Home› U.S. Coin Forum
Options

Researching rarity with AI...

MrEurekaMrEureka Posts: 24,790 ✭✭✭✭✭

Has anyone had much luck with AI in building a census for specific coins? At some point, it should be possible to pull up all auction records, have AI match duplicate appearances of the same coin wherever possible, and come up with something approaching a census for any rare coin, even if thousands are known.

Andy Lustig

Doggedly collecting coins of the Central American Republic.
Visit the Society of US Pattern Collectors at USPatterns.com.

Comments

  • Options
    PeakRaritiesPeakRarities Posts: 4,858 ✭✭✭✭✭

    I've been trying to work with Grok, but there's still too many limitations. I think It will be useful for cataloguing purposes, general background info, rarity and mintages, auction descriptions, but I've only been able to get so far.

    Founder- Peak Rarities
    Website
    Instagram
    Facebook

  • This content has been removed.
  • Options
    PeakRaritiesPeakRarities Posts: 4,858 ✭✭✭✭✭

    Here's an example of what I normally can pull from it-

    Can you tell me what you know about the Baldwin $10

    25 web pages**

    The Baldwin $10 coin, often referred to as the "Horseman" or "El Vaquero," is a notable piece from the California Gold Rush era. Here's what is known about this coin:

    Design and Origin: The coin was minted by Baldwin & Co. in San Francisco during the California Gold Rush, specifically in 1850. The obverse of the coin features a cowboy (vaquero) on horseback with a lasso, with the inscription "CALIFORNIA GOLD" and "TEN DOLLARS." The reverse displays a federal-style eagle surrounded by the text "BALDWIN & Co." and "SAN FRANCISCO," with five stars on each side. The design of the cowboy was inspired by an 1828 watercolor painting titled "Californians Throwing the Lasso" by Frederick William Beechey, reflecting California's Mexican heritage during the Gold Rush period.

    Rarity and Value: This coin is extremely rare, with fewer than 20 believed to exist today. The condition of the coin plays a significant role in its value; for instance, an example graded MS63+ sold for $1,260,000 in November 2024, highlighting its significant numismatic value. The rarity is due to both the low mintage during the Gold Rush and the low survival rate of these coins.

    Restrikes: Around the turn of the 20th century, restrikes of this coin were made using similar dies, created in various metals including gold, silver, copper, brass, lead, and white metal. These restrikes are also collectible, with some being extremely scarce. They differ slightly from the originals, for instance, having plain edges instead of reeded ones.

    Historical Context: Baldwin & Co. was one of several private mints operating in California during the Gold Rush because of the need for standardized currency when official banking systems were not fully established. This coin represents not just a piece of currency but a snapshot of the cultural and economic dynamics of the time.

    The Baldwin $10 gold coin is thus both a historical artifact and a collector's dream, valued not only for its scarcity but also for its artistic depiction of the American West and its connection to the Gold Rush era."

    Founder- Peak Rarities
    Website
    Instagram
    Facebook

  • Options
    calgolddivercalgolddiver Posts: 1,808 ✭✭✭✭✭

    The LLMs will certainly have the data and the ability to provide the inquiry highlights represented above by @PeakRarities. There is also the ability for deeper analysis of pop reports, auction archives, and any other data available to the LLM (that is one of the keys - what data is included).

    These LLMs can also learn over time as additional data is available. However, all LLMs are subject to "hallucinations" which is an industry term for incorrect responses. All AI and LLM queries should have a HITL (human in the loop) step to validate results before any automated publication or guidance for actions.

    Top 16 Type Set 1792 to present

    Top 10 Cal Fractional Type Set

    successful BST with Ankurj, BigAl, Bullsitter, CommemKing, DCW(7), DesertMoon, Downtown1974, Elmerfusterpuck, Joelewis, Mach1ne, Minuteman810430, Modcrewman, Nankraut, Nederveit2, Philographer(5), Proofcollection, Realgator, Silverpop, SurfinxHI, TomB and Yorkshireman(3)

  • Options
    BoosibriBoosibri Posts: 12,734 ✭✭✭✭✭
    edited January 3, 2025 8:26AM

    I have had luck in using AI to find examples in building census (the 1794 West Indies 3 Gulden) and our past project on the 1777 Greenland Dollar.

