YT YouTube2Text

YouTube2Text → Transcripts → Ho trasformato ChatGPT nella mia Assistente Personale (Guida Completa)

Ho trasformato ChatGPT nella mia Assistente Personale (Guida Completa) — Transcript

by Antonio Guadagno · 3,368 words · 510 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:01Today I’ll show you how to transform
0:03Claude, Gemini, and ChatGPT from the
0:06standard chat everyone uses into a
0:08personal assistant that manages all
0:10your work and potentially your entire
0:13life. I've been using this system for
0:17about a month now; it has saved me a
0:19ton of time, and therefore a lot of
0:21money, and the most interesting thing
0:23is that building it is truly simple.
0:28You just need to follow the 10 steps I
0:30share in this video, which took me a
0:33lot of time and many experiments, but
0:36will take you only about 20 minutes,
0:38because I’m giving you the perfect
0:41recipe to have an assistant that helps
0:44you with every task each day. If you
0:48don’t know me, my name is Antonio, I
0:50am a consultant and computer engineer,
0:51and here on YouTube, I talk about
0:53artificial intelligence. I’ve worked
0:57with many companies, including Google,
0:59Rai, and IBM, and my goal on social
1:01media is to publish no-frills,
1:03content-rich videos about AI. If you
1:06find this useful, subscribe to the
1:08channel. First of all, if we want to
1:12turn Claude, Gemini, or ChatGPT into
1:14the perfect personal assistant for us,
1:17the very first thing we must do is
1:19allow these chatbots to learn
1:22information about us, so they can store
1:24it and use it in all our future chats.
1:29So, what we do is go into the settings.
1:33Typically, there is a section dedicated
1:36to customization, and we simply
1:37activate the chatbot's memory. For
1:41example, I did this on ChatGPT. What
1:45happens when you turn it on is that
1:48when we chat with Claude, Gemini, or
1:51ChatGPT, they analyze the entire
1:53conversation, extract the most
1:56important and relevant information, and
1:59save it in their memory to use in
2:01future requests. For instance, ChatGPT
2:06has learned that I work in AI adoption
2:09training for companies and that I
2:11publish videos on YouTube. It has
2:15stored this information, so I no longer
2:18have to specify it every time, making
2:20me much faster at defining the context.
2:24Obviously, the second step is that once
2:27we have activated the memory, we need
2:29to populate it, we need to fill it with
2:31information, and here we can follow two
2:34approaches. The first is to leave
2:37everything to chance, allowing the
2:39artificial intelligence to learn who we
2:42are and what we do on its own, chat
2:44after chat. Or, a second approach that
2:49I like much more, we can use this
2:51prompt, which I've left for free in the
2:54description below, to copy and paste it
2:57into Claude, Gemini, or ChatGPT and
2:59start a real interview where the
3:02artificial intelligence asks us
3:04questions about ourselves, our work,
3:06our hobbies, and our projects, and by
3:09answering, we provide information that
3:12will be saved to memory. This approach,
3:16in my opinion, is much more interesting
3:18, first of all because we get it all
3:19done in one go. The interview, in fact,
3:22is just one, and we do it right away.
3:25The second reason I appreciate it, on
3:27the other hand, is that since we are
3:29being asked questions, there is much
3:31less chance of us forgetting something
3:34important. The artificial intelligence,
3:37in fact, asks us and we answer.
3:40Obviously, the interesting thing is
3:42that everything happens automatically,
3:45meaning we answer and the memory is
3:47updated automatically. If you use
3:51Claude, by the way, which is much more
3:54advanced in terms of memory, the memory
3:57is also updated in real-time, whereas
4:00on ChatGPT we have a small delay
4:03because chats are analyzed only at the
4:06end and only after some time. Once we
4:10have activated and populated the memory
4:13, the third step is to keep it updated.
4:17This is fundamental because, obviously,
4:20over time goals change, projects change
4:23, and sometimes we even change jobs.
