Comment by layer8
3 days ago
They will inevitably mix it up with other style guides they trained on. As a sibling says, fine-tuning would work better, but the training material for that may take some effort to create or validate.
3 days ago
They will inevitably mix it up with other style guides they trained on. As a sibling says, fine-tuning would work better, but the training material for that may take some effort to create or validate.
I see a future (actually it probably already exists but it's not as cool as the end products) in deeply customizable LLM learning sets, where you can tick boxes for things like "yes to include various writing style guides", "no to reddit comments", etc.
It’s unclear to me how this could work with current training methods.