Is the gap between open and closed models growing? Evidence from WeirdML — LessWrong