Transformers are highly parallelized, and generative AI appends. Even if you ignore that graph problems typically have more time complexity than LLMs offer dur

Transformers are highly parallelized, and generative AI appends. Even if you ignore that graph problems typically have more time complexity than LLMs offer during inference, they are circuit like and not turing machine like. Without getting lucky with pre-training matches, it would be surprising if they were good at it at all.