Researchers from multiple companies, including OpenAI, have called for deeper focus on how AI reasoning models work - an area that remains poorly understood.