Комментарии:
Thank you so much Amy. I want to do this with my ancestors as children for my grandchildren. I didn't know where to begin so now I have an idea.
ОтветитьGoogle isn't the only and arguably not even the best Internet search engine. Just sayin'.
Ответитьas for the "editorializing" Chat GPT is horrible for that... You can, however, tell it to keep it's tone/style NEUTRAL and OBJECTIVE and to cut out all the "flowery" language.
Ответитьone thing I use Chat GPT for is editing and grammar correction. I have also used it to help me start biographical sketches.. For instance, I wanted to write a bio of my mum, so I cut and pasted text from other obituaries and news articles about my mum's work etc (like there was an article about her in the 1950s being a life guard, and it had some interesting details about her family and other sports activiites) I threw all that into Chat GPT and asked it to compile a biography. Now you have to be specific, tell it what style you want, how you want the text to flow, etc.. And with Chat GPT you have to review what it writes, very careful... It does tend to leave out things, sometimes construes facts in a piece of text in a wierd fashion.. etc.. SO ALWAYS REVIEW. And you can prompt Chat GPT to redo it, change the style, include more facts etc.. It is a good way to get started on Biographies especially if you are having trouble getting started. I plan on throwing my grandfathers typed bio into it and see if Chat GPT can "Clean it up" a bit.
ОтветитьI’ve been trying to get more into my genealogy recently because one of my aunt did and brought a huge folder of lots of cool stuff over and used ancestry for a while . Now that I’m into I’ve found lots of stuff but when you mentioned the kingery’s I remembered finding them in my genealogy not sure if it’s any correlation but I’m related to a Samuel kingery jr (born:1 apr 1833-14 nov 1913) son of possibly Samuel sr and Catherine “Eve” Evelene and he married Catherine “Katie” kingery (maiden name was possibly Mullen) ❤
ОтветитьI found an old picture with many former MLB players from a long time ago and I wanted to know when the photo could have been taken, so I told chatgpt the names and chatgpt made a table with their birth and death dates to show me when exactly it could have been taken. Cool stuff
ОтветитьPerplexityAI not only scans the Internet but gives you the links to the articles it finds.
ОтветитьI tried this and it told me it can only help when the people you're searching for are famous, well known or notable. So, it did help me clarify some links on 'known' ancestors of mine.
ОтветитьI can't use ChatGPT at all, as you must have a cellphone to sign up. They won't accept a home phone, as some other services have in the past.
ОтветитьI wrote a simple script for ChatGPT that allows me to post the script, then copy and paste source info from an Ancestry census record, and it creates a perfectly formatted citation in the Evidence Explained style. I struggled until I learned I had to feed it a template first.
ОтветитьFamily Search's AI is not very strong on the "I" part. It needs improvement
ОтветитьI don't see Chromosome Painter or ChatGPT anywhere. Where's it located? It is NOT anywhere on Ancestry that I have seen.
ОтветитьGreat video!
ОтветитьChat GPT has changed significantly since you posted this video.
Ответитьi generated a family tree with JSON values with parent childs values chatgpt generated a text output works like a charm
but there is a limit for the values
so for small presentations it works fine
Tried everything you mentioned, and all I got was absolute nonsense repeating the facts I put in while adding some complete nonsense. Sorry
ОтветитьThat’s a great idea to use it to generate a narrative based on a set of facts. Really handy for someone wanting to publish their research for their family
ОтветитьChatGPT said Shirley Bassey was born in India!
ОтветитьExcellent video and great advice about ChatGPT being a tool and you need to check the facts and date it gives it. Learning how to "Prompt" correctly and with the right commands is key. Just starting my own family trace, so this was a great help. Thanks. New Sub and bell set.
ОтветитьYou just explained ChatGPT better than computer guys do.... excellent! Everything you explained is so spot on.
It is also good at writing, translating and correcting small chunks of computer code. I think that might be what it is still best at.
Your advice is so good.... hope everyone takes it. ChatGPT can be a big time saver in some instances.
Amy Johnson Crow! You are THE BEST. Thanks so much for addressing this topic, and the helpful comments your video has elicited.
ОтветитьFun Fact: Google used to demonize, ridicule, attack, and get fired anyone who used to talk about AI or mention AI or even suggest AI was possible. Google is evil.
ОтветитьThank you for confirming what I had hoped is possible. Besides the genealogy tips, I appreciate the mini tutorial on ChatGPT, which helps me understand what it is and a better way to ask for help.
ОтветитьAI has a huge problem -- it doesn't know how to say, "I don't know." Also, AI doesn't UNDERSTAND anything. ChatGPT essentially just guesses what the next word will be. And that means it is good at spewing plausible-sounding garbage. It SOUNDS right. But if you trust anything that chatGPT says, you've got a problem. That's not to say that chatGPT is ALWAYS wrong. It means that you cannot depend on anything it says without triple-checking it. And in genealogy, you often cannot check at all. After all, if you can't find the records in the first place, chances are, you can't find the records to check what chatGPT has told you.
ОтветитьChatGPT often invents hogwash to fit a narrative. Even regarding facts that are available in the public domain.
ОтветитьI haven’t heard of any horror story with disastrous results. Can you point out some of these disastrous results?
ОтветитьI’ve only dabbled with chatGPT thought I could use it for something like my genealogy. This was a great educational video. Thank you so much!
ОтветитьTHANKS. This is the best synopsis of ChatGPT that I have seen. I now think I can get some usefulness from it.
ОтветитьThe technical term for an AI “making up facts” is “hallucination”; it certainly fits.
ОтветитьThis is a terrific explanation that gives guidelines that are both specific to one type of use case - and general for any use! Bravo.
