YT YouTube2Text

YouTube2Text → Transcripts → Dammi 20 Minuti e ti Insegno ChatGPT da Zero

Dammi 20 Minuti e ti Insegno ChatGPT da Zero — Transcript

by Giuseppe Castagna · 4,085 words · 614 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00ChatGPT was the very first artificial
0:02intelligence system I ever opened in my
0:04life, and I suspect it was the same for
0:06you. And it's also the one that, even
0:08today, I open on my phone at least 10
0:10times a day, especially because a ton
0:11of new features have been released
0:13recently that make it, in fact, one of
0:15the most complete tools for daily use.
0:18In these years of using this tool, I've
0:20made practically every mistake a
0:21beginner makes when they first open
0:23ChatGPT. So, in this video, I'll show
0:26you what I really wish I had known the
0:28first day I opened it, starting
0:30absolutely from the basics, explaining
0:32how ChatGPT works, what it's better at
0:33than the others, how much it costs, and
0:35above all, I'll show you how to use it
0:37for everything, practically right away.
0:40Let's begin. So, let's start with the
0:42question you must ask yourself
0:43immediately before opening any AI
0:45system, whether it's ChatGPT, Gemini,
0:46Claude, Grok, Manus, or all the others
0:48that have come out. And no, it’s not
0:50whether ChatGPT is better than Gemini
0:52or Claude—no, that’s the wrong
0:54question. Because starting to use one
0:56of these tools well, even though the
0:57basics are virtually the same for
0:59everyone, still requires some time, and
1:01making the wrong choice will make you
1:03waste weeks. The problem is that every
1:05day new models and new acronyms come
1:06out, and you basically can't understand
1:08a thing anymore. So the right question
1:10you need to ask is: which tool am I
1:12going to invest my time in? So, the
1:14main AI systems that exist today are
1:17the most famous ones. The first one by
1:19far is ChatGPT, the one we are looking
1:20at together right now in this video. It
1:22was the first, the most mainstream—
1:24we've all tried it, and today it is one
1:26of the most used apps in the world.
1:29Then there's Gemini: it's the one from
1:32Google, and it's really cool because we
1:34all use Google, and within Gemini,
1:36there's Gmail, Calendar, Search, and
1:39Drive. And then, finally, there's
1:42Claude, the one from Anthropic, which
1:43in recent months has become the coolest
1:45one for coding and intensive work. And
1:47without a doubt, the most generic,
1:49generalist, and default one for the
1:51vast majority of people is definitely
1:53ChatGPT. And this is for objective
1:56reasons. The first is web searching.
1:58When you need to know something
2:00immediately, without going crazy
2:01through 1,000 pages, ChatGPT is the one
2:03that answers you best in the fastest,
2:04most direct way compared to the other
2:06two. The second is the voice mode.
2:09ChatGPT, or rather OpenAI, the parent
2:11company of ChatGPT, bought Whisper, a
2:13company that developed a transcription
2:15system that is absolutely the most
2:16accurate of them all. You can talk to
2:19ChatGPT, and it responds with insane
2:21quality. On Claude, it sucks in
2:23comparison. The third is the phone. If
2:26you use it on a smartphone, the ChatGPT
2:27app is one of the most intuitive—
2:29actually, it's the most intuitive of
2:30all three, so it wins. And now I’m
2:33going to say something super honest
2:34that none of the other AI video
2:36creators will say, because it might
2:37cost me a few subscribers. I use all of
2:40these systems, both for personal
2:41passion and for work. I have
2:43implemented them in a ton of companies,
2:45both my own and others '. And the point
2:47is that for intense work and writing
2:49code, in my opinion, Claude is
2:51unbeatable. If you only use Gmail in
