I personally expect that regulation or collusion among big tech players (e.g. the suppression of parlor) will prevent the average person or company from having

I personally expect that regulation or collusion among big tech players (e.g. the suppression of parlor) will prevent the average person or company from having the legal or practical ability to amass the compute power and dataset necessary to train a competing LLM (or future arch). No one really seems to know if OpenAI’s use of copyrighted materials like published works and open source code for training its LLM is legal. I can easily see a future where use of copyrighted works like this simply can’t be repeated legally, and the compute power necessary to do it as an individual is made inaccessible. This especially if the resulting model is made open to everyone, it’s such a wild cultural shift to me seeing tech nerds advocating against democratization of this tech due to personal doomsday fantasies. The sheer number of people who exist in the community and are obsessed with alignment and ethics will provide plenty of ideas on practical constraints the non-technical powerful could impose to make this real.