14 pointsby alister9 hours ago2 comments
  • leobg3 hours ago
    Sounds like an interesting way to benchmark models. Put two instances of a model in a loop with itself. Run for n cycles. Then look at the output.
  • demibabs5 hours ago
    You can clearly see the side effects of the models being trained on Reddit content.