We Can Ask Bigger Questions Now

I read a post by Chris Albon, who leads machine learning at the Wikimedia Foundation, about realizing that ChatGPT could have produced large parts of his PhD dissertation. He had worked hard for it and mourned for a few days.

Over time, he became sad for a different reason. With ChatGPT available from the beginning, he believed he could have pursued a much bigger research question, one he couldn’t have tackled without AI assistance.

OpenAI’s recent release of proposed solutions to hundreds of open mathematical problems brings that reaction into sharper focus. The results are at different stages of verification, and the model behind the release is still internal.

Oxford mathematician Ben Green said the claims covered roughly three quarters of the aims of a research grant he received in June. He thought the work of understanding the results was substantial and likely to be worthwhile. Bryna Kra, a Northwestern mathematician and former president of the American Mathematical Society, raised a concern about the way these results were released, closing off research directions while short-circuiting the understanding needed to build on them. The difficulty of checking and understanding these proofs deserves attention.

For the solutions that hold up, I see knowledge we can build on. When a problem that resisted decades of effort has an answer, I want us to understand it and explore what becomes possible because of it. I don’t think the loss of a familiar research direction should set the limit on what we try next.

If I were starting a PhD program today, I’d be looking at bigger questions and learning how to work with the models available to me to see what we could accomplish together.

I bring the same ambition to my work as a product leader. I want teams to take on customer problems that would otherwise be too costly to explore. In the builder model that I advocate, a small group with product, design and engineering depth can work through an idea in a running prototype and take it into an A/B test. They can change it together and learn from people using it, with AI extending what they can attempt.

For leaders, making room for that work means deciding which existing commitments can wait. I want to give people time to learn these tools and pursue ideas whose value is still uncertain. We have to build and evaluate enough to find out whether a promising idea actually helps someone.

I believe we are entering a period of growth and opportunity that could surpass anything in human history. I want to help people grow into the work that becomes possible, and build things we couldn’t have attempted before.

This writing reflects my personal perspectives on product management, AI, and content discovery. It does not represent the official position of my employer or any affiliated organization.