3 ms·
I do not have a grand unified theory of what is or isn't racist. Nor do I need one in order to understand the concept that racist biases implicit in an LLMs tra
by foldr 3y ago
I do not have a grand unified theory of what is or isn't racist. Nor do I need one in order to understand the concept that racist biases implicit in an LLMs training data can be reflected in its output, in a way that makes that output offensive to large numbers of people and unsuitable for general purpose use.
Fussing over the semantics of whether or not the LLM itself is 'racist' is a complete red herring. The article merely says that the LLM may 'use racist stereotypes' or 'offer racist replies'. It's perfectly clear what is meant by this, even if you disagree with some specific examples.
- _heimdall 3y agoI don't see it as just semantics here. > The article merely says that the LLM may 'use racist stereotypes' or 'offer racist replies'. Using racist stereotypes and offering racist replies are fundamentally different as I read it. Using racist stereotypes could be as simple as parroting statements learned without comprehending or meaning to replicate the intent behind it. Offering racist replies would imply that the speaker is themselves racist, meaning the context behind the replies is that they actually see one race as lesser than another simply because of the racial difference. Is this really not a meaningful difference for you? I honestly don't mean this to sound like a condescending question, I'm just surprised if what seems like a huge difference to me is effectively meaningless to someone else.
- foldr 3y agoThe LLM could reply “Ethnicity X is dumb and stinky”. That’s a racist reply. As I’ve said over and over, we agree that LLMs can’t literally hold racist beliefs (or beliefs of any kind), but it’s not difficult to understand what the article is talking about or why one would want to avoid it.