Interesting small Nature paper on how LLMs "perceive" the user based on the dialect used. Given how LLMs will be more and more commonplace going forward - often without the user knowing that they are interacting with a LLM - this might have relevant implications for how the user is treated by the LLM (e.g. imagine a LLM that filters out applicant suitability for a loan in a first pass black box filter based on dialect assumptions).

"LLMs produce racist output when prompted in African American English"

https://media.nature.com/original/magazine-assets/d41586-024-02527-x/d41586-024-02527-x.pdf

#AI #LLM #chatgtp #future #science