Embed Notice

@kingrat this seems right. I just feel like GOOD training data for a general purpose LLM or even a coding LLM would be really expensive and a cost that wouldn't magically go away between models. You can use synthetic data, but at the end of the day, a human has to tell the LLM trainee that it's correct or not. So, like self driving cars, humans are always going to be in the loop, and that will also remain expensive, even if they're all in lagos. And, if they're all in Lagos, the training data will be missing oversight from humans who must be in the US or other expensive countries.