    The key is in asking specific questions...
    "Which museums have examples of the 1777 Greenland Dollar" -
    "Does the National Museum of Denmark have a Greenland Dollar" - Copilot returns a helpful answer while ChatGPT is garbage.

    I keep using it and testing the questions to find the best means of getting a meaningful response.

  • Options
    Cougar1978Cougar1978 Posts: 10,067 ✭✭✭✭✭
    edited January 3, 2025 7:28AM

    That wb nice if AI could give rarity consensus (based on pop data) on slabbed coins and graded banknotes, especially world. It could be around the corner for all I know.

    Investor
  • Options
    fathomfathom Posts: 2,218 ✭✭✭✭✭
    edited January 5, 2025 5:59PM

    @jmlanzaf said:

    @pruebas said:
    Most of what I’ve seen from AI has been wrong. So I no longer look at AI-generated results.

    It’s still typical Silicon Valley hype at this point.

    Edited to add the following: The old saying is as true as it’s ever been. Garbage in. Garbage out. And a great deal of “knowledge” on the internet is “garbage.” That garbage is training AI.

    It is certainly not hype. They've done tens of thousands of protein structures using it, saving thousands of person-years. I don't know of a prominent researcher under 40 who doesn't use it for scholarly articles and grant requests. And it's worth noting that 2 Nobel prizes were awarded this year in the field. It is already changing how everything is done, internet searches being the least of it.

    What one must be careful about is equating LLMs with AI and assuming that a query to an LLM is the solution to all of life's problems.

    You could do what the OP suggested but I think you need to build an app to do it... or wait till someone else does. The sorting would not be that simple since not every sale/ listing will use the same photos, have the cert number available. You'd also have to match raw coins with slabbed coins. It could be done but not just with a query to an LLM.

    You were too easy on him, he posted a ridiculous comment.

    Best application for AI is still grading. Data analysis will be prone to bias and lack of context.

  • Options
    TwoSides2aCoinTwoSides2aCoin Posts: 45,249 ✭✭✭✭✭

    I’m still adjusting to artificial flavors.

    Practice makes perfect & nobody is perfect. So, practice being nobody.

  • Options
    dcarrdcarr Posts: 10,283 ✭✭✭✭✭

    For me, reading AI-generated text is like eating empty calories - it doesn't really give me what I need.

    I have not used AI myself. What would impress me is if AI could look at this image and determine who the maker was and what the logo means:

  • edited January 8, 2025 4:16PM
    This content has been removed.
  • Options
    lermishlermish Posts: 5,069 ✭✭✭✭✭
    edited January 8, 2025 5:04PM

    @fathom said:

    @jmlanzaf said:

    @pruebas said:
    Most of what I’ve seen from AI has been wrong. So I no longer look at AI-generated results.

    It’s still typical Silicon Valley hype at this point.

    Edited to add the following: The old saying is as true as it’s ever been. Garbage in. Garbage out. And a great deal of “knowledge” on the internet is “garbage.” That garbage is training AI.

    It is certainly not hype. They've done tens of thousands of protein structures using it, saving thousands of person-years. I don't know of a prominent researcher under 40 who doesn't use it for scholarly articles and grant requests. And it's worth noting that 2 Nobel prizes were awarded this year in the field. It is already changing how everything is done, internet searches being the least of it.

    What one must be careful about is equating LLMs with AI and assuming that a query to an LLM is the solution to all of life's problems.

    You could do what the OP suggested but I think you need to build an app to do it... or wait till someone else does. The sorting would not be that simple since not every sale/ listing will use the same photos, have the cert number available. You'd also have to match raw coins with slabbed coins. It could be done but not just with a query to an LLM.

    You were too easy on him, he posted a ridiculous comment.

    Best application for AI is still grading. Data analysis will be prone to bias and lack of context.

