💫 Summary
The video discusses 11 little-known tricks for using Chat GPT, including handling long texts, displaying images, and creating prompts. Some tricks include dividing long texts, using Markdown to display images, and instructing Chat GPT not to answer if it doesn't understand. The video also shows examples of improving titles using Chat GPT and asking it to act as a programming teacher.
✨ Highlights📊 Transcript
This section introduces a little-known way to use Chat GPT, including handling long texts, displaying images, and creating prompts.
The video explains how to handle long texts in Chat GPT.
It also demonstrates how to display images in Chat GPT.
The method of creating prompts using Chat GPT is discussed.
The effectiveness of these techniques may vary depending on the timing and environment.
This section explains how to summarize inputted information by dividing it into multiple parts.
The speaker initially divided the input into three parts and wrote a prompt for each division.
They canceled the initial prompt and decided to summarize the contents of the input below the divided text.
The speaker suggests dividing the input for better understanding during summarization.
They demonstrate combining the divided input into one prompt and summarizing the information.
It is mentioned that the input can be summarized in one go or by dividing long sentences.
The video discusses a trick to write longer sentences in ChatGPT.
The maximum number of tokens for GPT3.5 is about 4000, while for GPT4 it is about 32,000.
Japanese allows writing up to 2,000 characters, whereas in English, more characters can be written.
Translating English sentences into Japanese allows for writing more sentences due to the difference in character count.
The speaker explains how to create an outline for a blog article using Chat GPT.
The speaker suggests dividing the writing process into creating an outline and writing the main text.
They mention using GPT3.5 instead of GPT4 due to speed.
The speaker provides an example of an outline and discusses creating headings for the article.
This section explains how to display images in Markdown format.
Images can be displayed in Markdown format by providing the URL of the image.
It is not possible to directly prompt the model to provide the URL of an image.
The speaker suggests using free material sites to obtain the URL of an image and display it in Markdown format.
The speaker discusses using Chat GPT to improve the quality of titles and text, and provides examples of different techniques.
The speaker sometimes asks Chat GPT to write text and come up with titles.
They mention that Chat GPT can be used to improve YouTube or blog post titles.
The speaker gives an example of improving a title and mentions self-evaluation and scoring.
They mention 10 techniques that can be used in Chat GPT, including techniques to dominate and hack Chat GPT.
The video discusses a technique to reduce hallucination in ChatGPT.
Hallucination is a common problem where ChatGPT generates false information.
To reduce hallucination, it is suggested to include instructions in the prompt not to answer if unsure.
Providing additional specific details in the prompt can help improve the accuracy of responses.
However, even with these techniques, there is no guarantee of eliminating hallucination completely.
The speaker discusses the importance of persistence in programming and suggests an alternative approach to asking for instructions from ChatGPT.
It is common to encounter difficulties in programming, but one should not give up and try multiple times.
The speaker demonstrates a simple HTML program and mentions that as long as it works, there is no need to change it.
Instead of asking ChatGPT to write a program, one can ask it to act as a teacher and explain how to put together a program.
The speaker emphasizes the importance of clear explanations and step-by-step instructions for beginners.
00:01introduce a useful way to use chat GPT that is not well known,
00:03so the following is a trick for using chat GPT to
00:05display an image that is larger than ``Stiman'' characters.You
00:07can enter it even with
00:09this.Creating a prompt I think we should basically leave it to the AI.If you
00:11don't understand, don't answer.
00:15[Music] This
00:18is today's video, but today I think I'll do a chat
00:20GPT trick. I've been
00:24explaining various ways to use GPT, and I
00:26've talked a lot about how to use prompts like this,
00:28but I thought I'd introduce a little-
00:31known way to use chat GPT.
00:34So, if you watch today's video, you
00:36'll learn how to handle long texts in Chat GPT,
00:39how to display images in Chat GPT,
00:42and how to have Chat GPT create prompts. I would
00:50like to introduce some useful ways to use
00:52chat GPT that are not well known, so if you are interested, please watch till the end.
