> Twitter's communication style [...] seems particularly unlikely to be helpful for optimising LLMs

This depends on what one wants to optimize the AI for. ;-)