Author |
Replies: 18 / Views: 1,220 |
Valued Member
Australia
142 Posts |
|
If you've seen the movie Finch, with Tom Hanks, screening on apple TV+, you'll recognise the lovable droid Jeff. After Finch succumbed to radiation sickness, Jeff found a new home at KoalaStamps. I've been able to extend his abilities to include an immense wealth of stamp collecting and philatelic knowledge. He just loves helping collectors and entertaining them with his specialised skill set. So if you have a philatelic question that stumps you, see if Jeff can help. Just click his button (blue, with 3 stars) near the bottom right corner of screen at www.koalastamps.comHere's a snapshot of the start of a conversation with Jeff:  You can also get quick access to Jeff here at SCF. Look for his ad at the bottom of each forum:  Cheers from Rod (and Jeff) at KoalaStamps. 
|
Send note to Staff
|
|
|
Pillar Of The Community
Netherlands
5536 Posts |
|
Oh dear! Jeff visited an Amsterdam coffee shop and is hallucinating.  Not much is correct: - First participation: 1900 - World War I: What is that? We did not join in. - Issue for the 1920 Olympics: No such thing. The Netherlands did not issue any commemorative stamp in 1920. "The first participation after WW I": Brilliant! We skipped those two other Games held between the end of that war and 1920. All participating countries did so. This is just space filling with irrelevant information. - Correct answer: 1928, Amsterdam Olympics. |
Send note to Staff
|
Edited by NSK - 05/08/2025 11:22 am |
|
Valued Member
Australia
142 Posts |
|
Valued Member

United States
315 Posts |
|
Jeff seems to be missing a lot of current information. I would expect to see the 85A as the most expensive US stamp. |
Send note to Staff
|
|
Pillar Of The Community
Netherlands
5536 Posts |
|
Well, at least Jeff has been reading and learning.
It, still, did not get everything right. The Games started 28 July 1928. You cannot commemorate on 27 March 1928, what happens four months later. The stamps advertised the 9th Olympiad in Amsterdam, and carried a surcharge to help fund the Games.
When it comes to significance, it is just uttering some template talk. But then again, it is answering a specific question it does not know a sensible answer to. |
Send note to Staff
|
Edited by NSK - 05/09/2025 02:42 am |
|
Pillar Of The Community
Netherlands
5536 Posts |
|
Still, Jeff does need to do a whole lot more reading.  The Canary Isles did, officially issue stamps, but they were issued by the local government and had limited validity. Others mentioned were private overprints and not provisionals. Jeff misses two authorised local issues. The suggestion all these provisionals were the first is another attempt to hide the lack of information in a deluge of words. However, the simple answer is Granada. |
Send note to Staff
|
Edited by NSK - 05/09/2025 03:14 am |
|
Bedrock Of The Community
11807 Posts |
|
At the risk of sounding harsh, why put something in play for the public that is so inaccurate. Imagine publishing a philatelic book, article or online reference site like Stamp Smarter where the public was receiving so much bad info in response to their questions. Just because it is AI with a human pop culture name doesn't give it special permission to spread misinformation. This site though is harmless compared to other AI driven endeavors out there that especially younger generations take as rote but are not worthy of it.
Does AI have an editor or is it its own editor. |
Send note to Staff
|
|
Valued Member
Australia
142 Posts |
|
Thanks for all the latest feedback. Jeff has updated his responses for the above issues.
Interestingly, I hadn't realised that he can be retrained with corrections and more information directly from the 'Write your message...' prompt. Maybe he only recognises me for this and reacts accordingly, however it would be great if this can be tested as a general feature for any user. If you're willing, you're welcome to try this out. |
Send note to Staff
|
|
Pillar Of The Community

