YT YouTube2Text

YouTube2Text → Transcripts → Нейросеть нашла моих предков! Я теперь богат!? | Семейное древо через ИИ | Генеалогическое древо

Нейросеть нашла моих предков! Я теперь богат!? | Семейное древо через ИИ | Генеалогическое древо — Transcript

by Влад | Про нейросети · 2,950 words · 457 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Just look at this, the neural network
0:01found my ancestors all the way back to
0:03the 15th century. The agent finds data
0:04on the internet by itself and deciphers
0:06these archival records. All you need to
0:08do is install ChatGPT or Claude on your
0:10computer and send it these instructions
0:12. But first, you need to set up the
0:13browser-controlling neural networks.
0:15Then it will be able to open archive
0:17websites, find the necessary data, and
0:19even send official requests. I also
0:21helped my agent with additional
0:22resources like Perplexity. Anyway, I'll
0:24show you how to do this in this video.
0:26All the links will be down in the
0:27description or on my Telegram channel.
0:29Be sure to like this video and subscribe
0:31to the channel. Thanks a lot in advance
0:33. So, what's the deal? I started coming
0:35across various clips in Reels and Shorts
0:38.
0:39About this skill that allegedly searches
0:41for information about your relatives.
0:44And I decided to test it out on my own
0:46family. This is what I ended up with.
0:49I also created an additional small
0:51website that I can open on the internet
0:54and send to my family to look at.
0:57For certain branches, I managed to get
0:59back to about the end of the 15th
1:01century.
1:02In some places, I even found documents
1:06with full addresses, but in others, I
1:09only had names to go by, like here with
1:12Andrey, because it's just a patronymic.
1:16So, some of the earlier layers remained
1:18more as assumptions that still need to
1:20be verified. But one way or another, it
1:22searched for all of this using these
1:23types of scans. It went into archives,
1:26parsed documents like these, and scanned
1:29them for my name, surname, and so on—
1:32meaning my ancestors.
1:35But simply installing this skill, like
1:37everyone in the Reels says, and
1:38expecting it to do it with just one
1:40prompt...In reality, that didn't work.
1:43The total research time took me about a
1:45day. I had to help it out quite a bit,
1:48so now I want to share the entire
1:50process from start to finish: what I did
1:54, where I went, and how to set it all up
1:57. Look, I'll be working through the
1:59Codex neural network, which is basically
2:01ChatGPT from OpenAI.
2:03You can download it by just typing "
2:05download Codex for PC" into your browser
2:07, and this window will appear.
2:09You can download it here for Windows or
2:12Mac. The only thing is, when you open
2:16Codex on your computer, it will
2:18immediately ask you to log in.
2:21You log in via your browser, but it will
2:24additionally ask you to verify via your
2:27phone number.
2:29Actually, it won't just suggest it, it
2:31will insist that you enter your phone
2:33number and receive an SMS there.
2:35You can also do this using any online
2:37SMS receiving service. So, in principle
2:40, there are no difficulties with this.
2:42Receiving SMS messages costs just a few
2:44cents, but there will be a small
2:45difficulty later on. That is with the
2:47payment. If you need to pay for ChatGPT
2:50or a subscription, I use this Zarub bot
2:52. This is not an advertisement, I just
2:54use it myself. And a link to it will be
2:56in the description. You can use it.
2:59I think they give you some free points
3:02when you open a card.
3:04So, you issue a card here and use it to
3:07pay for any purchases on foreign online
3:09services.
3:10In general, this genealogical skill is
3:13designed for other environments as well
3:15, like Claude Code, for example.
3:18But I will continue to show you using
3:20the example of Codeium, my own, so I
3:22don't have to jump between different
3:24interfaces. Look, when you first log
3:26into Codeium, let's configure it a
3:28little bit. To start, we'll go right
3:30here into the settings. And here we will
3:33enable this mode. It is called full
3:36access. This is done so that you can
3:38select this full access mode right here
3:40in your chat. This is so that when your
3:42agent is working, it doesn't interrupt
3:44you and has full capabilities to
3:46navigate to the pages it needs. The
3:48second point is that we need to
3:49configure permissions so it can use your
3:52browser. For the browser, you will need
3:54Google Chrome. Any other one won't work
3:55.