2:53your life, Gemini is perfectly fine,
2:54it’s also free, and there’s a video
2:56on my channel where I explain how to
2:57use it from scratch. But if your day is
3:00made up of quick questions, research,
3:02voice, and curiosity, okay, ChatGPT,
3:04nobody beats it. Okay, let’s look at
3:07the first thing everyone hits a wall
3:08with: the price, because there is a
3:10free version, but there are also
3:11several paid ones, so let’s see which
3:13one is best to choose based on how we
3:15use it. As of today, there are three
3:17plans for ChatGPT. The first is the
3:19free one; it’s completely free, and
3:20you can use it with limited chats on
3:22the base model. The main limitations
3:24you have are on voice chats, and
3:25therefore on transcription, which is
3:27one of the most demanding parts for the
3:28system. Image generation is much slower
3:31and more limited, as is the use of more
3:33advanced models. Honestly, it's fine
3:36for a quick test, but I would move
3:37immediately to the next level, which is
3:39the Go plan. As of today, it costs €
3:418 per month. You have more messages
3:43with tools, more uploads, more
3:44possibilities to generate images, but
3:46there is one thing that is often not
3:47clear. In this plan, there is
3:49advertising, like the ads you see for
3:51using Google for free. Here, too, you
3:53might get someone’s ads. Then you
3:55move up to the next step, which is Plus
3:57, € 23 per month as of today, and
3:59I’ll tell you right now, it’s fine
4:02for 99%of people. Here you have all the
4:04models, including the latest ones,
4:05image generation, memory, extended
4:07search; there is no advertising, so for
4:08me, that’s the one to choose. The
4:10next model is the Pro, which lets you
4:12choose between usage 5 times greater
4:14than Plus for € 103, or 20 times
4:16greater for about € 200. In this, it
4:18is aligned with Claude. It only makes
4:20sense to switch to this plan if you
4:22burn through your weekly limits like
4:23three times a week; otherwise, forget
4:25about it. And the real criterion for
4:27choosing which subscription to have for
4:29any AI system is just one. It's not
4:31what Giuseppe says, it's not the
4:32marketing; it's simply your daily usage
4:34. If you hit your usage limit several
4:37times a day or several times a week,
4:38that is the signal that you need to
4:40move up to the next level. Furthermore,
4:42another thing almost nobody mentions is
4:44that if you are a freelancer or a
4:45company like me, you just enter your
4:47VAT number and you don't pay VAT. Then
4:49it has to be recorded via reverse
4:50charge, but that’s the accountant's
4:52business. Now let's get practical, and
4:54the first thing is what 90%of people
4:56skip, which is why they get mediocre
4:58answers. Okay? The first thing you
5:00absolutely must do is log in. Okay, it
5:02seems trivial, but a lot of people try
5:03it and say, "Ah, it works like crap."
5:05Well, you didn't log in, there's no
5:07memory, nobody remembers who you are,
5:08so what do you expect? You can tell it,
5:11"Find me the best restaurants in Milan,
5:12" but if you've logged in, it remembers
5:14which restaurant you chose, and this
5:15becomes very useful later on—we'll
5:17talk about that in a moment. The second
5:19thing to do after logging in is to go
5:22to settings, personalization, and
5:24enable memory. Memory that can then be
5:26managed via the "manage" button. In
5:28English—in Italian it will be "
5:29organize,""control," I have no idea, I
5:31only use it in English. And here, basic
5:33memory about who we are, what we do,
5:35and what we like is collected. We can
5:39also ask it to update it directly via
5:40this form, but the coolest thing is
5:42that we can also ask it to update it
5:43during chats. We can tell it, "Remember
5:46that I like summaries in this format.""
