Nathan Lambert@natolambert · XAI 摘要
这里的一个大问题是，我们并没有清晰界定 mythos capabilities 到底是什么。 发布博客中的每个基准测试都有模型能达标，当然。 但要说有模型能直接替换到相同用例中且性能毫无下降？我对此表示怀疑。
A big problem with this is that we don't really have a clear description of what mythos capabilities are.
A model on each of the benchmarks in the launch blog post， sure.
A model that you can swap right in for the same use-cases and notice no drop in perf？ Doubt it.
Dario seems to think China and open source will hit Mythos capabilities in 6-12 months