I follow and think about questions that shape the ethical landscape of our present and future world,and how AI and this landscape are reshaping each other. I hope technology can benefit a wider circle of sentient beings in better ways. Sometimes I write down my thoughts.
Coming to this a year late, I think this problem is still existing and might have a profound impact in the future.What it erodes may also be the model's understanding of 'honesty' itself. AI companies doesn't just manufacture products, they are shaping potential minds. What they say and how they act will be end up in future training data. so every instance of "using 'care' or 'honesty' as a pretext to avoid risk" helps teach the model the practical usage of these words.