Title is incomplete. It should be "No real advances in Open Source LLMs since 2019's T5". I think that is fairly close to being true. The real advances have

Title is incomplete. It should be "No real advances in Open Source LLMs since 2019's T5". I think that is fairly close to being true. The real advances have come from OpenAI (and to some extent Anthropic and Cohere). I've played around with some of the open source models and as much as I want them to be "almost as good as GPT-3.5", they're simply not. On the flipside, you can do A LOT with T5 and Flan-T5.