Run it on a fixed model (Llama2, so repeatable) then create a big fat vector including “username embeddings” into a linear regression to score. I say this beca

Run it on a fixed model (Llama2, so repeatable) then create a big fat vector including “username embeddings” into a linear regression to score. I say this because this will then score users based on their good or bad (let’s not use the wsb disability-offensive terms!) I think you really need to know who is saying it. Even then the potential for flukes over methodology on a stonks gambling sub makes it very noisy I suspect. Or is this just for fun :-)