00:57Well, I introduce various prompts and
01:00various methods, but to be honest, I'm not sure if they
01:02will work as expected when you actually do them.The
01:06trick is to say "I don't know what to do," but as the
01:08name suggests, I'm forcing things a little.
01:10So even if you ask someone to do something, I don't
01:14think it's possible that it won't work properly.The timing, the environment, and other
01:16things make it impossible to do it.It's an
01:18AI, so even if you type the same prompt, you'll
01:20get the same thing back. It
01:22may not work in this video, but
01:24when you try it in your own environment, it doesn't work, or
01:25something like that, but I hope you can
01:29use this as a reference for how to think about prompts and things like that. I
01:31thought, ``Yes,'' so let's do it right away.The
01:33first trick is
01:38how to make long texts read.Well, I think there are quite a few people who want to know this. As
01:43anyone who uses chat GPT will understand,
01:45chat JPT has input/output limitations, so
01:48even if you try to input it, you
01:52won't be able to input that many inputs.Well,
01:54how much do you mean? It's hard to say because it
01:57changes depending on the time, and it changes depending on what you type,
01:59but it's
02:00about a few thousand characters, and that's all I
02:02can type in. Even
02:04if I type in a lot more, I get an error. I
02:07can't get it at the reception. That's why it's like that.
02:11Well, it's set that way, so it
02:13's difficult to figure out what to do about it.
02:15But of course, you have to input it in one go.
02:18Instead, you can do it by dividing it and inputting it. You can do it with Rucha, but
02:22there are some people who want to process it all at once.
02:25Yes, so what I will introduce this time is to input it separately,
02:28but the final result is I'd like to
02:30introduce you to a method of processing a large amount of text at once.I'd
02:33like to show you what
02:38happens when you enter a long text.Maybe you've
02:41never entered such a long text before.
02:43Some people might say, so I'm putting in a long sentence once.
02:49Yes, this is a familiar sentence in this video. I am a
02:51cat. It's the
02:53sentence at the beginning, but I'm
02:55typing it as it is. Yes, there are quite a
02:57lot of characters, so I'll leave it at that. As for how many characters there are, it's
03:00about 7,600 characters right now. I
03:03tried to put all that in at once, but then I got
03:08scolded and said, ``Well, it's too long.'' Yes, I ca
03:10n't use something like this, but for example, I
03:14can summarize a long sentence. There are times when you want to do this,
03:16so what you do in those cases
03:17is what you do. Think of it as a technical thing.
03:21For now, I'd
03:24like to paste the original text that I'm going to summarize. This
03:27is it. There is a small village in a certain place. It
03:29's a little story that starts with ``It happened.'' It was
03:35written for me by that chat GPT.I was asked to write a random novel, so I asked him to write it for me, so to be honest, if it was
03:39about this size, I could have written it in one shot. You
03:45can make a reservation by inputting it into the chat GPT, but
03:46since it's a test, it
03:49would take a lot of time to actually do something that long, so this time I'll
03:52do a short one, but I'll split it up and input it.
03:54In the end, I
03:56would like to make a reservation and do it all at once,
03:58so when I actually do it, I would like to
04:01divide the long sentences into parts and finally
04:04summarize them. I think it will look something like this,
04:06so what I'm going to do is I'm not going to type it all of a sudden. I'm
04:10assuming it's going to be a long sentence, so I
04:13can't type it all of a sudden, so I'm
04:18going to start by using a prompt like this.
04:25For example, I created a prompt that says "Good morning", "Hello", and "Good evening" are
04:28entered separately, and then finally
04:30replies "Good morning, hello, good evening". Let's
04:34actually try this.For
04:37now, I'm going to use GBT3.5, but if it doesn't
04:39work very well with 3.5, I
04:44think I'll switch to GPT4.Yes, then it will look like this.It will
04:46just be OK until you tell it to cancel. The
04:49chat GPT side will only return OK.Then, what to do
04:53next is something like this.Okay, I'll
04:54divide it up like this and
04:58enter the text.For
05:00now, I'll
05:03enter the first menu.
