“The current era of AI models using interpretable, English-based chain-of-thought is expected to end in a few years as more efficient but less transparent paradigms are developed.”