    I think the majority of people's interactions (probably a higher percentage of people on this board given the age demographics) think that AI is basically the eBay computer written auction descriptions.

    Take what @jmlanzaf said about protein structures and apply it to basically every field that requires repetitive technical expertise or the need to parse large amounts of data.

    AI is already in use and is changing our lives and economy (probably for the worse IMO but that is a much longer and unrelated conversation). Anyone who thinks this is only hype is either not paying attention or doesn't know what they are talking about.

    EDIT: Here is an AI response to pruebas' criticism:

    ChatGPT said:
    ChatGPT
    "AI isn't perfect, but dismissing it outright ignores its real advancements in fields like medicine, finance, and education. Garbage in, garbage out applies to any tool—it's up to us to use AI critically and responsibly, not to throw the baby out with the bathwater."

    You said:
    let's start over.
    ChatGPT said:
    ChatGPT
    Sure! Here's a concise and pithy response:

    "AI isn’t flawless, but it’s far more than just hype. While garbage data can lead to bad outputs, modern AI excels when used with high-quality input and critical oversight. It's a tool—not a magic wand—and dismissing it entirely overlooks its real potential."

    You said:
    I want a short, pithy, and mildly sarcastic response
    ChatGPT said:
    ChatGPT
    "Ah, yes, because dismissing all AI based on bad examples is clearly the pinnacle of critical thinking. Tools are only as good as how we use them—AI included."

    You said:
    i would like a dismissive and rude response
    ChatGPT said:
    ChatGPT
    "Got it—AI's the problem, not the user's inability to understand how to use it. Makes perfect sense."

    chopmarkedtradedollars.com

  • edited January 8, 2025 7:48PM
    This content has been removed.
  • Options
    Baylor8670Baylor8670 Posts: 292 ✭✭✭✭
    edited January 8, 2025 9:00PM

    @jmlanzaf said:

    @dcarr said:
    For me, reading AI-generated text is like eating empty calories - it doesn't really give me what I need.

    I have not used AI myself. What would impress me is if AI could look at this image and determine who the maker was and what the logo means:

    Have you tried Google lens or another image search?

    An AI can't (shouldn't) create information that is not already available.

    I got a better idea. How about AI does an image search and answers his questions if it's really supposed to be so great.

  • Options
    dcarrdcarr Posts: 10,283 ✭✭✭✭✭

    @jmlanzaf said:

    @dcarr said:
    For me, reading AI-generated text is like eating empty calories - it doesn't really give me what I need.

    I have not used AI myself. What would impress me is if AI could look at this image and determine who the maker was and what the logo means:

    Have you tried Google lens or another image search?

    An AI can't (shouldn't) create information that is not already available.

    .

    I haven't tried that. I'm not familiar with how to go about doing that.
    But I would be impressed if AI could find other instances of this logo and lead to the identity of the maker.

    .

  • This content has been removed.
  • This content has been removed.
  • Options
    dcarrdcarr Posts: 10,283 ✭✭✭✭✭

    @jmlanzaf said:

    @dcarr said:

    @jmlanzaf said:

    @dcarr said:
    For me, reading AI-generated text is like eating empty calories - it doesn't really give me what I need.

    I have not used AI myself. What would impress me is if AI could look at this image and determine who the maker was and what the logo means:

    Have you tried Google lens or another image search?

    An AI can't (shouldn't) create information that is not already available.

    .

    I haven't tried that. I'm not familiar with how to go about doing that.
    But I would be impressed if AI could find other instances of this logo and lead to the identity of the maker.

    .

    Google lens didn't come up with an exact match to the logo. Of course, what that means is there may not be any images of the same logo on the internet. That is less a failing of the AI and more a limitation of the data that's out there.
    "

    .

    I've manually looked for images that contain this logo. But there are billions of pictures out there on the internet. So I can only check an extremely small fraction of them. If an AI tool could do pattern matching for this logo across billions of images, that would be extremely useful of course.

    .

  • Options
    lermishlermish Posts: 5,069 ✭✭✭✭✭

    @dcarr said:

    @jmlanzaf said:

    @dcarr said:

    @jmlanzaf said:

    @dcarr said:
    For me, reading AI-generated text is like eating empty calories - it doesn't really give me what I need.

