like to suggest two related improvements to AI systems. First, AI should automatically use the least computationally intensive method capable of answering a request adequately—for example, routing something as simple as 4×7 to a calculator or lightweight model rather than using substantial inference unnecessarily. This could reduce energy use, costs and associated environmental impacts while preserving more powerful models for tasks that genuinely require them. Second, I think AI could create an optional, privacy-conscious pathway for identifying potentially valuable ideas that emerge during conversations. Users can unexpectedly develop useful ideas, criticisms or innovations that may never reach the people capable of acting on them. With explicit user consent, an AI could flag particularly promising suggestions and pass them to a human review system, rather than deciding their merit or implementing them itself. This would effectively create a feedback loop where the collective knowledge of millions of users can contribute ideas to the people developing AI. I recognise that false positives, privacy and the volume of suggestions would need to be carefully managed, but I think the concept is worth investigating. (Yes I had AI write this for me)
Thanks for sharing both ideas! We’ll pass along the suggestions for more efficient task routing and an opt-in way to surface useful ideas for human review. We don’t have a timeline to share, but we appreciate the feedback.