This is a tricky one. I suspect that the leading models by this date will use a similar approach, but will it be close enough to count as this algorithm? I also wonder how much the likes of OpenAI will disclose about how they trained their frontier models...