• hydroptic
    link
    fedilink
    arrow-up
    3
    ·
    9 months ago

    It’s a bias in the training data, not a lot of pics of people eating eating compared to all the stock photos of women smiling at salad etc.

      • hydroptic
        link
        fedilink
        arrow-up
        1
        ·
        edit-2
        9 months ago

        Oh yes my point was that you’re exactly right. Current neural net architectures need absolutely ridiculous amounts of labeled data to “learn” anything, and they’ll just repeat the biases in their training data.