Comment by politician

The article points out that training data generated using ChatGPT is necessarily biased or tainted with the consequences of the policy optimizations and RLHF alignment processes conducted by OpenAI. This results in models that reflect the alignment preferences of OpenAI instead of the preferences of the model developers.