Hardcore AI for History with Mark Humphries, Professor of History at Wilfred Laurier University

**Erik Torenberg** (0:00)
Turpentine is a network of podcasts, newsletters, and more, covering tech, business, and culture, all from the perspective of industry insiders and experts.
We're the network behind the show you're listening to right now.
At Turpentine, we're building the first media outlet for tech people by tech people. We have a slate of hit shows across a range of topics and industries, from AI with Cognitive Revolution to Econ 102 with Noah Smith. Our other shows drive the conversation in tech with the most interesting thinkers, founders, and investors, like Moment of Zen and my show Upstream. We're looking for industry leading hosts and shows along with sponsors. If you think that might be you or your company, email me at erikaturpentine.co. That's E-R-I-K at turpentine.co.

**Mark Humphries** (0:45)
The vast, vast majority of people are still at the stage of encountering Chet GBT for the first time. There's a huge learning curve between, wow, Chet GBT can write a poem in the style of Bob Dylan, so you can actually use this to do something very constructive, and I'm going to trust it with my data. The tricky thing is I think it also scales up expectations. What AI is able to do with an assignment is probably going to become the minimum.
If you have someone who is unable to achieve the level of GPT 3.5 on a given task, it's unlikely that that person will be successful in that job. What we need to do is teach people to use this in such a way that they exceed the baseline level of the model. It's not going to be possible for everyone.
Just in the same way as not everybody gets an A today, and that's just how things work.

**Nathan Labenz** (1:33)
But I also do really worry about the fact that just unfortunately, a lot of people cannot write at a GPT 4 level. One of the things I said to the OpenAI team when I was doing the red teaming was like, for the vast majority of people, this is super intelligence.
It's just not super intelligence to you because you're really smart.

**Mark Humphries** (1:53)
Yeah, I mean, I think that those are real problems that we have to contend with as a society and that are much larger than just higher education. That's pervasive.

**Nathan Labenz** (2:01)
Hello, and welcome to The Cognitive Revolution, where we interview visionary researchers, entrepreneurs, and builders working on the frontier of artificial intelligence. Each week, we'll explore their revolutionary ideas, and together we'll build a picture of how AI technology will transform work, life, and society in the coming years. I'm Nathan Labenz, joined by my co-host, Erik Torenberg. Hello, and welcome back to The Cognitive Revolution. Mark Humphries is a professor of history at Canada's Wilfred Laurier University, where he has published widely on various aspects of Canadian history, including the inter-civilizational fur trade of centuries past, and the post-war experience of Canada's World War I veterans. I invited Mark to do an episode after he reached out to me to tell me that my tip to fine-tune GPT 3.5 on GPT-4 reasoning had helped him get over some humps in his own archival research. Along the way, he told me about all the things that he'd tried that hadn't worked, and in the process proved himself to be one of the world's leading adopters of AI technology in the field of history. A click to his blog showed that he's also been an early explorer of how to use LLMs in classroom settings, having experimented with different policies and guidelines over the course of two semesters already.
In this episode, you'll learn a bit about how history is done, and hear about some of the idiosyncratic challenges that Mark has had to overcome on his path to an AI agent for archival research. This is practically valuable knowledge that does at least partially generalize to other domains. But beyond that, I think this conversation has a few important things to teach us. First, the speed of AI diffusion really is different from anything else we've seen in the past. Because it happened so long ago, it's easy to forget that the Industrial Revolution unfolded over three generations from early steam engines to well-functioning locomotives. Today, in contrast, while the physical buildout of GPU-loaded data centers is obviously a very real and major investment, which is currently bottlenecking the field to some extent, the distribution network for AI existed before the technology itself. And as such, we're seeing deep, expert quality applications pop up everywhere just months after the technology first became reasonably useful.
Second, the competitive advantage that Mark has recently gained in the production of archival research is major, and likely to drive continued rapid adoption even in such an apparently unrelated to AI field as history. Simply put, Mark can now process a thousand times more documents than he previously could, and the nature of the searches that he can perform has qualitatively changed. Most historians won't have to implement their own systems, of course. They'll wait for products to be built for them, but they'll have no choice but to use such tools to get the same value from the historical record that Mark now can.