ОтветитьAmy, you’re a very good speaker. Obviously very intelligent.
ОтветитьI don't trust ChatGPT to get facts correct, but using it to create a first draft of either a table or a bunch of paragraphs after inputting facts is a pretty interesting idea.
ОтветитьI am going to try this right away. I have been doing genealogy for years, and I've been wanting to write my ancestors' stories but it is very laborious. Thank you so much for bringing this to the genealogy community. The charting aspect is useful too. Thank you so much for sharing this.
ОтветитьIf I have to provide the facts about a person, place or thing to ChatGPT I don't see the benefit of it as far as furthering my research. If I'm writing a book or article then I guess it would be helpful. To be honest I have no experience with it. I guess I'll try it one of these days. I did enjoy the video though. Wonderfully presented as always. Thanks
ОтветитьAmy, in all honesty, I learned more about the basics of ChatGPT from your video here than any of the ones I have viewed that were made by CIS scholars. Thank you.
ОтветитьThank you, this was great and easy to understand!
Ответитьthis video gave me a better understanding of AI, and introduced me to 'Chat GPT' something I had never heard of.
ОтветитьThis was cool to use, with great results, but it did extrapolate and fabricate additional information beyond my initial facts... and it did so with great confidence. ;-). So it still has to be double checked.
ОтветитьNice explanation and also nice practical uses of hose ai tool. Nice also the focus you have made in explaining what a language model is. It was pretty important and correct.
I will just add some corrections.
Comparing these large language models with chat bots is not completely wrong in a first approach just to get started. But it is really stretching it. And most of their essence is lost in that analogy. So one should give a quick follow up or at least leave a little warning.
You made the excellent distinction between facts and what ChatGPT (and his cousins) gives. But there is a catch here.
Those uncountable texts that they were trained with are full of data/facts embedded in them. And this means that they can indeed be used as a searching tool.
One just has to realize that the legitimacy of those facts is very far from being credible and that the one who is giving them has a strong editorial power over them. Someone with a 2 year old kid's concept of what being factual is.
If used like a search engine, another way to think of those language models, is as interacting with one of those experienced old geezer full of stories and full of "knowledge" about the world. One that worked as a storyteller before retiring. One who always thrived in telling stories.
So. This means that he could be entertaining and even interesting. This means that he could even know some old forgotten things that happened who knows when.
But in all of it, what is truth and what is story?
probably even he doesn't really remember (or care that much).
In fact when using google search and co, one should already have a very similar attitude.
Who really knows how much credible the net is? maybe 0.001% of the facts in the internet can be believed (or some other statistic invented by me)? I mention this because when hearing some people declaiming some "real facts", it seem that some of those people forget where they got them from.
Excellent explanation. The majority of us have been engaging with AI/Machine Learning in different capacities through the products we use on a daily basis. ChatGPT was just the product that made it more accessible and triggered a mini AI revolution. I'm looking forward to a consumer-friendly chatbot that understands my personal genealogy data - my gedcoms, downloaded research files, and notes. Imagine asking it to find relatives who lived in DeKalb, NY between 1804 and 1860 and put those into a spreadsheet with birth and death dates where available, and it just spits that out, even if the data isn't properly added to a database somewhere. Maybe I scribbled it on a piece of paper and took a photo with my phone, but it still ends up on that spreadsheet. I want that.
I do think worries about AI rendering all our blood, sweat, and tears from the last 40 years obsolete is probably unwarranted at this point. Unless ChatGPT (or its cousins) can negotiate with the Nebraska Historical Society on my behalf and persuade them to provide me access to some 1911 records from the Geneva reform school (presumably stashed away in a dusty basement filing cabinet) there's ample non-AI work left for us to undertake, including of course the gathering of family stories, as others have mentioned.
Interesting and very useful article. My own use of ChatGPT has been for idea generation, especially in the domain of fiction writing, where the hallucinations matter less.
One current deficiency is that interactions are size-limited, and if you are working with larger documents or datasets, it is sometimes very tricky to figure out how to slice things up while maintaining continuity.
It is also relevant to point out that ChatGPT, in spite of not being a search engine (by the way, Google isn't the only or the best search engine out there), does contain a not insignificant amount of information. The problem is that there's no way to tell whether output is based on valid information or not. One potential method of improving this that I read about is to use ChatGPT itself to compare a reasonable sample of outputs to a given prompt generated by different instances of ChatGPT. The logic here is that facts that are training-based will be relatively constant (i.e., signal), while made-up facts (i.e., noise) will vary, so the comparison will be able to filter out the noise to produce a more accurate signal. This filtering could be fully automated as part of a future chat app.
(The icing on the cake for me of this video was that Lawrence County, Ohio is also an important location in my own family tree.)
Thank you for the video.
ОтветитьWhat "horror stories"?
ОтветитьI will need to watch again.....didn't hear specific instructions for doing genealogy research.
ОтветитьI think data driven individuals would be drawn to the chart first.
Ответитьbest explanation I’ve seen of how to approach chatgpt for nontechnical users. Very concise and accurate l
ОтветитьThank you for such good information about AI tools. I've started using AI for my own research (unrelated to genealogy) - and everything you discussed will help me immensely. Thanks again!
ОтветитьThank you for this. Ive been doing genealogical research for over 25 yrs and this is new info to me. Thank you again
ОтветитьNew subscriber. Learned a lot about chatGPT from your video. I was wanting to use chatGPT as background research on the history of a place or event but not a person or ancestor. For example, I research the lives and times of members killed in WWII and write stories about where they grew up and the military engagement in which they lost their lives. Would chatGPT be a good and trustworthy source for that kind of information?
Ответить