3:56To do this, you go to this menu item, it
3:58is called computer use.
4:00And you will need to click on connect
4:03Google Chrome right here. The Google
4:05Chrome web store will open.
4:08And after installation, you will need to
4:10authorize; accordingly, you authorize
4:12through the browser, and a window like
4:14this will appear, and a checkmark will
4:16light up green here, indicating that you
4:18can use the Google Chrome extension.
4:21This is absolutely mandatory. Also, this
4:23is not mentioned anywhere in this skill
4:25. Various MCP CLs are listed here, which
4:27it ultimately failed to connect to for
4:30me. Either they were paid somewhere, but
4:32I decided to do everything for free via
4:33Google Chrome. The extension handled it
4:35perfectly, in principle. Next, you will
4:37need to select the model here,
4:38accordingly. I recommend that if you
4:41have a subscription for around $ 100,
4:43you can also use Astra. It will work
4:46better.
4:47And if you really want to go all out and
4:49do some proper, solid research,
4:51constantly running tasks quickly, you
4:53can activate this plan for speed, and it
4:55will handle everything quickly.
4:57But personally, I worked on the 6.1
4:59model. Sonnet. It’s a bit cheaper and
5:02it’s a bit slower.
5:03You can set its reasoning to very high
5:05here, and in principle, it will handle
5:07the task, it will just work a bit slower
5:09.
5:09And at some stages, it might lack a bit
5:11of intelligence, but I was constantly
5:13nudging it, so, in principle, it handled
5:15the task. Next, look, we need to create
5:18a folder on the desktop. I named the
5:21folder "tree".
5:23So, I just create an empty folder here,
5:26right, and in the agent, here in ChatGPT
5:29Code, I just select it, click new
5:31project, and choose this folder from my
5:34desktop.
5:36In my case, it was the "tree" folder, I
5:39selected it, and it appeared here
5:41accordingly.
5:42This is how it will look. So, you have
5:44this folder connected, you're working on
5:46the computer, you have full access, and
5:49you have the power, say, the power of
5:51ChatGPT 6 or Astra. You can also use
5:53Astra on a very high setting. And then
5:55you paste this GitHub instruction into
5:56the chat.
5:57So, you copy, paste, and write something
6:00to it in roughly this format.
6:02Here is how I did it in the project. So
6:03, create the basic documentation in the
6:05project and clone this skill for us. It
6:07clones it so that it doesn't constantly
6:10have to run to the internet for
6:12information, so that it's always right
6:14here in front of it.
6:16And in accordance with this, I say: "
6:18Ask me the necessary questions."
6:20Then it set everything up for me here
6:22and asked me the questions I needed.
6:25Meaning about my closest relatives,
6:27about my dad, mom, brothers, sisters,
6:29and so on and so forth.
6:31So, I wrote something to it in roughly
6:33this kind of format. I gave it just a
6:36little information about my father, then
6:38gave it some about my mother, and about
6:40my grandmother. But naturally, that
6:42wasn't everything.
6:44Later, in the process of the dialogue, I
6:46kept adding more information.
6:49I did this to see how the skill works,
6:51observed where it goes and what it does
6:53, and then a little later I added a bit
6:55more information about myself.
6:58So, you don't really need to wait until
6:59you have the complete story.
7:01Start with what you already know,
7:03whatever first comes to mind; the agent
7:06will clarify with you and move forward
7:08anyway.
7:10Also, notice that at the very beginning
7:12I told it: "Create some project
7:14documentation." This is done so that we
7:17have such documentation implemented in
7:19our project. I'll show it to you now. So
7:21, this is my project folder. And the
7:23project documentation is basically the
7:25agents.md file. This is the most
7:26important one for Claude. The file will
7:28be named agents.md. Here I have all the
7:31main information, meaning everything it
7:33needs to follow for this skill.
7:35There is already some information about
7:37me, information about my relatives, and
7:39notes it copied from the skill on how to
7:41use the browser, and so on.
7:43Basically, it will create all this
7:45documentation itself so that when you
7:47give it a new task, it follows those
7:49instructions. Including when you point
7:51out any mistakes in the chat later, it
7:54will immediately record them in its
7:56documentation and won't repeat those
7:58errors. That is why this file is created
8:01, called, yes, agents.md.
