Sources: bias

Thoughtful and provocative response to claims of implicit gender bias in GPT-4. The critical and refreshing claims:

1. researchers are too prone to anthropomorphizing LLMs in interactions and interpretations, and
2. LLMs are not as good at language disambiguation as we think.

The lesson we need to keep in mind: language is full of ambiguity that we don’t appreciate because we can negotiate it without much effort, but computers, no matter how impressive, can’t yet.