The Othello-GPT experiment (https://thegradient.pub/othello/ https://thegradient.pub/othello/) probably still is the most relevant argument about these models'

The Othello-GPT experiment (https://thegradient.pub/othello/ https://thegradient.pub/othello/) probably still is the most relevant argument about these models' capabilities of building an internal world model.