8:03Also, it will save various pieces of
8:05information it finds in this folder,
8:07including document scans, links, and
8:10various files it creates for itself.
8:14As a result, it did the work, compiled a
8:16report with a scheme including sources,
8:18and I opened various links—those are
8:20roughly the kinds of links it sent me.
8:24Meaning, it found my great-grandfather
8:27here, along with his photos.
8:30And it also linked them to sources,
8:32explaining exactly why it chose that
8:34link and how it relates to me.
8:36So, all of that was there, and I
8:38reviewed it once more. And pay attention
8:40.
8:41Then I write to it: "Now write me a
8:42prompt in accordance with this skill and
8:45give me an MD file with the data you
8:46already have."
8:48I do this in order to later send all
8:51this information for additional in-depth
8:54research.
8:55You can do it using the same ChatGPT you
8:58already pay for, or you can use a site
9:00called Perplexity.
9:03It is known precisely for being good at
9:04conducting research.
9:06So, in addition to this agent's work, we
9:08also use web research because sometimes
9:11there is a bit more information on the
9:13surface than this agent can find, as it
9:15will specifically look at certain sites
9:18, parse superficial data faster, and
9:20find additional information.
9:23If you are doing this with ChatGPT, you
9:25feed it the prompt that it gives you.
9:28You take this exact prompt, copy it here
9:30, and paste it into the web version of
9:33ChatGPT.
9:35You don't need to configure anything
9:37here; just set the model to Pro, or if
9:39you have the $ 20 subscription, set it
9:40to extra high.
9:42Paste it here and make sure to click
9:43right here. A button will appear to
9:46convert all of this into text. Click
9:49that button exactly like that, and
9:51additionally, upload this file here as
9:53well. So, you find it in the folder here
9:57, drag it in, and click to add it. You
10:00need to click the plus sign and then
10:02select "deep research" so that it starts
10:04the actual deep research process.
10:06Then, we send it all off for review. It
10:08will work for some time and then give
10:10you a document just like this one. You
10:11can download it; or rather, you need to
10:13download it as a Markdown file. And then
10:15, we immediately send it back to him
10:17right here. That's it. Next, I ask him
10:19what we can actually do next from here.
10:21Let's keep going. So, I kept telling him
10:24: "Let's keep searching, let's dig
10:26deeper."
10:28And he continues to open various
10:29archives, save documents, and analyze
10:32different records.
10:34He keeps digging deeper because if you
10:36don't keep him working, he will search
10:38for a bit, find something, stop, and
10:40give you the result.
10:42It looks like a final result, but
10:44naturally, it will never be enough for
10:45you. Therefore, you have to look at the
10:47information he found and, roughly
10:49speaking, keep guiding him further. For
10:51instance, if you notice a familiar
10:53surname or name, or remember a specific
10:55year, just add that information here for
10:57him. By the way, I performed these deep
10:59research sessions three times during the
11:01process. He kept finding additional
11:04files on his own and then recorded them
11:06for me here. And then I would perform
11:08those deep research sessions again.
11:10So, I had about three attempts at this
11:13additional research.
11:15And one important clarification. I
11:16forgot to mention that during the
11:18process, he will naturally open Google
11:19Chrome here on your computer and use it
11:21.
11:22So, you need to have it open beforehand
11:24; otherwise, he won't be able to use it
11:26.
11:26Pay attention to this. It will have its
11:29own cursor here, so it's better not to
11:31interrupt it and let it work on its own
11:33. You should use, I don't know, either
11:34separate tabs or a different browser.
11:36Then I ran into a slight limitation.
11:39It kept getting stuck because in some
11:41documents, the patronymics were somehow
11:43cut off.
11:44And because of this, because this skill
11:47limits any kind of guessing of the
11:49patronymic.
11:51If it isn't explicitly visible, it
11:53doesn't develop that motive or that
11:55chain further.
11:58That is, it doesn't link this person in
12:00any way to, say, your grandfather.
12:03Even though the connection might be
12:05obvious, because, for example, the
12:07patronymic is cut off as "Pavl," but in
12:10reality, it's obviously "Pavlovich."