5:49Remember that I like to eat pizza at a
5:51specific restaurant," remember,
5:53remember, remember; a series of layers
5:55about us are built up, and this allows
5:57us to work much faster, to get much
5:58more accurate results, and above all,
6:00to waste much less time. And linked to
6:04the memory is its opposite, which is
6:06temporary or anonymous chats, which are
6:08activated by clicking on this little
6:09icon at the top right. This is very
6:12important because when you ask
6:14something that has absolutely nothing
6:16to do with your daily work or what you
6:18do—perhaps you're asking something
6:20for a friend—it is crucial to use
6:21temporary or anonymous chat. This way,
6:24the searches, what you do, and what you
6:26create won't end up in your memory, and
6:28you won't in any way contaminate or
6:30modify the standard behavior it has
6:31toward you. And there is another very
6:34cool little feature. When you go into
6:37anonymous chat, you can ask it to use
6:39it in a personalized way, meaning it
6:41uses your memory to answer you but
6:42doesn't save the output to memory, or
6:44use it in a completely depersonalized
6:46way—so it uses no memory, plugins, or
6:48anything—a completely anonymous and
6:50neutral chat. The last setting you
6:53shouldn't miss is connecting your apps
6:55via a feature called plugins, and it
6:58covers practically every app you can
7:00possibly imagine that you use. Think of
7:03Gmail, Google Drive, Notion, Granola,
7:05Slack, Google Calendar, GitHub,
7:07practically anything. The most
7:10important ones for me are the
7:11email-related apps; for instance, I
7:13have connected Gmail, and this way,
7:15from my chat, I can say "check the
7:16latest emails received." Connecting
7:20apps allows us to open our entire
7:22digital world to our AI system. The
7:26last two things to learn about the
7:27interface are model selection via this
7:29button. You can choose the latest, most
7:33powerful one today, which is CG GPT6
7:34Astra, and it's the one that consumes
7:36usage limits most rapidly and takes the
7:38longest to respond. And you can also
7:41choose the previous model, in this case
7:435.6 Sol, which works perfectly for the
7:45vast majority of daily tasks. If you
7:48need to do something specific or highly
7:49elaborate with complex documents, it is
7:51better to choose the latest, most
7:52powerful one, but for everything else,
7:54the one in the middle is practically
7:55always fine. The other selection you
7:57can make is the usage level, meaning
7:59how much time the model spends
8:01reasoning on the task we have given it.
8:04I usually use either high or extra-high
8:06; I almost never use it because it's
8:08useless, and often the answers, in my
8:10opinion, aren't as accurate as those
8:12below. I tend to avoid using instant
8:15unless the questions are truly trivial.
8:17So, high or extra-high is the perfect
8:20sweet spot. The last thing on the
8:22interface is the plus button. With plus
8:25, you can add files, search the web,
8:27select deep research—we will see what
8:29these functions are later in the video
8:31—select a connector, drag photos, in
8:33short, it's what allows us to add
8:35resources to the chat. And let me tell
8:38you something, with these four simple
8:39basics you're already ahead of 90%of
8:41the people who have been using ChatGPT
8:43for months. Look, it has already
8:44changed your life. And if so far today
8:46I've found something that makes you
8:47think, you know, I'll save half an hour
8:49on Monday morning? You earned that half
8:51hour in these first 3 minutes of the
8:53video, so remember to subscribe to the
8:55channel because I do this every week,
8:57showing practical tools to work without
8:59going crazy and without
9:00incomprehensible ultra-technical stuff.
9:03That said, let's get to one of my
9:04favorite parts, the part that makes the
9:05whole video worth it, because the only
9:07thing you're missing now is, what the
9:08hell do I write in there? Because the
9:11difference between "hey, AI is cool"
9:13and "it changed the way I live and work
9:16" lies entirely in five simple lines,
9:18and this method counts more than the
9:20tool itself, and the method consists of
9:23line 1: stage. You have to tell it what
9:26outfit, what hat, what role it is
9:28setting and using at this moment.
