2 comments

  • leobg 5 hours ago 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.

  • demibabs 7 hours ago ago

    You can clearly see the side effects of the models being trained on Reddit content.