1 Comment

Even if GPT-4 is a combined model, we now have single models close to or even beating some of GPT-4's test metrics. The open source Orca model for one, matches GPT-4 on some metrics while only being 13B parameters. Take this snippet from the technical report: "Orca retains 95% of ChatGPT quality and 85% of GPT-4 quality aggregated across all

datasets as assessed by GPT-4.". It also performs 100% better than Vicuna-13B on zero-shot reasoning benchmarks.

Expand full comment