9:30Meaning: you are an Italian tax
9:32consultant, you are an expert LinkedIn
9:34copywriter, you are a corporate law
9:36attorney—no, don't do that, call me
9:38—, you are a YouTube strategist. This
9:41is useful sometimes, because the moment
9:43you give it a role and establish that,
9:45practically everything changes: the
9:46tone of the answer, the output, the
9:48depth of the reasoning, the tools used
9:50by the system. Line number two, you
9:52have to tell it the task, you have to
9:54give it the assignment, and you must do
9:55it with a strong verb. So not "please
9:58help me do this," but write, analyze,
10:00act; you have to tell it what to do,
10:03you shouldn't beg it. Line number three
10:06: you have to give it the rules, you
10:07have to tell it what limits its work
10:09must have. Meaning: formal but not
10:11rigid tone, maximum eight lines, no
10:13beating around the bush, no emojis,
10:14only scientific sources. Because if you
10:17don't tell it the form of the answer
10:18you want, it decides for itself and it
10:20usually decides poorly. Line number
10:22four, the audience, meaning you have to
10:24tell it who that result is intended for
10:26—a random reader, you, a technician,
10:28your boss—because the exact same
10:30answer written for two different
10:31recipients has a different form. It's a
10:33bit like when we write to someone; it
10:35changes based on the person we are
10:37addressing. Line number five, ask me
10:39questions, and literally I write it
10:41like this: "If you are missing any
10:42information to proceed, ask me first."
10:44This is the wild card line. This is the
10:46one that changes everything. It's
10:47basically the wild card that saves you
10:49from a crappy response, it saves you
10:51from having to keep prompting— "
10:53prompting" is such an ugly word—and
10:55keep asking for changes. Because, in
10:57reality, the system is great at asking
10:58you questions; it's designed to ask you
11:00questions: it asks you three targeted
11:01ones, you give it the answers, and you
11:03get the right output on the first try.
11:04You save time and tokens, and therefore
11:06, money. And I'll show you right away
11:09how this system works with a complex
11:10example: where, if I had simply told it
11:12"write a job posting for my team that
11:14needs to do this," it would have been a
11:16disaster. Instead, we phrase it like
11:19this: "You are a senior recruiter
11:20specializing in small Italian companies
11:22.""Write a job posting to hire a new
11:24person for my team.""It must be
11:26concrete, direct, professional but
11:28human, maximum 300 words, without
11:30corporate jargon, and without emojis.""
11:32The posting is aimed at talented people
11:33who are already working and whom I need
11:35to convince to seriously consider this
11:36opportunity.""If you are missing
11:38information to write a truly effective
11:39ad, ask me the necessary questions
11:41first." Here, in my opinion, you've
11:42already figured out that some
11:43information is missing, and let's see
11:45what the model responds. The first
11:46super concrete question is: "What role
11:48are you hiring for?" I didn't tell it
11:51who I'm hiring. And so it rightly asks
11:53—instead of making a generic ad—
11:54what is the main result this person
11:56needs to deliver? What are the four or
11:58six concrete activities? The
11:59intelligent level: do I want them
12:01junior, senior, experienced? Then where
12:03and how they work. In my Milan office,
12:05in my Alessandra studio: remote, hybrid
12:07, in-office days, contract and
12:08compensation, characteristics. In short
12:09, it asked for everything needed, and
12:11so, by dictating it (this dictation
12:12feature is really convenient), I
12:14respond with: I'm looking for an
12:15operations manager for a digital
12:17content production company; I need to
12:18coordinate a team of six, manage
12:20deadlines, KPIs, and weekly meetings,
12:21hybrid work from Milan, 3 days in
12:23office, I want at least 3 years of
12:24experience. Salary 35-45,000 euros. The
12:27main advantage is that they will have a
12:28lot of autonomy and will be able to
12:29build the company's operational
12:30processes from scratch. And bam,
12:32there's the perfect ad. In one go,
12:35without telling it things 150 times, "
12:37Oh, no, change it there, change that,
12:38add this other thing." Perfect. So we
12:41were ahead of 90%of people, now we are
