It takes a bit to extract the relevant story from the paper (21 page PDF) but happily this article captures the gist: "the AI models struggled to match the level of casual negativity and spontaneous emotional expression common in human social media posts, with toxicity scores consistently lower than authentic human replies across all three platforms." So, um, I guess that's a good thing, though it speaks poorly about the human contributions to social discourse. Interestingly, it's not really a finding that has much applicability outside those social media forums where casual negativity and toxity are commonly found. Via Apostolos K.
Today: Total: [Direct link] [Share]
