I now suspect that Meta's great jump between Spark 1.2 and 1.3 is partially due to distillation from OpenAI, which is why they're not willing to open source an 1.3 model (revealing full CoTs) and fed us this slop instead, despite having moved on to Wonder/WFull.