Matthew Barnett comments on Forecasting progress in language models