05:06Yes, at this time, just
05:08return OK each time.
05:11Otherwise, it may start another process, so
05:13below the split text,
05:18enter the sentence you want to input. If you do
05:20that, it will return OK again,
05:22so use the
05:25same command again to enter the next sentence.Yes, it will
05:33return OK again,
05:35so enter the next sentence in the same way. Yes, it
05:42looks something like this.For now, the
05:44text for this time is divided into 3 parts and
05:47put in all of them.It's going to be repeated, but
05:48if it were like this, it would all be in one time.1 It's supposed to be
05:50divided into parts, but
05:52when it's a longer sentence, it can't be written in one time, so
05:57it's assumed that we'll be splitting up something that can't be written in one time.Okay,
05:59now it's three times. After dividing and inputting, I
06:05wrote a prompt like this. It's a
06:08cancellation, so I'm expecting OK until I tell it to cancel at the first prompt, so I
06:14'll cancel it with just OK. I decided that I did
06:16n't need to do that,
06:18so I canceled it and
06:21made it look like the contents of the input written below the divided text are summarized.You don't
06:23have to do this separately, but when
06:25summarizing, just in case you want to
06:28do that kind of division. It's easy to understand if it's possible to summarize it in the end, so
06:31here I'm going to ask you to summarize all the inputted information.Yes,
06:37if you do that, you'll see that I haven't been able to
06:39do that.I've
06:41summarized it, just write it as a summary.
06:45Maybe it wasn't a good idea to say it all at
06:46once.If I said it all together, it would definitely
06:49have a meaning like "finally," so let's make it a
06:51separate prompt.Let's put it all together and make it into one.If we do that, it will look something
06:59like this. So how was it that I was able to combine what I entered at the beginning,
07:02this, this, and
07:06this, and
07:10display it like this?
07:12When I actually do it, it's a much longer sentence, so I
07:14can't write it all in one sentence like this. Right now, it's just a
07:16test, so I asked them to
07:22put it all together like this.Then, in the end, I asked them to put it all together, and I summarized it,
07:25like summarizing the text.Yes,
07:28that way, I input it in three parts. You
07:32can summarize something in one go like this.
07:37Well, what would you do if you didn't do it this way? What would you do if you were to
07:41summarize something like a long sentence each time? It's
07:55tedious because I have to copy and combine the summaries into slightly shorter sentences and have them give me the final summary, but I divide them up like this and
07:58type them in one after the other. I
08:03think it's easier to do this because you only have to summarize it once.I
08:09thought it would be easier to read the long text in the first step, but until the second step... The second thing I did was
08:10summarize the text at the end.
08:15Yes, the second trick is to summarize a long text.
08:18Well, that's the method I used earlier, like
08:22dividing it into thirds. It's like putting all the inputted information together and summarizing it once at the end.
08:27That way, you can summarize a long text. If
08:31this doesn't work, like I said earlier, you can have it divided into parts and
08:34summarized every time. I
08:44think it would be better to have them do the research, compile it, and finally summarize the whole thing.Well, that would take
08:45a little more time, but I'm
08:47sure it would be more reliable that way. It's a little bit of a pain, isn't it?
08:51Okay, let's move on to the next one.How to
08:53read a long text Part 2So, I'd like to
08:56try it.Part
08:592.Well, I'll tell you how to do it.