12:13So I additionally told it that let's not
12:17use these restrictions, these frames of
12:20this skill.
12:23That is, we will still build connections
12:25even based on patronymics that aren't
12:27displayed.
12:28Or, for example, if only initials are
12:31displayed, we will also consider them as
12:34a connection.
12:35And thanks to this action, I ended up
12:37with a lot of connections.
12:39I added a lot of different sources.
12:42And it started offering me many more
12:44options and reaching some really distant
12:46names.
12:48Then, at some point, the agent even
12:50offered to prepare requests to various
12:52institutions, like writing to a registry
12:54office, for instance.
12:56Well, most often, a document or request
12:57is submitted to the registry office.
13:00But I stopped at this stage and told it
13:04that let's pause with what we have and
13:08just make it all into a beautiful design
13:12in Figma.
13:14This time, with the help of design, I
13:16made a more convenient layout for myself
13:18, and then asked it to lay it all out in
13:21this format.
13:22So it actually made it into a little
13:24site for me.
13:25You can do this yourself as well.
13:29I will have a skill in my Telegram,
13:31which is basically, loosely speaking, an
13:33instruction.
13:35You just take that file I'll have there
13:38, download it from Telegram, upload it
13:41here into your project chat, and say: "
13:45I want to stop with what I have for now
13:48, and let's lay this all out in this
13:51visual style according to this
13:54instruction."
13:57You send him this instruction here,
13:59click "execute," and he will build you a
14:01site like this, where you can click on
14:03various things—for example, let's say I
14:06want to select him and click on "
14:07descendants" here.
14:09He will immediately map out all the
14:12lines, roughly from me down to my child
14:14, so to speak. There are various other
14:16cool features here as well. You can take
14:18a look; it’s very convenient to use.
14:20Hold down the space bar, a hand will
14:22appear, and you can drag to move around
14:24, basically, across this area.
14:27Depending on the skill, he will either
14:30assemble a version without photos or
14:32gather photos that were found, for
14:34example, like with my great-grandfather
14:36—he found him, meaning he took his photo
14:39from the source, so he didn't alter it
14:41in any way.
14:43Other photos were generated in ChatGPT,
14:46as it has a built-in image generator
14:48here.
14:50I initially uploaded a number of my own
14:52photos, that is, photos of my family.
14:56I sent him a series of my own photos,
14:58some from my wedding, photos of my
15:00brothers and sisters from VKontakte.
15:04And I told him, based on this—based on
15:07each person's photo or based on a pair—
15:10to generate either their father or their
15:12mother.
15:14That is, to repeat and try to capture
15:16their facial features and try to apply
15:17them to another person. So the photos
15:20here are naturally mostly fictional, but
15:23some photos he simply cleaned up. He
15:25took, for example, a photo of my father
15:27and just generated it on a white
15:29background.
15:31It looks cleaner that way, a bit more
15:33logical, just to make it easier to view
15:36everything.
15:37Well, as I already mentioned, I handled
15:39the archival photos differently.
15:42If the agent found an actual old
15:44snapshot, I asked him to keep the
15:46original and not process it in any way.
15:50Therefore, using my instruction, you
15:52first create a design—this kind of board
15:55—and look at it; if everything is fine,
15:57he will then suggest that you upload a
15:59few photos, just like this, into the
16:01folder. You upload them here into your
16:04main folder where you are working,
16:07upload a few photos, and tell him what
16:09you want. To also generate photos. The
16:12style is already locked in there, so it
16:14will use the exact same photo generation
16:16style.
16:17In total, the whole project took me
16:19about a day of work, with a few sessions
16:21for research and refining the design.
16:24So, do set aside time for searching and
16:27verifying what you find.
16:29Start with your closest relatives, the
16:32ones you know the most about.
16:34Then, gradually work your way through
16:37each branch. Thanks for watching. If you
16:39liked the video, be sure to hit like,
16:41subscribe to the channel, and leave a
16:43comment. It helps promote my channel.
16:45Bye-bye.

About this transcript

This page contains the full transcript of Нейросеть нашла моих предков! Я теперь богат!? | Семейное древо через ИИ | Генеалогическое древо by Влад | Про нейросети, generated from the public captions YouTube serves with the video. The transcript has 2,950 words across 457 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.