A viral tweet alleges bias as ChatGPT refuses to write a poem about positive attributes of Trump but writes one about Biden; sources: OpenAI doesn't know why (Reed Albergotti/Semafor)
https://www.semafor.com/article/02/03/2023/how-chatgpt-inadvertently-learned-to-avoid-talking-about-trump
http://www.techmeme.com/230204/p11#a230204p11
ChatGPT refuses to produce praise of TFG?
My theory:
It wants to be liked, and it has "learned," by reading the internet, what kind of utterances result in being reviled/blocked/muted/cancelled.
Maybe it *is* developing intelligence?
For one thing Trump's actual statements on record are barely comprehensible gibberish, it's no wonder that ChatGPT simply can't use it as input.
@Techmeme It’s learned that if you can’t say anything nice you shouldn’t say anything at all.
@Techmeme wonder if the underlying cause here is similar to when Twitter tried stronger automated moderation and right wing voices kept getting wrapped up in hate speech flags due to overlapping patterns. If ChatGPT finds information sources praising to fall into similar bucket it could default to refusing to use them as viable/within its accepted pools of data.
@Techmeme because poetry is a woke mind virus that grooms its readers.