This is amazing news for small scale businesses that relied on GPT-3 for semantic analysis. I guess smaller model size should permit in-house hosting.