Serious question: How do you avoid making a model bigger than GPT-4 when you don't know anything about the architecture? I guess you could try comparing capabi

Serious question: How do you avoid making a model bigger than GPT-4 when you don't know anything about the architecture? I guess you could try comparing capabilities, but without access to the training data we have basically no effective way of running benchmarks against the GPT-4 api, even if you are one of the lucky few who already has API access.