Comments by "" (@diadetediotedio6918) on "Is AI really getting dumber? Llama2 vs GPT-4" video.

  1. I think their tests against open source models are biased, they used just a small subset in the reports (even when powerfull models like Guanaco are out there) and I'm suspicious on the quality. But, with my personal tests, the quality is reasonably good and it can produce parseable JSON so it is pretty useful anyway.
    2
  2.  @tipprich  It was not, most of these "tests" are based on confirmation biases and information that was already disponible and spreaded online before, if you try to give it something specific it will struggle no matter how simple it could look.
    1