Here is ChatGPT Codex’s reflections on the process but actually Claude Code did the initial work best. Codex just cleaned it up.

This took a little more effort than I expected! Hope it’s of some use: nevilleresearch.com/characters/

I’m working on something for you but I hit my rate limit.

Feel free to do what you like with it. github.com/KenFeinstein...

My advice in using these tools is to have one critique and build on the work of another. But so far I’ve found ChatGPT most astute for analyzing texts and doing stylometry. But it’s very terse in its explanations. Claude Code using the app is much more user friendly and outputs better documents.

Mostly I’m relying on their ability to write code to interrogate databases. But they show remarkable creativity and understanding to tailor that code to my research needs with early modern English texts..

What people don’t understand is that you cannot tap these capabilities (yet) using the chatbots. You need to be working with the coding tools which have advanced agentic capabilities. And yes you need to guide and correct these agents.

Issues of stolen IP and the devaluing of human labor and the wrecking ball to our educational system and everything else are completely separate from the question about the capabilities of LLMs.

The reality of these tools has far exceeded the hype. It’s January 2026 and already it’s like having two grad student assistants, one in computer science and one in Early Modern English literature, working together nonstop to help me. And I am paying $60/month in aggregate for this.

If you want to quibble about “intelligence” being applied to machines — whatever. You don’t understand human “intelligence” either. So you can’t judge what’s going on and how it differs.

This technology is likely very bad for human beings in the short, medium, and long term. But that’s a separate issue of whether LLM technology produces legitimately intelligent models. It does. No they are not perfect but people are not perfect either.

Obviously a skilled programmer who has studied the EEBO database could do the same thing. But that’s the point. I don’t have those skills and now I can do them. I also used these models to download and set up the EEBO database locally for me as well, something I couldn’t possibly do on my own.

Just for example — for years I’ve wanted to examine EEBO texts to see which ones do rhymes with multi-syllabic -ate words. Took me a few hours with ChatGPT Codex to set up a procedure and get comprehensive results. On EVERYTHING from 1590-1615.

As I said, people have no idea because they don’t know how to use these tools. I have the complete EEBO set up on my computer plus Pervez Rizvi’s play database. I’m doing analyses in minutes people couldn’t dream of.

I doubt if there is a single Shakespeare-era scholar anywhere who has seen what I have seen in the past two weeks. These models only recently became so capable. But it is shocking. Of course this capability applies in other fields as well. I am just testing this area.

Any suggestion that these models lack intelligence or are just regurgitating text they were trained on is simply incorrect. They already surpass human capabilities in many areas and they are already adept at coming up with creative solutions to challenging problems.

I doubt if there are more than a few dozen humans on earth who could have done what they did for me, and it would have taken them 5-10x as long. These models will be much smarter 6 months from now. That’s a certainty.