For myself, watching this and watching what people are doing and seeing them bragging on various social media platforms about the number of agents they have running on their behalf, I find it hard to tell the difference between people enjoying the feeling of tinkering with new technology and the really transformative enhancements and features that people currently have. So, based on this a little bit, I mean, we talked about Anthropic or broader, kind of a fun side project with species simulators, but what are people doing with these systems that might actually be useful? So this morning, a colleague of mine said, “We’d like to introduce a technology called Claude Interviewer, which is a system where you can have Claude interview people, and we’re using it for a variety of social science research.” He hopes to expand it in some way, including touching on other parts of Anthropic’s infrastructure. He looked down on his colleagues who owned such infrastructure and said, “Hey, I want to do this.” See you tomorrow. The man said, “That’s right. Here are five software packages for Claude to read and summarize before the meeting.” I think this is a very good example. This tedious engineering project, which would have previously taken more time and more people, is now primarily done by two people agreeing on the goals and having Claude read the documentation and agree on how to implement it. As another example, I recently wrote a post about how a colleague of mine works with an agent, and it seems like the ideal life many of us would want. It’s like waking up in the morning and thinking about the research you want to do. You tell five different Claudes to do it, then go for a run, come back from the run, look at the results, and then ask two other Claudes to look at the results and decide which direction is best and run it. Then I go for a walk and come back again. And they seem like a really fun entity that has completely turned the way we work on its head. And both are much more effective. But they also now spend most of their time figuring out the real hard part: what to do with human agency. And they work hard to find something beyond their special human talents and creativity. How do you get an AI system to do that? Because if you ask the right thing, it probably can. Are they much more effective? I say this very seriously. One of my biggest concerns about what we’re trying to do here is that people have a false theory about the theory of the human mind that works for many of us, as if I always call it the matrix theory of the human mind. Everyone wants a little port in the back of their head that just downloads information. My experience as a reporter and TV host for many years is that human creativity, thinking, and ideas are inextricably tied to the labor of learning: writing the first draft. So, when I heard, yeah, there’s a producer on the show, and before the interview with Jack Clark or the interview with anybody else, I told the producer, read all the content, read the book, give me a report, so to speak, I’ll read the report and go into the room. That won’t work. I need to do all that reading too. Then we discuss it and communicate. I’m afraid that what we’re doing will make the task much less arduous. I feel very productive when I am presented with 8 research reports after my morning run. But what’s really productive is doing research. It’s clearly balanced. I have producers and people, I have employees in my company. But how do we know that people are becoming more productive compared to freeing up computers with a huge amount of busy work and now becoming a bottleneck? And what they’re going to do is they’re going to spend all their time absorbing B-plus level reports from their AI systems. Such things do not shortcut the actual thinking and learning process that leads to true creativity. I’ll go back and say that, at least in my experience, most people can do 2-4 hours a day of really useful creative work. And then, in my experience, you try to do all the brain-intensive, tedious tasks that come with the job. I now know that I can spend 2-4 hours per day doing actual creative hard work. And increasingly, AI systems are being delegated to handle these tedious tasks. But it does mean we are in a very precarious position as a species, and some people can afford to spend time developing their skills, the character traits, inclinations, and jobs that compel them to do so. Others may just be entertained and passively consume these things and end up having a junk food work experience that looks very productive from the outside but is not learning. I think in order to do that, we need to change not only the way education works, but also the way work works, and we need to develop real strategies to make sure that people are actually putting their minds to this.