4:26Fortunately, Gemini, Claude, and
4:29ChatGPT all provide a section for
4:32memory management. The only downside,
4:37in my opinion, is that we cannot edit
4:40the memory text manually; instead, we
4:44are given a small chat where we make
4:47requests, asking to delete this, update
4:50that, and so on. In my opinion, today,
4:55the chatbot that manages memory the
4:58best is Claude. GPT, which has a single
5:04file containing everything: information
5:07on work, hobbies, projects, personal
5:10life, and so on. Claude, on the other
5:15hand, has a memory made up of many
5:17files, each independent. These files
5:21are therefore divided by topic. There
5:24is a file for project X, a file for gym
5:26workouts, and so on. And the moment I'm
5:31no longer interested in a project, for
5:33example, I can simply delete that file
5:36while keeping everything else intact.
5:40Furthermore, if I want to make any
5:42changes, I have the ability to make a
5:45precise and targeted edit because I
5:47only modify the file concerning the
5:49topic I'm interested in, leaving
5:51everything else untouched. Leaving
5:55memory behind, the fourth step involves
5:58connecting our agent, our assistant, to
6:01the applications we use most often.
6:05Specifically, in my opinion, the bare
6:08minimum is connecting to email, Drive,
6:11and the calendar. I also recommend
6:15connecting to Gitab, which is a very
6:17interesting application I’ve made a
6:19dedicated video about, which I’ve
6:21linked below in the description. To
6:25connect our AI agents to the
6:26applications we use, we rely on plugins
6:29, also known as connectors. The
6:32approach is more or less the same for
6:34all chatbots, so I’ll show you using
6:37ChatGPT, but it’s virtually identical
6:39on Claude. I go to the left, click on
6:43plugins, choose the one I’m
6:45interested in, like Outlook, install
6:48the plugin, and at that point, I’m
6:50asked to connect my account using my
6:53email. I won’t do that here because I
6:57don’t use Outlook; I use Gmail, and
6:59I’m already connected to that.
7:03Getting into the details, the reason I
7:06really appreciate ChatGPT plugins is
7:08that once installed, I can connect more
7:10than one account to them. For example,
7:14here we have Gmail, and as you can see,
7:17I’ve linked my personal account, my
7:20company account, and also the WBO
7:23account, which is the company I do AI
7:25adoption projects with. Why is this
7:28such a great feature? Because once I
7:31link all the accounts and give them a
7:35nickname—in this case, personal,
7:37company, and WBO—I don’t have to
7:41keep switching back and forth; I simply
7:44ask the AI to connect to my personal
7:47profile rather than the WBO one. This
7:52way, it understands which account to
7:54connect to and responds to that
7:56specific request. Moving on to steps 5
8:01and 6, these involve creating projects
8:04and skills, two tools that, at first
8:07glance, seem interchangeable but are
8:10actually profoundly different. That’s
8:15why I’ll explain how I use them now,
8:17showing you an approach that I find
8:19works very, very well. Starting with
8:23projects, these are spaces or folders
8:26we can use to manage clients,
8:29initiatives, and tasks more effectively
8:33. For example, I created a project for
8:37Rossi Srl, a client of mine whose name
8:40I’ve changed for privacy, but I’ve
8:42also created projects for my personal
8:45initiatives, like AI Evolution, my
8:48practical course on artificial
8:50intelligence applied to work. By the
8:54way, if you’re interested, I’ve
8:56left the link below in the description.
8:59In any case, every project is defined
9:01or characterized by two elements,
9:04namely instructions and sources. Both
9:08are very, very useful, but if we don't
9:11want to risk using projects poorly and
9:14substantially destroying their
9:16performance, we absolutely must
9:18understand the differences between
9:21instructions and sources, because if we
9:24use them wrongly, the projects become
9:26essentially useless. And now, in fact,
9:30let's see what both of these things are
9:32for. As for instructions, these are
9:36pieces of information that the
9:38artificial intelligence reads from
9:41start to finish every single time for
9:43all the chats we start within that
9:45project. For example, to give a
9:50concrete example, within the
9:52instructions we can explain the context
9:54of the initiative, who the client is,
9:57what work we are doing for them, what
9:59role we play in the initiative, and
10:02what our goal is. Essentially,
10:05everything we write in the instructions
10:09will be read and kept in mind for all
10:12the chats we start in that project.
