The next article in this issue opens a new multi-part series with an interesting and remarkable backstory. It started out as a series of articles by Joel Orosz in our print journal, The Asylum. That's interesting to many of us and worth remarking on, but the really interesting part is the journey from there to here.
I initially suggested the idea to Joel as an excerpt from the articles, in order to give readers of The E-Sylum a taste of the great product they're missing out on if they're not members of the Numismatic Bibliomania Society. Joel was on board, but there was a parallel effort that would eventually combine with this one.
Newman Numismatic Portal Project Coordinator Len Augsburger had been experimenting with ChatGPT to summarize NNP content. He'd generated a summary of Q. David Bowers' book on dealer Abe Kosoff. He commented, "It's not the most trenchant analysis (there is no discussion of Kosoff's appraisal of the Lilly collection for example), but it is accurate and publication-ready. Really amazing." Here are the first few paragraphs:
"This volume by Q. David Bowers provides a comprehensive biography of Abe Kosoff, a prominent American numismatist who made significant contributions to the coin collecting community. The book is divided into eight chapters that cover Kosoff's life and career, including his upbringing, early career, and founding of several coin companies.
In the first chapter, Bowers provides an overview of Kosoff's family history and upbringing. Kosoff was born in Russia in 1913 and immigrated to the United States in 1918. Bowers describes Kosoff's early years in New York City and his love for collecting coins and stamps. He also discusses Kosoff's early career, which included working for several coin dealers and eventually founding his own company, Numismatic Gallery.
The second chapter focuses on Kosoff's founding of Numismatic Gallery in 1935. Bowers describes how Kosoff used his knowledge and connections in the coin collecting community to build a successful business. He also discusses some of the notable coins that Kosoff sold through Numismatic Gallery, including the 1913 Liberty Head nickel.
The third chapter covers Kosoff's role in the coin market during World War II. Bowers explains how Kosoff's business suffered during the war but how he continued to collect and sell coins to fellow collectors. He also describes how Kosoff's experience during the war led him to become involved in fundraising for Jewish causes."
I responded with a link to an amazing post about what can be done with these tools already (linked below). I took a few minutes to reflect and wrote this, which I later dreamed up a headline for.
I Was a Teenage AI Programmer
Long ago in my second job after college, I had a business card that read, "A I Programmer". Everyone would ask, "What's an A I Programmer?" I had to explain that "AI" stood for "Artificial Intelligence" and that we wrote computer code to do cool stuff so people don't have to. We built what were then called "expert systems" using rule-based algorithms, and got paid to play with language understanding and neural networks. We had some minor successes, but it was hard to live up to the hype surrounding the field. In hindsight, the algorithms were good but not yet good enough, and we didn't realize the role big data could play - and the era of big data was still decades away.
I stayed with the company as the "AI Winter" descended and we had to find contracts in other areas. Along the way I discovered and taught myself to use some early internet tools, long before browsers and the World Wide Web. One day I found some code on a server in Switzerland that would do some task I needed. I downloaded it, ran it and saved myself a month of work. Millions of people download apps every hour today, but then it was like magic. I grabbed our sales guy Steve. I'm sure he thought I was goofy, but I excitedly told him, "This will change the world."
I had the same feeling when I first saw ChatGPT. It's an orders-of-magnitude improvement over all that went before, and a milestone on the way to artificial general intelligence. Our company was an early casualty in the war, but in another hindsight revelation I see us now as well-paid cannon fodder, fighting for territory inch by inch, not knowing our fate but faithful that our cause was just. The future we dreamed of is now on the horizon - and it's scary and exciting as hell.
To put this event into perspective, consider this chart posted by Kyle Hailey on Twitter. It shows the number of days it took popular internet platforms to reach their first million users, and their first 100 million users. ChatGPT leapfrogged to 100 million in practically no time at all.
This is astonishing.
Could a team of humans have done a better job? Without question, across the board. The AI output is highly competent, but since, as I understand it, AI is essentially a super-duper autofill application (given this word, what would be the highest probability for the next word?), competence is to be expected. What AI output lacks is sparkle--there is nothing like wit, no jolt of unexpected insights, or twists of unforeseen interpretation.
But, when it comes to speed and output, AI beats us humans all hollow. In the example, it churned out in 30 minutes a competent set of eight documents and electronic outputs that would have taken a team of humans days, if not weeks, to do. You would have created a better website, Lianna would have created a better video, etc., but each individual component would have taken more like 30 hours than 30 minutes--and then there would have been the task of coordinating them so that they spoke consistently, accurately, and authoritatively.
For most businesses, the comparison between the two would result in an easy decision: do I want one guy plus AI to produce a suite of B-graded projects in 30 minutes, or do I want a team of people to produce a similar suite of A-graded projects in 30 days? The cost savings of the first option alone would make it the obvious choice. Even better, if you want A-graded work, would be to have one guy plus AI churn all of the B-graded projects out in 30 minutes, then assign a team to improve it to A-graded output. That would add time and costs, of course, but you would end up with more creative and better quality outputs at a fraction of the time and costs that would be required to build them from scratch in the ancient fashion. It's sort of like a machine that can build the basic framework of a house in 30 minutes, and turn a team loose to add embellishments and paint it.
So, this raises an interesting question. We have been discussing the possibility of my writing, for The E-Sylum, a concise summary of the seven "Associations" columns that I have written for The Asylum on Charles Ira Bushnell's Flandin / Groux / Howard compendium catalog. This seems to me like a job that AI could do in--well, 30 minutes, and that I could then edit to interject creative insight, and choose which of the photos should accompany it. Would this be a good numismatic test of the technology--or am I smoking dope?
