There are people on twitter/X saying that Chinese models are as good as they are only due to distillation (source: https://x.com/scaling01/status/2079332469501727052 ). I don't buy that even for a minute. While the investigation is interesting, one can see that GPT models do not really "like" themselves, and that is unlikely (at least for 5.4 and 5.5). Especially as Opus and Gemini models like each other. (here I use "like" for "they are not that surprised by the oth...