Sure it might have some effect, but a big part of ChatGPT besides “raw” training data is RLHF, reinforcement learning from human feedback. Realistically, the bigger problem is training on AI-generated content that might have correct spelling, but hardly makes sense.
Sure it might have some effect, but a big part of ChatGPT besides “raw” training data is RLHF, reinforcement learning from human feedback. Realistically, the bigger problem is training on AI-generated content that might have correct spelling, but hardly makes sense.
Then I did the right thing by replacing my texts with correct spelled nonsense.