Of course, I would need a basic tutorial on how to use such tech (I assume that I would have to feed it the seven columns, and then phrase the instructions so that the AI would summarize them within some parameters, but I haven't a clue how to make all this happen, and which AI I should/could use). What do you think?
I thought it was well worth a try, and Len offered to run Joel's articles through ChatGPT for summarization. Joel was on board, but wrote "My nagging fear is that the summary will come out much more creative and witty than the originals!"
It took a few tries to wrangle the logistics and Len's effort succeeded, although Joel needn't have worried about the machine's originality and wit. Here's an excerpt.
7 / THE CATALOG
The article is a review of Charles Ira Bushnell's Flandin's Catalogue of Coins and Medals. The book is a
collection of three rare numismatic sales from the mid-19th century, curated by Bushnell, and contains
detailed descriptions and prices of each item, as well as the names of the buyers. The article notes that
the book is unique and contains one-of-a-kind ephemera, including a bookplate, an ALS from Bushnell to
Flandin, and newspaper clippings. The author, Joel J. Orosz, considers the book to be his most prized
possession and an essential document for American numismatics.
7 / THE CREATOR
Charles Ira Bushnell was a New York City-based attorney who became an advanced coin collector in the
late 1840s. He actively gathered a vast collection of coins, medals, and tokens, although he kept a low
profile. He built a comprehensive library of catalogs and articles on American coinage and published An
Arrangement of Tradesmen's Cards, Political Tokens, Also Election Medals, Medalets & c. Current in the
United States of America, for the Last 60 Years in 1858, which became the first substantive work on
American tokens. He also wrote An Historical Account of the First Three Business Tokens Issued in the
City of New York in 1859. Bushnell was known for his extensive knowledge and was highly regarded in
the numismatic community. He participated in coin auctions, and his dedication to collecting led him to
become a leading figure in the hobby.
Still lengthy at seven pages, but we could split it into two or three E-Sylum articles depending on where natural breaks fall. With images that could work well. It summarizes at the writing level of a
10th-grader, but still impressive.
First, many thanks to you, Len, for your persistence and creativity in getting ChatGPT to do its job.
It is absolutely fascinating to see what it summarized and what it did not. It glommed onto a lot of detail, and left out a lot of connective tissue (lots of prices realized, nothing about the scandal that led to the creation of the only known portraits of Pierre Flandin, as one example).
The writing is, as you say, Wayne, at about the 10th grade level; it's more a recitation of facts than an essay; and occasional errors creep in; but overall, I am astonished by how much complex detail it got right. It would serve the purpose of summary quite well, and as you have noted, Wayne, may intrigue a few E-Sylum subscribers sufficiently to take a closer look at NBS membership.
So I am inclined to proceed as follows. First, I would edit the output lightly to correct half a dozen errors of fact. Second, I would write a couple of introductory paragraphs defining this as an experiment in using ChatGPT to create a summary of a series of articles in The Asylum, and then briefly sketch out the significance of the Bushnell compendium catalog. I would think that it would take three installments to do it all—I would choose 4-5 pictures to illustrate each installment.
And a big shout-out to you, Len, who can now add ChatGPT Meister to your many other accolades!
Well, the World Wide Web begat the job description of "Webmaster", but the term I'm seeing for today's Large Language Model Wranglers is "Prompt Engineer." Maybe it soon won't even be necessary to 'splain it to people.
And someday, maybe just a few years from now, someone will say, "You know, you young people don't know how good you have it. Back in my day, we had to write prompts by hand. Here's my old email signature that says 'Prompt Engineer.' Remember e-mail?"
For grins, I asked ChatGPT to "Please make me a poster for a movie called "I Was a Teenage Prompt Engineer", in the style of a 1950's horror film poster." Here's what it came back with, almost instantly:
Title: I Was a Teenage Prompt Engineer
Subtitle: They thought they had the perfect computer program...until it started to think for itself!
The poster features a menacing robot looming over a terrified teenager who is cowering in fear. The robot's eyes are glowing red, and its metal arms are reaching out menacingly towards the boy.
The background is a dark and ominous sky with lightning bolts striking in the distance. The title is written in bold, red letters that appear to be dripping with blood. The subtitle is written in smaller white letters beneath the title.
At the bottom of the poster, there are a few quotes from movie critics:
"A terrifying tale of technology gone awry!" - The New York Times
"Edge-of-your-seat suspense!" - Variety
"Don't miss this thrilling ride!" - Rolling Stone
Finally, the release date is listed at the bottom of the poster in bold, white letters: "Coming Soon to Theaters!"
ChatGPT itself only produces text, but that was marvelous. Another OpenAI product, Dall-E produces images. As you can see, that's got a ways to go yet. Maybe I need to hire a Prompt Engineer...
To read some articles on ChatGPT's power and promise, see:
Superhuman: What can AI do in 30 minutes?
Attention is All You Need
ChatGPT sets record for fastest-growing user base - analyst note
chatGPT : something different is happening.
To read the earlier E-Sylum article, see:
OH E-SYLUM, NUMISMATIC HIVE
Wayne Homren, Editor
The Numismatic Bibliomania Society is a non-profit organization
promoting numismatic literature. See our web site at coinbooks.org.
To submit items for publication in The E-Sylum, write to the Editor
at this address: firstname.lastname@example.org
To subscribe go to: https://my.binhost.com/lists/listinfo/esylum
Copyright © 1998 - 2023 The Numismatic Bibliomania Society (NBS)
All Rights Reserved.
NBS Home Page
Contact the NBS webmaster