A rant on LLaMA: please stop training giant language models
severelytheoretical.wordpress.com
severelytheoretical.wordpress.com
We saw this with Dall-E2, there was an incentive to make it seem impossible for this tech to be ran or even safe to be ran outside of the OpenAI walls. This is in part the claims of size and power required but also clever use of safety theatre to make it seem too dangerous for you to be allowed to use it while they claima business advantage.
I believe the same will happen with language models, they will have their stable diffusion moment too when someone is incentivized to make something that would work for the masses rather than incentivized to lock up the value.