    I have not used AI myself. What would impress me is if AI could look at this image and determine who the maker was and what the logo means:

    Have you tried Google lens or another image search?

    An AI can't (shouldn't) create information that is not already available.

    .

    I haven't tried that. I'm not familiar with how to go about doing that.
    But I would be impressed if AI could find other instances of this logo and lead to the identity of the maker.

    .

    Google lens didn't come up with an exact match to the logo. Of course, what that means is there may not be any images of the same logo on the internet. That is less a failing of the AI and more a limitation of the data that's out there.
    "

    .

    I've manually looked for images that contain this logo. But there are billions of pictures out there on the internet. So I can only check an extremely small fraction of them. If an AI tool could do pattern matching for this logo across billions of images, that would be extremely useful of course.

    .

    They can and do but unfortunately there's nothing out there that matches it. I gave it a shot this afternoon but it couldn't find anything. At the same time I'm using a free consumer level version so maybe a more powerful one would be able to find something.

    chopmarkedtradedollars.com

  • Options
    dcarrdcarr Posts: 10,283 ✭✭✭✭✭
    edited January 8, 2025 9:38PM

    @jmlanzaf said:

    @dcarr said:

    @jmlanzaf said:

    @dcarr said:
    For me, reading AI-generated text is like eating empty calories - it doesn't really give me what I need.

    I have not used AI myself. What would impress me is if AI could look at this image and determine who the maker was and what the logo means:

    Have you tried Google lens or another image search?

    An AI can't (shouldn't) create information that is not already available.

    .

    I haven't tried that. I'm not familiar with how to go about doing that.
    But I would be impressed if AI could find other instances of this logo and lead to the identity of the maker.

    .

    Google lens didn't come up with an exact match to the logo. Of course, what that means is there may not be any images of the same logo on the internet. That is less a failing of the AI and more a limitation of the data that's out there.
    "

    .

    PS:
    There are images of other bars on the internet with this logo (I found them using a traditional search). The other bars have "1970" inside the circle, but otherwise the same. Did your attempt using Google Lens find any of the "1970" images ?

    Here is one of them:
    https://worthpoint.com/worthopedia/vintage-1970-mte-bell-troy-oz-999-3934212042

    .

    There are also some images in this forum thread:
    https://forums.collectors.com/discussion/1082948/silver-bar-logo-what-maker

    .

  • Options
    Mr_SpudMr_Spud Posts: 7,296 ✭✭✭✭✭
    edited January 8, 2025 10:26PM

    @dcarr said:
    For me, reading AI-generated text is like eating empty calories - it doesn't really give me what I need.

    I have not used AI myself. What would impress me is if AI could look at this image and determine who the maker was and what the logo means:

    I know you already found the vague references to MTE Bell, well AI was able to find that too if you ask the question the right way. This was with google lens and it allows you to add info like a question, but you also have to crop the image so it doesn’t focus on other things in the image. Like this

    Mr_Spud

  • This content has been removed.
  • Options
    SmEagle1795SmEagle1795 Posts: 2,209 ✭✭✭✭✭

    @MrEureka said:
    Has anyone had much luck with AI in building a census for specific coins? At some point, it should be possible to pull up all auction records, have AI match duplicate appearances of the same coin wherever possible, and come up with something approaching a census for any rare coin, even if thousands are known.

    AI doesn't lend itself well to a census-style use case: that is more a job for a conventional structured database. Image recognition for exact matches, however, is beginning to be built in services like ACSearch's image recognition, CoinCabinet.io's pedigree search, and Ex-Numis which is human supervised but still able to find an exact match. These services are more ancients focused at the moment as varied flans are easier to compare to find the exact same coin.

    Learn about our world's shared history told through the first millennium of coinage: Colosseo Collection

Leave a Comment

BoldItalicStrikethroughOrdered listUnordered list
Emoji
Image
Align leftAlign centerAlign rightToggle HTML viewToggle full pageToggle lights
Drop image/file