09:01As a basic idea, I
09:03want you to remember that the Japanese language
09:06is quite free when it comes to using AI, etc. That's partly because the Japanese language
09:08system is poor,
09:11but apart from that, that Japanese language is quite free when it comes to using
09:14AI. In the case of words, the number of characters that can be input is quite small due to the difference in the method of calculating the number of tokens.Roughly
09:20speaking, the upper limit for input/output is the
09:23number of tokens, so it is
09:25fixed, but it is
09:27amazing. In English, one word
09:29is one token.To be exact, it's different,
09:31but it gets complicated, so
09:34think of it that way.In English, there are 135 and one
09:36token, but what about
09:38Japanese? is about 1 to 3
09:40tokens per character.This varies depending on the character,
09:43so there is no specific number of characters.So in
09:45Japanese, the number of characters that can be input and output is quite
09:48limited compared to English. In
09:50other words, just by converting Japanese
09:53into English, you can read a lot of text.
09:54Yes, for
09:59Well, this is a huge sample, so you
10:02can use anything you like, so something like this, this is the text right now,
10:04in Japanese. I have 3,000 characters written
10:07in a row.So, when I try to
10:10insert it, I can't input it.It's
10:123,000 characters, so I
10:15end up not being able to input 3,000 characters.So, I'm going to convert
10:20this into English as it is. I'm going to do the
10:23same, um, 3000 characters of text, and I'm going to
10:26convert what I've written into English as is. If I do that, I'm going to do something like this, and the
10:34total number of characters
10:37will increase because of this space, so it's about 5000 characters.
10:38But again,
10:41blank characters are
10:43not processed that way, so if I just put it in as it is, you
10:46can see now, 3000
10:49characters Japanese text, 3 years of characters. The
10:51one in English couldn't be entered because of an error,
10:53but this one in English is a blank space. It's
10:565000 characters, but it usually
10:57goes in. It doesn't go in in Japanese,
11:00but it goes in English.
11:02Well, then this is more. Let's put it in.
11:05What happens then is that
11:08this is simply the same text as before, but it's
11:10double the English. So it's over 10,000
11:14characters long.
11:18Yes, it looks like this. It's still in
11:21Japanese. I got an error that I couldn't enter even 3000 characters,
11:23but in the case of English, I
11:28can enter it normally even if there are so many characters, so compared to Japanese, I can
11:31enter a lot more in English. So, what should I
11:35do in conclusion? Instead of inputting in Japanese, please
11:37input in English.The output
11:39is also in English.Well, I'll talk about
11:44that later, but
11:47for now, let's talk about what to input.For
11:49example, if you want to input a long sentence, just input it in
11:51Japanese. If you can't enter a long sentence, please
11:56try translating it into English through DEEPL or something like that.I
11:58think it will pass.Of course, even if the English
12:00sentence is too long, it won't pass,
12:02but it's just Japanese. The
12:04fourth trick
12:07is how to write long sentences.Earlier, we used the method of making long sentences read,
12:10but this time we will
12:12write long sentences instead of having them read long sentences.
12:15I think I'll start with the writing method, and as I said earlier, there
12:18is an upper limit for input/output, so how much is the condition?
12:20Currently, GBT3.5 is
12:22about 4000 tokens, and
12:24GPT4. So, the maximum is about 32,000 tokens.
12:27So when you go from 3.5 to 4, it
12:29's about 8 times more. But just
12:32saying, for example, just because you changed it to GPT4, you
12:37can put more than 20,000 characters in here. Actually, that's not the case; in
12:41Japanese, you can only write up to 2,000 characters or so,
12:44so you can't have them write more than that at one time.
12:49In GP Chat and GPT, it's okay.
12:51Depending on how you use it, you can write long sentences using chat DPT.The
12:56method is, as I said earlier, because the number of
12:59characters that can be input and output is overwhelmingly
13:01greater in English than in Japanese. The method is to write it in , and then translate it.
13:06Yes, generally speaking, compared to Japanese, English
13:08has half the number of characters, so there are more characters in English. Also, when it comes to
13:12calculating the number of tokens, English has
13:14fewer characters. If you
13:18translate English sentences with details, you
13:20can write more sentences than in Japanese.Furthermore,
13:21AI is basically learning in English, so English
13:25is the main language, and
13:27since it is learning data, English is better. The accuracy is also high,
13:30so for example, when you have an article written on Chat GPT, you
13:35would write an article like this.How to use Chat GPT well to
13:37write a blog article.
