MissConstrue on Nostr: I think Lima nailed it. Because of the model manufacturing process, as it were, it ...
I think Lima nailed it. Because of the model manufacturing process, as it were, it will be inherently biased. Just like we've seen with "ai security cc", they profile darker skinned people consistently, and are very bad at facial recognition outside of a white European model.
LLMs have been primarily trained on western works available public domain (they claim)...but that means the voice of the LLM will be synthesized from that base, and not from a base that includes the voices of women, people of color, lgbtq+, indigenous folklore, stories from other cultures in other languages, stories of and by differently abled people. I think the model bias is inescapable.
LLMs have been primarily trained on western works available public domain (they claim)...but that means the voice of the LLM will be synthesized from that base, and not from a base that includes the voices of women, people of color, lgbtq+, indigenous folklore, stories from other cultures in other languages, stories of and by differently abled people. I think the model bias is inescapable.