“Increased computing performance in AI is primarily channeled into safety and functionality, such as enabling models to research answers, reflect on their certainty, and reduce hallucinations.”