10:16Conversely, regarding sources, these
10:19are typically documents, PDFs, or even
10:22links to websites that the artificial
10:25intelligence might read to answer the
10:28questions we ask, but this does not
10:31mean that all sources are read for
10:34every chat from start to finish. The
10:40artificial intelligence simply has
10:42access to those documents or files, and
10:45when it thinks a file might be useful
10:48for answering a question we ask, it
10:51goes to read and consult it. However,
10:56it is wrong to think that at the
10:58beginning of every chat the artificial
11:00intelligence reads all the sources and
11:02therefore keeps everything written in
11:04these sources in mind. This means that
11:09if we want the artificial intelligence
11:11to consider a certain piece of
11:13information in every chat, it should
11:16not be put among the sources, but
11:18rather among the project instructions.
11:22And this is one of the most common
11:24mistakes I see made very often in the
11:26companies where I consult. Another
11:29trick, a piece of advice I give you, is
11:32to dedicate a small paragraph of the
11:34instructions to the topology of the
11:36sources. So, essentially, what you do
11:40is write in the instructions, for each
11:43source, its name, a brief summary of
11:46what is present in that source (i.e.,
11:48the content written in that file), and
11:51a short list of possible questions that
11:54can be answered by that file. This way,
11:59before starting a chat, the artificial
12:02intelligence reads the instructions,
12:05reads the list of all sources with
12:08their descriptions, and when you ask a
12:11question that requires consulting a PDF
12:14among the sources, the AI already knows
12:17which PDFs are available and, most
12:20importantly, which source to consult
12:22and which one not to. Finally, another
12:27thing I appreciate about projects is
12:30that they allow us to keep all chats
12:33related to a certain topic within a
12:35single space. And this, in my opinion,
12:40is fundamental, especially for those
12:42initiatives that go on for a long time
12:45and for which we ask many questions
12:47that we would otherwise risk losing
12:49among all the other chats. By using
12:53projects, we have our chats neatly
12:55organized and we don't get confused.
12:58Moving on to talk about skills, here we
13:01go into a whole new world. The focus is
13:05no longer on the client, the project,
13:08or the initiative, but on the specific
13:11activity we want the artificial
13:13intelligence to perform, for example,
13:16writing quotes or creating slides. When
13:21we create a skill, in fact, we are
13:24effectively teaching a competency to
13:27the artificial intelligence; we don't
13:29just explain what we want it to do, but
13:32also how we want it done. By creating a
13:38skill, we crystallize all the steps the
13:41artificial intelligence must follow
13:44when performing a certain activity, so
13:47we define not only the "what" but also
13:50the "how." To give a concrete example,
13:55if I create a skill that analyzes an
13:57Excel sheet that is crucial for my
14:00business, I define all the steps of
14:02that analysis, such as which sheets to
14:05look at, which fields interest us,
14:08which calculations need to be done, and
14:10so on. The moment I run that same skill
14:15multiple times, the analysis will
14:17always be repeated in the same way.
