Raw LLM Responses

Generic LLMs are almost never useful for anything domain-specific. This study looked at versions of Mistral fine-tuned for the basic tasks of document retrieval, classification, and clustering. It appears no effort was made to pre-train or post-train the models to remove bias. What this shows is that AI must be developed and used intentionally and responsibly, not that it’s impossible to make AI unbiased. Companies developing hiring-specific solutions would hopefully be incentivized enough to address these problems, and ones that don’t hopefully face a litany of lawsuits.