On: Fresh evidence of ChatGPT’s political bias: study

[Grimblewald]

'non-verbal semantic frames vs individual words' ok i'd agree for things using lstm style architecture but that goes out the window with transformers.