GPT4 is more difficult to measure I think. The value I get from GPT4 is in the details it gets right on very obscure, complex questions. I'm not sure benchmarks

GPT4 is more difficult to measure I think. The value I get from GPT4 is in the details it gets right on very obscure, complex questions. I'm not sure benchmarks are capturing how far GPT4 is ahead of other models. For simple stuff it's not that much better than 3.5.