Question Do you want OpenAI or ChatGPT analyzing your site?

Do you want OpenAI or ChatGPT analyzing your site?

I would guess that some of us do and some don't.

For the latter:  platform.openai.com/docs/gptbot

Yes and no. I will try to use a personal LLM trained with the data in my tree. But because my tree contains information about still living persons it would be a bad idea (and impossible) to use ChatGPT.

My dream is to ask my personal genealogical assistant:

1. When married Uncle George my Aunt Magdalena?
2. Who is the oldest person in my tree?
3. Tell me all the children of Grandfather Peter.
4. ...

I'm using AI for my genealogical research in many ways (see zenodo.org/records/10844181 ), but the intelligent personal genealogical assistant is still a dream. Maybe the time will be ready for that at the end of this year or next year.

Thanks, but no thanks. I'll pass on the hype. Ten years ago it was blockchain, now it's LLM.

I can get an answer to most (if not all) those questions from the Statistics page.

What LLMs are lousy at, is doing calculations and logic puzzles. Which is understandable, since it's a language model.

webtrees easily answers questions like that.

For questions about details not yet in my database, things like openAI might be very useful, except for the fact that their answers will be as reliable as the data used to "train" them—which might well include the plethora of genealogical garbage that is floating around on the web—including on ancestry.com, familysearch.org, and wikitree.

I've got good answers on some non-genealogical questions. I've also seen "answers" that were the opposite of facts I was able to find myself easily.

On one question, perplexity.ai gave a good answer and cited five sources.  Four of them had NO CONNECTION to the question, and the answer was merely a plagiarizing summary of the fifth one. So its usefulness was not in the answer, but in pointing me to a decent source.  DuckDuckGo can usually do that!

In my view, employing AI as a web search engine is not advisable. AI can be utilized to transcribe ancient documents, translate them from Latin to a contemporary language, summarize the content, or compile a list of individuals mentioned in the document and create a GEDCOM file for import into webtrees. This approach is exponentially more efficient than current methods.

Prüfung der eigenen Genealogiedaten durch KI? Sicher nicht! Wenn KI die Datenbankinhalte in die Hände bekommt, dann sind sie in diese jeweiligen Systeme übertragen. Es gibt keinen Weg zurück. Was das Abgreifen von Daten anbetrifft, verhält sich KI möglicherweise radikaler als MS, Google und so weiter. Woher sollte KI sonst Genealogiedaten bekommen. Wäre KI das was sie verspricht, könnte man Archivrecherchen an sie delegieren. Das geht aber nicht, denn KI ist kein Supergehirn, vielmehr handelt es sich um eine Datenbankstruktur, deren Wissen man mit geeigneten programmtypischen Abfragen entlocken muss.

I think the original post was about  Disallowing GPTBot  to deter AI from using your site to "train".
But it opens up a very interesting topic.
For me I'd like to use AI to transcribe old handwritten documents. 
I've just now had my first look at Transkribus  and that looks promising.
 

@JGerardi: Wenn Ihre Daten durch ein Passwort geschützt sind, kommt da keine KI ran. Weder für das Training noch bei einer Suche im Web. Die KI hat die gleichen Rechte wie ein normaler webtrees „Besucher“. Darüber hinaus blockt webtrees alle Zugriffe von KIs, die sich korrekt melden, ähnlich wie bei unliebsamen Webcrawlern.
Wenn die Archive es unterstützen, kann man eine KI sehr wohl zu Recherche nutzen, siehe www.compgen.de/2023/07/chatgpt-kann-auf-...-archives-zugreifen/
So etwas nutzt auch FamilySearch um aus Quellen automatisiert riesige Stammbäume zu erzeugen.
Und eine KI ist keine Datenbankstruktur; Sie verwechseln das mit dem Index einer Suchmaschine.

hermann wrote:

So etwas nutzt auch FamilySearch um aus Quellen automatisiert riesige Stammbäume zu erzeugen.
 

Which will undoubtedly make their "world family tree" even worse than it already is.

Which will undoubtedly make their "world family tree" even worse than it already is.

Yes, the FamilySearch FamilyTree is as good as the users made it. Some parts are excellent, in other parts there is potential to improve the quality.

The automatically generated trees of FamilySearch are not connected to that world tree (see my blog post  www.compgen.de/2023/06/ki-generierte-sta...me-bei-familysearch/ ). As a user, you can prove all the sources and manually copy the information you have checked to the world tree if you like it.

hermann wrote:

The automatically generated trees of FamilySearch are not connected to that world tree (see my blog post  www.compgen.de/2023/06/ki-generierte-sta...me-bei-familysearch/ ). As a user, you can prove all the sources and manually copy the information you have checked to the world tree if you like it.

So that is how they generate the list of suggested additional sources on the RHS of each individual in the world tree, and presumably Ancestry uses some similar technology to generate the list of related documents they display below each source.

Quite frankly, those suggestions (along with warnings of potential duplicate individuals) have become pretty good -- much more reliable than the ones suggested by Ancestry, which Wes loves to hate.... Starting about 1 or 2 years ago, it has become more than worth my while to consult the world tree when trying to find information about some "lost" individual. And if I find something, I always make a point of leaving the information in better shape than I found it.

drblam wrote:

.... Starting about 1 or 2 years ago, it has become more than worth my while to consult the world tree when trying to find information about some "lost" individual. And if I find something, I always make a point of leaving the information in better shape than I found it.

I also look at WFT frequently, hoping someone has attached some source I haven't found.  Occasionally, that is the case.  But far more often, none of the sources attached prove anything about the person they are attached to.