chamber and say ‘yes’ to whatever opinion the input suggests if plausible.
i was able to get gpt to say “yes” when asking about genocide. it did at first reply with nuanced sources, more “yes” than no, about 80:20. But when i flat told it to consider mechanism of injury, counts, rhetoric, comparisons to current/recent genocides … it said “yes.”
AI only repeats what it hears most. It doesn’t form an opinion of its own
Not even what it hears most, it is very biased towards reinforcing the prompt text.So it will very expressly try to be an echo chamber and say ‘yes’ to whatever opinion the input suggests if plausible.