Anyone tried ChatGPT3

Question

Has anyone tried ChatGPT3 at OpenAI.com?

I asked it to write a bio for an ancestor, but it crashed. I suspect that some python skills are needed to "embed" the WikiTree query API and link to the Chat GPT question  "Completion API".

Of course, I think we should ask if this is a good idea.

On the other hand, Chat GPT made a very good attempt at explaining Quantum Loop Gravity.

Chris

Comments

See my comment here.

I think ChatGPT would often produce incorrect results for genealogy, all the more dangerous for looking plausible. This is because it will know nothing about particular cases we are interested in, apart from what we explicitly give at the time; but it has a vast array of stored information, including biographies, which it reuses without regard to meaning but only to similarity of wording. So it could easily insert material filched from someone else's biography into one you asked it to write.

Answer 1

I think that I am now happy with how to use ChatGPT to ask it to produce text limited to only the info that I give it. Using the WikiTree API, it produced this bio of Little-4639, in French. for a change:

"William Little était un homme né le 9 juillet 1676 à Cumberland, en Angleterre. Il est décédé le 27 mars 1763 à Dormansteads, Stapleton, Cumberland, en Angleterre. Il avait un fils et était le fils d'Edward Little."

This issue has now turned into one of how best to use the WikiTree API with curl, https POST, etc.

Thanks for the responses and discussion.

Answer 2

No, but I just created an account and used it to write a reference note about “night riders” in Missouri. I changed it up a bit and made it more accurate and relevant to the profile, but am kind of excited about the possibilities of using it to help write better biographies.

Answer 3

I've tried it, but I think it's pretty useless for generating a genealogical profile. ChatGPT just generates text based on any relevant material that it has consumed/been trained on. In the case of most ancestors, it doesn't seem to know anything about the person and will either say the info is unknown or it will make up something wrong. As an example, I entered "Write a genealogical profile of Edward Brewer, who emigrated to Massachusetts in 1632 aboard the Lyon" and this is what it produced:

Edward Brewer was an early settler of Massachusetts who emigrated from England in 1632 aboard the ship Lyon. He was born around the year 1600 and lived for approximately 60 years. Edward was married and had several children, although the exact names and number of descendants are not well documented. He was a farmer and owned land in the Massachusetts Bay Colony, where he was a member of the Puritan community. Edward was involved in the political and religious life of the colony and held several civic positions. His legacy continues through his descendants who still reside in New England. Despite limited records from the 1600s, Edward Brewer is remembered as one of the early settlers of the Massachusetts Bay Colony and a founding member of the Puritan community in New England.

In fact, according to "Great Migration," Edward Brewer was born in 1602 not 1600, he died at the age of 84 not 60, and the names of his children are well documented.

Comments

I don't know sorry, Chris. Embeddings for ChatGPT are described here, here, and here, but I don't understand what those pages mean, except that it costs money. In particular, I don't see where it says how to "embed" your own information. WikiTree holds many gigabytes of information. How would we embed all of that? Or would we do it only for one family line at a time?

---

Hi Jim,

On the current scale of things on the Web, I don’t think WikiTree’s gigabytes are very big.

The OpenAI embedding is not really embedding, but an API, but I hadn’t spotted the price lists!

I think that I need to do some more homework to see if ChatGPT can be made reliable for genealogy.

---

Chris, I still don't know what "embedding" means, API or not. Is there a way we could upload or transfer the many gigabytes of WikiTree public data to ChatGPT and then keep it up to date as profiles are improved? Could we even do just a few generations of a single family?

---

Jim, the WikiTree database exposes its data, read-only, via this URL: https://api.wikitree.com/api.php . See the example pages at: https://github.com/wikitree/wikitree-api

I posted this URL into my browser: https://api.wikitree.com/api.php?action=getProfile&key=Little-4639

I then saw this page, in JSON format, convenient for computer software and some humans, starting: [{"page_name":"Little-4639","profile":{"Id":12301674,"Name":"Little-4639","FirstName":"William","MiddleName":"","LastNameAtBirth":"Little","LastNameCurrent":"Little","Nicknames":"","LastNameOther":"Litle","RealName":"William","Prefix":"","Suffix":"","BirthLocation":"Cumberland, England","DeathLocation":"Dormansteads, Stapleton, Cumbe…. Etc.

I then asked ChatGPT to create a bio based on that JSON which I pasted in to the message.

ChatGPT wrote:

“William Little was born on July 9th, 1676 in Cumberland, England. He was the son of Edward Litle. He lived in Dormansteads, Stapleton, Cumberland, England, where he passed away on March 27th, 1763."

HTH , Chris

PS hopefully the JSON and URLs will display correctly

---

Interesting, Chris, though copy and paste would need to be automated somehow. Here's a hopefully more readable format for some of the JSON data.

"Name": "Little-4639",
"FirstName": "William",
"MiddleName": "",
"LastNameAtBirth": "Little",
"LastNameCurrent": "Little",
"Nicknames": "",
"LastNameOther": "Litle",
"RealName": "William",
"Prefix": "",
"Suffix": "",
"BirthLocation": "Cumberland, England",
"DeathLocation": "Dormansteads, Stapleton, Cumberland, England.",
"Gender": "Male",
"BirthDate": "1676-07-09",
"DeathDate": "1763-03-27", 

ChatGPT gives a passable result in this case, except it has assumed William lived in the same place where he died, which may not be justified.