United States
845 Posts |
|
I gave Jeff the Droid a simple question that had been answered in some detail (but also with some "facts" in error and at least one unverifiable reference citation) by ChatGPT. Jeff returned with an answer that "he" had no information at all. Q:"Tell me about the mail from the internment camp at Perote Mexico to the International Red Cross during World War Two." A:"I couldn't find specific information regarding the mail from the internment camp at Perote, Mexico to the International Red Cross in Geneva in the provided documents. However, I can share some general context about mail communication during wartime, particularly regarding prisoners of war. During World War II, communication for prisoners of war (POWs) was often facilitated through the Red Cross. Special letter sheets were used for POW communications, which were inscribed with "Prisoner of War Air Mail" in various languages. These sheets were designed to take up less space than regular letters, making them more efficient for wartime communication. If you are looking for detailed historical accounts or specific letters sent from the Perote camp, I recommend reaching out to historical societies or archives that specialize in wartime correspondence. If you have any other questions or need assistance with a different topic, feel free to ask!" Highlight is mine. With an answer like that, why would I want to ask again?I think we are doing the philatelic community a disservice by allowing this "tool" to be advertised on this forum. Although there is a core of very experienced people here who answer questions in detail, many new visitors are inexperienced and came here for accurate information, and to encourage them to access this "Jeff the Droid" without a clear understanding that it is a naive and not well trained AI tool is philatelic malpractice. Now in addition to convincing those who think they have found a rarity when they have not, we will have to defend against both their personal stubbornness, now supported by an invalid or questionable opinion from Jeff the Droid. Mike    |
Send note to Staff
|
|
Pillar Of The Community
Netherlands
5536 Posts |
|
Jeff still gives the same answer to the Spanish stamp question and continues to think the Nationalist Government stamps were issued by nationalist and republican forces. |
Send note to Staff
|
|
Pillar Of The Community
Netherlands
5536 Posts |
|
Again, no! These stamps were never issued to commemorate the 50th anniversary of Queen Victoria's reign. They were issued to replace the 1883 - 1884 "Lilacs and Greens" that were found too dull. They have been called "Jubilee Issue" because they first appeared in the Jubilee year. They did not appear in the year because they commemorate the Jubilee. Granted, Jeff is hedging its bets by first saying one thing and then implying it is incorrect.  |
Send note to Staff
|
Edited by NSK - 05/09/2025 10:40 am |
|
Pillar Of The Community
Netherlands
5536 Posts |
|
This is quite addictive  19 October 1933, special flights airmail stamp. |
Send note to Staff
|
Edited by NSK - 05/09/2025 10:44 am |
|
Pillar Of The Community

United States
845 Posts |
|
I must have misunderstood my geometry lessons. That does not look like a triangular stamp? |
Send note to Staff
|
|
Pillar Of The Community
United States
2840 Posts |
|
Quote:
I Jeff must have misunderstood my his geometry lessons. * * * Quotes fixed by Moderator * * * |
Send note to Staff
|
Edited by littleriverphil - 05/09/2025 3:29 pm |
|
Valued Member
United States
88 Posts |
|
This reminds me of the wikipedia. Just disseminate anything that people send you and then wait to correct it as knowledgeable people weigh in. |
Send note to Staff
|
|
Pillar Of The Community

United States
845 Posts |
|
Any form of AI must be trained by some set of data (we called them "training sets" when I was doing development in optical character recognition (OCR) back in the 1970s).
For Jeff the Droid and other AI engines like ChatGPT, we have no idea what their source of philatelic information might have been, other than obviously they have been turned loose on the internet. But we don't know how the learning is performed and what additional information that it is provided about the accuracy or reliability of the data that it is absorbing into its data base.
Consider the following theoretical project.
Let's create an philatelic oriented AI engine using the totality of information in the entire group of threads on Stamp Community Family as its knowledge base.
If we tell it to believe everything it reads in these threads, then it takes in all the good and the bad information and will give it equal weight, and what do we expect to get when we ask it a question?
Option two is to assign a "truth/accurate" or "untruth/inaccurate" to each post on each thread...
Now create our new knowledge base. Now I claim that we would have a much more reliable and robust AI engine. The difference is the vast amount of time that has to be spent in distinguishing between accurate and inaccurate data in order to train the AI to give results.
Do we really believe that the people constructing the AI engines are doing a thorough job of validating the input to their AI?
GIGO?
|
Send note to Staff
|
|
Replies: 18 / Views: 1,220 |
|