Dolly 2.0, a recently released open-source LLM, is a big deal, but not because there's anything new or special about it, but rather specifically because it's legally airtight. Unlike the models that were fine-tuned based on the leaked LLaMA model (many of which Meta takes down), it's based on the EleutherAI pythia model family. The training data is CC licensed and was crowd-sourced from 5000 Databricks employees.
Since it's on Hugging Face Hub, you can use it with Langchain, and I expect that it will become the Stable Diffusion of LLMs. I think especially companies that legally can't send their data off to the US will flock to using Dolly.
I kind of like how there's still this theme of woolly hooved animals (in this case a sheep), but still a divergence from the LLaMA strain of models (llama, alpaca, vicuna). I don't like how it sounds too similar to "DALL-E 2" when spoken though.