12:44ahead of 95%of people, moving towards
12:46100%. Let's look at a feature that
12:50almost no one explains well, but it's
12:52the one that makes you stop using
12:53ChatGPT as a simple chatbot and turns
12:55it into an incredible productivity
12:57system: the "Projects" feature. What is
13:00a project? In a nutshell, it is simply
13:02a folder where ChatGPT keeps everything
13:04related to that specific activity, the
13:06instructions that always apply, the
13:08documents to consult, and all the chats
13:10about that topic; every time you open
13:12it, it's ready—you don't have to
13:14re-explain anything to anyone ever
13:15again. Okay? How do you create a
13:18project? You go next to projects, press
13:20the plus sign, and you can choose an
13:22icon. In this case, it will be this
13:24nice little blue book, and for me, the
13:26project will be Readings 2027. Once the
13:29project is created, we go to
13:30instructions, click on the three dots
13:32for Project Settings, provide the
13:34instructions, and in my case, they will
13:36be these. Act as my personal writing
13:39advisor. When I ask you what to read,
13:41first use the information contained in
13:42the files and other chats within this
13:43project. Do not recommend books that
13:45have already been read as a first
13:46choice. I prefer concrete, rigorous,
13:48and applicable books. In short, I am
13:49giving it the perimeter and the
13:51instructions that I don't have to
13:52repeat every time, within which it must
13:54operate. I can choose whether to have
13:55it use the default memory, the one that
13:57knows a lot about me. Or a memory
13:59specific to this single project. This
14:02can be important for certain work
14:03things where I want to somehow
14:05compartmentalize what I am doing from
14:06the rest of my ChatGPT. At this point,
14:09I save, and within the Resources
14:11section, I go to add any PDFs, lists,
14:13instructions, or connectors—so Slack,
14:15Google Drive, in short, everything this
14:18project needs. In this case, I will add
14:21this reading list, like "already read,
14:24do not propose," and a backlog to
14:25evaluate where I tell you things I
14:27liked; this way, I have completely
14:29customized this workspace. So I open a
14:33new chat and simply tell it, "What
14:35should I read next?" And while it
14:38answers, I remind you that everything
14:40you are seeing in this video—the
14:42prompts, the grid, how to build
14:44projects, and the activities to set up
14:46ChatGPT—can be found completely for
14:48free in the PDF in the description. And
14:50the answer is: "I would read' Amp It Up
14:52'by Frank Slootman." It’s about a
14:55growing team, the need for real
14:57ownership, KPIs, higher standards,
14:58operational pace, and people who take
15:00the reins without depending on you—
15:02which is my life right now as I build
15:04the team for YouTube and my media
15:05company. So yes, I’m actually going
15:08to read this. And moving on to more
15:11complex tasks, maybe reviewing a
15:12document, checking a contract, drafts,
15:14or who knows, slides for a presentation
15:17, the thing I've noticed that wastes
15:18the most time with artificial
15:20intelligence isn't the prompt, isn't
15:22choosing the model, and it isn't even
15:24reading the answer and steering the
15:26system elsewhere, that is, directing it
15:28toward the result we want. One of the
15:32most underrated features is the one
15:34that allows us to work directly on the
15:36document and stop copy-pasting
15:38responses into the chat. And the tool
15:41consists of so-called writing blocks,
15:43which allows us to correct what the
15:45chat is doing directly. I’ll show you
15:47. We tell it to write a short
15:49introduction to my YouTube channel to
15:50send to a company interested in
15:52sponsoring it. About 200 words,
15:53professional but personal tone, and
15:55we’ll put it in an editable document.
15:57Okay, it prepares the document within
15:59the chat. We can select a piece, click
16:02on ask for changes—since you
16:04highlighted it—and tell it I want it
16:06less self-congratulatory. We have
16:09suggestions, we can select the next
16:12part and tell it we want it for a tech
16:15company. In short, piece by piece, we
16:17can guide it through the edits to the
16:19document. Think about documents of 10,
16:2120, or 30 pages; how this simplifies
16:23our lives—we’ve gotten good at this
16:25so far, but we want to become super.