However, "artificial intelligence" is not needed for this. It's not hard to write algorithmic code which produces a biography from the same data. (Ian Beacall's Auto Bio uses such an approach, and I have my own private code to do this.) As distinct from a language model like ChatGPT, an algorithm will give a predictable result, without risk of inventing things.

Because much information, e.g. from censuses, is not in WikiTree data fields, it can't be made available in this way. It can be extracted from another database (I do that) or an existing biography (I think Ian does that in some cases), but the latter makes the process circular. ChatGPT could use events templates, but so can algorithms, and anyhow those have not been authorised.

I think "AI" would be massive overkill for something that can be done algorithmically.

---

Actually. ChatGPT just used what was in WT for birth and death locations.

I agree it is overkill just to get  a pleasing variation in English.

Whether we get similarly acceptable results for deeper concatenation queries allowed by the WikiTree API needs to be investigated. I think it could be a way of short-cutting algorithm building.

---

Chris, you can get a little more information with this query:

https://api.wikitree.com/api.php?action=getRelatives&keys=Little-4639&getParents=1&getSiblings=1&getSpouses=1&getChildren=1

---

Thank you Ian,

Have you thought about how to do ‘deeper' queries? By that I mean having retrieved nearest relatives, as you suggest, then creating another query based on the first results?

One hurdle is this switching between Profile IDs and page numbers in the WT API responses.

And we want as light processing as possible.

---

It's easy enough to add another (follow-up) query (with JavaScript), but it might be possible to get what you want in one query. The API is very good. What other query do you want?

Edit: You can also get the current bio text through the API, so you could give that to ChatGPT, too. Some biographies are really quite poor in terms of the prose. ChatGPT's English is very good.

---

Hi Ian, 

Yes I saw the variety of 'actions' in the WT API. And JavaScript should be straightforward. I am not sure I need to ask anymore for simple text creation from profiles. I suppose I am now thinking how could we get ChatGPT (or Bard, Lamda, etc) to assess whether a WT 'fact' is justified by the sources quoted in a profile.

Answer 4

I tested chat GPT vs Perplexity on a question on my 5th great grandfather as just a fun test  - and the latter vastly outperformed the former 

Who was Dr. Henry Hunt of Dublin

(1707-1796)

CHATGPT - wrong on anything and didn’t honour even basics like dates 

Dr. Henry Hunt was an Irish physician who served as a surgeon in the Irish Army during the French and Indian War. In 1759, he was elected to the Royal College of Physicians in Ireland, and he served as its president in

1779. Hunt was also a noted scientist, and he is credited with the discovery of the anaesthetic agent nitrous oxide. He died in 1896. 

————

Perplexity blew me away - even recognizing my well researched correction of his wife from Burke’s Landed Gentry of Ireland 1912 - of which considerable residual remains on the internet and absolutely correct on everything - even the fun fact of qualifications. It seemed better oriented for the geography and the date range supplied 

Dr. Henry Hunt (1707-1796 was the first Governor of Apothecaries' Hall in Dublin, Ireland (2]. He was born at Curraghchase Forest Park, Toreen, County Limerick (3] and married Elizabeth Kiernan in 1742(4). In 1784, he was appointed the first State Apothecary of Ireland (2][11.

In order to become Governor of Apothecaries’ Hall, Dr. Hunt had to possess certain qualifications as outlined in the Act[2]. These qualifications included being a member of the Worshipful Company of Apothecaries and having a Certificate of due qualification under their hands[5].

Comments

Another example of ChatGPT making things up. The discovery of nitrous oxide is credited to  Joseph Priestley, and Humphrey Davy first demonstrated the "laughing gas" effect.

I hadn't heard of perplexity.ai before, but I wouldn't trust it to be reliable either. There's a disclaimer at the bottom of the page: "Accuracy is limited by search results and AI capabilities."

If you ask it a question like Chase's, "Who was Edward Brewer, who emigrated to Massachusetts in 1632 aboard the Lyon?", it says:

Edward Brewer was a passenger on the ship Lyon, which sailed from London in 1632 and arrived in Boston in 1632^[1][2][3][4]. He was born circa 1600^[5] and settled at Roxbury, Massachusetts where he died in 1646^[5].

The sources include geni.com and Rootsweb. That's not as wildly speculative as ChatGPT, but it's still wrong about death date. Chase, is Roxbury correct?

---

Interesting. Daniel Brewer (abt. 1596 - 1646) travelled on the Lyon and lived in Roxbury. Perplexity is conflating Edward and Daniel.

What are your own sources for Edward Brewer, Chase? I can't find a profile.

Edited to add: If Chase used the following source, it seems unlikely that ChatGPT would have access inside the Ancestry paywall.

Ancestry.com. New England, The Great Migration and The Great Migration Begins, 1620-1635 [database on-line]. Provo, UT, USA: Ancestry.com Operations, Inc., 2013.

Original data: Anderson, Robert Charles. The Great Migration Begins: Immigrants to New England, 1620-1633, Volumes 1-3; The Great Migration: Immigrants to New England, 1634-1635, Volumes 1-6. Boston: New England Historical and Genealogical Society, 1996-2011.

---

Right! - so I was impressed in Perplexity is that it derived the final info from the Act of Parliament 1791

Answer 5

It is a tool like anything else. It is not always right, and it can ad lib (subjective statements). On the other hand, it can be a tool for ideals to explore and write about your family. For example, ask it what happened between 1848 and 1890 in Arkansas that would be significant to the people during that period or now has historic value. It can give you topics to research or explore.