Remix.run Logo
secretsatan 38 minutes ago

Simply? It’s flawed if you can’t train on new data past 2022 without first checking ai hasn’t tainted it. Dismissing that as simply avoiding model collapse seems to miss the point.