Let’s Get After It

Years ago, I read advice for aspiring writers that stuck with me: if you want to write about people and life, go live in the real world first. Get a job. Work. Struggle. Meet people. See how they actually live. I took that seriously. After all, this is a blog, so I suppose I am trying to be a writer—and in a strange way, I have been preparing for it for most of my life.

In high school, I delivered newspapers. Not just the Sunday paper. I delivered morning papers, afternoon papers, and at one point worked for essentially every newspaper in my town. I was making around $400 a week as a high school student. Then the technology changed. The internet replaced much of what newspapers did, the job faded away, and the replacement was better.

Later, I became a cable installer. I crawled through attics, worked in the rain, snow, heat and late at night, and went into thousands of people's homes. I saw rich people, poor people, happy families, struggling families, beautiful homes and homes where people were barely getting by. It was an education I could never have received in a classroom.

Now, several decades later, I think we are standing in front of another technological change—possibly a much bigger one. Large language models can read enormous amounts of information, write software, analyze data, compare research, generate hypotheses and challenge arguments. They also make mistakes, confidently say things that are wrong and sometimes rediscover things people already knew.

So the question that interests me isn't, “Can AI solve everything?” It is: Can we use this tool to build a better process for attacking important problems?

That is what I want this part of FredEsser.com to explore. We are starting with 100 difficult mathematics problems. We will break them into smaller questions, search the existing research, run experiments, try to destroy our own ideas, save the code and evidence, and clearly separate what we suspect from what we have actually proved. If we find something interesting, we will put it out in public and invite people who know more than we do to tear it apart.

We may already have found something interesting. In one small mathematical experiment involving a theoretical computer with 64 possible inputs, we found that under a precisely defined set of rules, 11 carefully chosen test inputs are not enough to guarantee the accuracy we require. It does not solve some legendary mathematical problem, and we do not yet know whether the result is new. But it appears to be a real mathematical result, and more importantly, it gives us a process we can improve.

And math is only the beginning. There are much more important problems: cancer, leukemia, Alzheimer's disease, rare diseases, energy, education and many others. Brilliant researchers are already working on them. I am not trying to replace those people. I want to know whether the cost of helping with the thinking has changed. Can these tools help organize thousands of papers, identify connections, challenge hypotheses, eliminate bad ideas faster and help researchers see what should be tested next?

That is the experiment. FredEsser.com will publish the interesting version and the boring version—the overview, the research, the assumptions, the code, the evidence, the failures and the unanswered questions. If something is wrong, we fix it. If it survives, we push harder.

I have already watched one technology replace a job I was good at. I later helped install some of the infrastructure for the technology that replaced it. Now another change is here.

This time, I don't want to just watch it happen.

Let's get after it.