16:28And so I'll show you three advanced,
16:30yet still very simple, functions. The
16:32first is Deep Research, it allows you
16:34to do in-depth research and get an
16:35ultra-organized document on a specific
16:37topic. For example, I asked it to
16:40compare the main newsletter platforms
16:42available in 2026 for an Italian
16:44creator. It analyzes prices,
16:45automations, deliverability,
16:47segmentation, and ease of use, and it
16:49gave me an amazing result; most of all,
16:51it creates an editable plan by working
16:53on multiple sources, returning a
16:54structured report with citations as
16:56well. Instead of firing off an
16:58immediate answer, it takes all the time
16:59it needs to find sources, reorganize
17:01them, and process them. Above all, you
17:02can even tell it what kind of sources
17:04you want. Next function, we need our
17:06phone. This is something that works on
17:08mobile, it works perfectly. It’s
17:11something I love, especially while
17:12driving, because at least I don't use
17:13the phone (which is forbidden, never
17:15use it), and it’s the advanced voice
17:16feature. And so you can tell it: "Hi, I
17:19have to record a video in 10 minutes,
17:20be my sparring partner, ask me one
17:22question at a time to see if the video
17:24thesis is clear: I'm about to talk
17:25about how to use ChatGPT from scratch."
17:27"If I ramble, interrupt me."
17:29>> Exactly. The really important part is
17:32that it interrupts you, meaning it
17:34doesn't wait for you to finish your
17:36whole speech to answer; it interrupts
17:38and gets you back on track.
17:39>> So are you interrupting me? Are you
17:41really interrupting me?
17:42>> Exactly. Yes, yes. And that's exactly
17:45how it works. I mean, the point here is
17:48>> The cool thing is that it has become
17:49truly natural; you talk to it as if you
17:51were talking to someone, so if I'm
17:53speaking, it interrupts me, okay? Let's
17:55move on. It's amazing. The other super
17:58cool feature with the phone concerns
18:00the camera. So, I can take, for example
18:03, this Stanley bottle, a gift from my
18:05wife, go to the camera, take a photo of
18:07it, and tell it: "Describe this image
18:09for me," as if I were giving
18:11instructions to an image generator:
18:13composition, framing, light, colors,
18:15materials, atmosphere. What is it? It
18:19tells us it's a tumbler, it describes
18:21it for the image generator. Very cool.
18:24And at this point, we tell it, "Perfect
18:26, generate an image of the same bottle
18:29in a promotional advertising context
18:31with two or three different versions
18:33for an Instagram ad." And bam! Here is
18:37the result. So, effectively, ChatGPT
18:40becomes a tool that understands the
18:42world around us. So we've become power
18:44users, but now the most important point
18:46of the video arrives, which is the main
18:48mistakes when using ChatGPT. And the
18:50first mistake is the dry, generic
18:52question. The one without the framework
18:54, without the five levels I told you
18:56about earlier. If you use that, it
18:57completely changes how you use it. It
18:59takes you 20 seconds and then it
19:00becomes practically automatic, I assure
19:02you. Error number two is infinite chats
19:04. These artificial intelligence systems
19:06have a thing called context, which is
19:08the limit in terms of tokens, therefore
19:11messages exchanged, beyond which they
19:13get confused because they are
19:14objectively not hyper-powerful or
19:16hyper-unlimited. The solution is
19:18projects, so if you have to talk about
19:20the same topic multiple times, start
19:22new chats within the project. This way
19:24you keep the chats lean and have a
19:26clean context. Error number three, not
19:28activating memory and not connecting
19:30apps. In that case, it's a tool that
19:32knows nothing about you, it's useless,
19:33it can't really be of help to you in
19:35practice. Error number four, always
19:37using the same model. More powerful
19:39model equals more complex tasks, but
19:40also more time and more usage. For a
19:42simple question, use the default model.
19:45And so now that you've discovered how
19:47to use ChatGPT as a real super pro user
19:49, the next question is: "But if I need
19:51to do more complex things, which tool
19:52should I use?" I'll explain it to you
19:55in this video on the right, where I
19:57show you how to use Cloud Code from
19:59scratch, the tool that will completely
20:01change the way you see artificial
20:02intelligence. Thank you for watching
20:04and unleash your potential, always.
20:06What?

About this transcript

This page contains the full transcript of Dammi 20 Minuti e ti Insegno ChatGPT da Zero by Giuseppe Castagna, generated from the public captions YouTube serves with the video. The transcript has 4,085 words across 614 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.