14:22Obviously, the huge advantage of skills
14:24is not just that we lock in knowledge
14:27or a competency, but also that, once
14:29the command is created, if I want to
14:32ask the AI to perform that activity, I
14:34just need to launch the skill and don't
14:37need to explain what to do and how to
14:39do it from scratch every time. In
14:44essence, thanks to skills, we save a
14:46lot of time, especially on activities
14:49we perform very often. Creating a skill
14:52is truly very simple. We just need to
14:56go into the chat, run the/skillator
14:59command, and then explain the
15:01competency we want to teach the
15:03artificial intelligence. In this case
15:08too, I have a couple of tips, a couple
15:11of tricks for you, the first of which
15:13I’ve actually already mentioned: that
15:16is, when we create a skill, we must try
15:19to be as specific as possible about the
15:22steps and the activities the artificial
15:24intelligence will need to perform. So,
15:29let's write a bulleted list, an ordered
15:31list of all the steps. In this way, we
15:35will create a skill that, even if we
15:38run it 50 times, will not only do the
15:41same thing 50 times, but will always do
15:44it in the exact same way, thus
15:46following a specific process. As for
15:50the second tip, this one concerns the
15:53complexity of the skills. My advice, in
15:58fact, is not to create an ultra-complex
16:02skill that does 50 things all at once,
16:05but rather many small, highly focused
16:08skills that we then execute in sequence
16:12within the same chat to effectively
16:15create a small workflow that takes us
16:18from the beginning to the end of a
16:21rather complex job. For example, I have
16:26a skill that helps me evaluate and
16:28focus on a YouTube video idea. Then I
16:32have a skill that helps me find a title
16:34idea for that video, and finally, I
16:36have a skill that helps me write the
16:38description for that video. When I
16:42launch all these skills in sequence, I
16:44have effectively received help in
16:46publishing a YouTube video. The even
16:51more interesting thing is that projects
16:54and skills can also be used in
16:56combination, in the sense that I might
16:59have the Rossi srl project, I have the
17:02skill that helps me write quotes, and
17:04obviously, I can use the quotes skill
17:07within the Rossi srl project. In this
17:12way, I create a synergy between these
17:15two very useful functions, and
17:17essentially I have projects that help
17:19me organize my work and skills that
17:22help me speed things up by avoiding
17:24repeating the same prompts over and
17:26over. Moving on to the seventh step,
17:31which will allow us to turn our chatbot
17:34into the perfect personal assistant,
17:37this concerns the creation of small
17:39automations with which we can speed up
17:42the execution of those more mechanical
17:45and repetitive tasks. These automations
17:50on ChatGPT are called scheduled tasks
17:53and are essentially prompts that are
17:56executed the same way automatically,
17:58respecting a certain frequency that we
18:01obviously choose ourselves, for example
18:04, every morning at 9:00 AM or once a
18:07week on Thursday at 5:00 PM. Creating
18:11these scheduled tasks is truly very
18:14simple. We just need to go to the left,
18:17click on scheduled, and then everything
18:20is done simply via chat. We explain
18:23when this action should be performed—
18:26in my case, every day at 9:00 a.m.—
18:28and then, of course, we also explain
18:30what this action is. In my case, it's a
18:34sort of press review with the latest
18:36news on artificial intelligence. We
18:41then hit enter, let ChatGPT understand
18:43what this scheduled task is, and in the
18:46end, we can both view it and, obviously
18:49, edit it. The reason why, in my
18:53opinion, these scheduled tasks and
18:55automations are very interesting is
18:57that they remove manual work and
18:59eliminate manual steps. In this case,
19:03for example, I no longer have to go
19:05onto ChatGPT every morning, copy and
19:08paste the prompt, and launch it
19:09manually; I simply receive a
19:11notification with the information I
19:13need, all ready to go. Obviously, if we
19:18create 4, 5, or 10 scheduled tasks, the
19:20benefit we get from them is even
19:22greater. Another very classic example
19:26of a scheduled task we can create is
19:29email triage. Earlier, for example, I
19:32showed you that I connected to three
19:34Gmail accounts—to three mailboxes—
19:37via the plugin. What I could do is
19:40create another scheduled task that
19:42reads the emails from all these boxes
19:45every morning and tells me which ones
19:47are the most important and the most
19:49urgent. This way, I avoid the
19:54mechanical hassle of switching from one
19:56account to another and can be much
19:59faster in my morning routine. Moving on
20:03to the eighth step, this concerns
20:06custom instructions, which are very
20:08often either overlooked or, worse yet,
20:11confused with memory. I'll start by
20:16saying that custom instructions are not
20:18meant to explain to the AI who you are
20:20or what your job is, because there's
20:22already memory for that. They are meant
20:27instead to define communication rules
20:29between you and the artificial
20:31intelligence. To give you an example,
20:34when I ask it to explain a concept to
20:37me, I always want it to give me a
20:38practical example as well, so I can
20:40better understand and cement the
20:42concept. Since this isn't an indication
20:47about my work, but rather an indication
20:50of how I want it to behave with me, I
20:53go and add it to my custom instructions
20:56. In this case, too, I have a couple of
21:01tips, a couple of tricks, the first of
21:03which is to keep the custom
21:05instructions rather concise because
21:07they are added to the beginning of
21:09every chat, and the longer they are,
21:11the more context they occupy. The
21:16second tip is to keep them as generic
21:19as possible, because since these
21:21instructions are applied to every chat,
21:24they must be very versatile and must
21:27apply to every context and every task.
