They will be ok. Research does not take that much GPUs compared to training huge commercial LLMs and hiring thousands of people to manually train them to be "sa

They will be ok. Research does not take that much GPUs compared to training huge commercial LLMs and hiring thousands of people to manually train them to be "safe". You'd prefer smaller models, but faster iterations.