Is there a way to always stay up to date with the latest and best performing models? Perhaps it's me but I find it difficult to navigate HuggingFace and find models sorted by benchmark.

That's according to this (https://lmsys.org/blog/2023-03-30-vicuna/) promotional blog post and just cited by the google memo right? Which isn't really even a doc, just a memo that was circulating inside google.

I also find it strange they don't contrast gpt4 and gpt3.5

This assessment is based largely on GPT-4 evaluation of the output. In actual use, Vicuna-13B isn't even as good as GPT-3.5, although I do have high hopes for 30B if and when they decide to make that available (or someone else trains it, since the dataset is out).

And don't forget that all the LLaMA-based models only have 2K context size. It's good enough for random chat, but you quickly bump into it for any sort of complicated task solving or writing code. Increasing this to 4K - like GPT-3.5 has - would require significantly more RAM for the same model size.

Against GPT3.5 perhaps the gaps aren’t too big for your use cases, but I wouldn’t say it’s in the GPT4 league. It looks close in the benchmarks but the difference in quality feels (to me) huge in practice. The others models are simply a lot worse.