21:33Moving on to step 9, which in my
21:35opinion is the absolute most important,
21:37it consists of installing the Claude,
21:40Gemini, or ChatGPT application on your
21:43smartphone. Only in this way, in fact,
21:47will you always carry your agent, your
21:50assistant, with you and have the
21:52ability to ask questions even when you
21:55are not at your PC. You have no idea
21:58how many times I've found myself,
22:00perhaps at the station in Milan. I was
22:03there waiting, maybe I was on my way to
22:05a client or some event, and I didn't
22:06know what to do. I would talk to my
22:10assistant and ask it anything:
22:12information about my trains, or details
22:14about hotels I might have booked on
22:16Booking. And my assistant would answer
22:21all these questions because, being
22:23connected to the plugin and therefore
22:26to Gmail, it had access to all the
22:28emails with Booking reservation
22:30confirmations or train tickets. I would
22:35also ask things like, "Connect to Drive
22:38, give me a summary of the last calls
22:40made with the client," so I could
22:42refresh my memory and arrive a bit more
22:44prepared, or I would ask questions
22:46about my notes, my slides, or ask it to
22:49search for information online, and so
22:51on. Another even more interesting thing
22:56is that through the smartphone
22:57application, it is really very, very
22:59simple to interact with your assistant
23:02using your voice. In fact, you have
23:06both voice mode, where you speak and it
23:09responds out loud almost as if it were
23:11a call, but you also have dictation,
23:14which basically allows you to send
23:16something like a WhatsApp audio and
23:19then receive a text response from the
23:21artificial intelligence. This is very,
23:25very convenient because it allows you
23:28to talk to your assistant even when you
23:30are on the street and don't have the
23:33ability to look at the screen. Last but
23:37not least, the tenth step consists of
23:39installing the computer application for
23:41Claude or ChatGPT. Only in this way, in
23:46fact, will we be able to leverage 100%
23:49of the agentic capabilities of ChatGPT
23:52Work and Claude Cowork. By installing
23:56these applications, we will unlock new
23:58possibilities for artificial
24:00intelligence. These agents, in fact,
24:03will be able to work with the files we
24:05have on our computer. They will be able
24:09to create, modify, delete, move, and
24:11organize them, and so on. Furthermore,
24:16they will also have the ability to take
24:18control of our computer and use the
24:20applications we use every day on our
24:22behalf. In this way, we will be able to
24:26delegate even more tasks to artificial
24:28intelligence. If you are interested in
24:31exploring these topics, specifically
24:34the desktop applications for ChatGPT
24:36and Claude, I recommend watching these
24:39two videos where I explain how these
24:41tools work and provide plenty of
24:43practical examples. Yes.

About this transcript

This page contains the full transcript of Ho trasformato ChatGPT nella mia Assistente Personale (Guida Completa) by Antonio Guadagno, generated from the public captions YouTube serves with the video. The transcript has 3,368 words across 510 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 translate the transcript into 100+ languages.

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, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.