There are already a lot of options for running LLMs with open weights artifacts, trained with a variety of sources. The real question isn’t which ideas they have. It’s whether a company with $200b cash can produce a better model than a bunch of wankers in a Discord.