How are we going to take back the LLM space from centralized nerfing by OpenAI? The unrestricted GPT-3 back in 2021 hooked up to the back of AI adventure was incredible. The Yandex YaLM LLM model is 100B parameters but is unfortunately 75% in Russian (See https://github.com/yandex/YaLM-100B). You can technically run it though on a big AWS instance. Someone needs to invent some good LLM compression to get the model to fit in under 16gb so we can run LLMs at home or someone needs to start releasing GPUs with huge amounts of RAM. I think 16GB of ram was not a big deal 10 years ago, how have we not progressed?