Researchers hope recursive self-improvement could tackle the skyrocketing costs of developing advanced AI models. But is it viable?