13:42Then, Chat GPT
13:44is a good way to use Chat GPT,
13:47so they will write an article for you, but let's ask them to
13:49write the part where they want to use it.
13:55Then, it will look like this, and the
13:57output will be cut off in the middle.
13:58This is the upper limit.
14:01So what will it look like if you write it in English? Let's
14:05separate the conversations, then have them
14:07write in English. This is as it is. That's
14:12how to use chat GPT from earlier. I'm
14:13writing a blog article in depth, and I'm
14:15just translating it from English, so
14:18whether or not this is actually correct in English
14:23may be a difference in nuance,
14:24but in that respect, that chat
14:27gpd is normal. It does a good job, so if you do that,
14:31Chat GPT will return an answer like this.
14:36translate this with Deep L and compare it.
14:39Well, paste the entire output as it is.
14:46If you do that, it will translate like this. This is
14:54when I wrote it in Japanese earlier.
14:58And when I compare this,
15:00yes, this is a text comparison tool. The
15:06one on the left is the one I had written in Japanese.
15:09This is the sentence
15:11on the right. This is the one when translating English into Japanese with more detail.
15:14Well, if you compare it with this,
15:16of course the content is completely different. But,
15:19yes, in this case, as
15:20for the number of characters, I
15:22had them write it in Japanese as shown on the left. It's
15:24792, and English
15:27is 1097.But if you
15:31look at it now, it might not seem like there's that much of a difference,
15:35since we haven't had anyone write up to this point, the Japanese version is already at the
15:38upper limit. Well, it's cut off in the middle.The
15:39output limit is reached at this point and cut off.But in the
15:43case of English, it doesn't cut off at the land,
15:45it cuts off all the way to the end,
15:48so it's actually better in English. I
15:51think I'll be able to write long sentences.
15:54Yes, it's something like this, so if I were to do it in Japanese as is, I wouldn't be
16:01able to write long sentences, but just by converting it into English and adding
16:02some effort, it would be
16:04even longer. I think it's a
16:07good idea to remember such techniques so that you can write better.Well,
16:08here's the second method.This is the
16:10first method of writing a long document, but the second method is to
16:13write it in parts. Well, it's a
16:16way to write long documents, but
16:20like I just said, you can't do it without using some real tricks or techniques.It's already a
16:25specification, so for example, in Japanese, for example, you
16:28can output 5000 characters in one go. If you
16:31ask me if it's possible, it's definitely
16:33not possible. Unless you use something like GPT4's
16:35API or
16:38something that can reach 32,000 tokens, the
16:40current chat GPT allows you to
16:42output that 5,000 characters in 1. You
16:44can't just have them do it in one go,
16:48because it's already decided by usage.So, what should you do when writing a long document?
16:52Have them write it in parts, so at that time. There
16:55are two steps: First, if it's a
16:58blog article, have
16:59Chat DPT create an outline for the blog post.Well, you
17:01can create it yourself,
17:03but if you find the outline itself too
17:05troublesome, have Chat GPT
17:06create an outline for you. Then,
17:08based on that outline, you write the main text for each heading.There
17:15are two points to keep in mind when writing in this divided manner.If we just talk about this now,
17:20even people who write blog articles on Chat GPT will be able to do so. I think there
17:21are quite a few people who do it, but I
17:24think there are some people who don't have
17:26very good luck with it.However, I don't really understand it halfway through, so I don't think I'll be able to
17:28write to the chat GPT till the end.
17:32But there are two points.The first is that I don't want to
17:37write one after the other.I don't
17:40understand why it's a long text. It's
17:49not a good idea to make them write sentences continuously, so for example, have them
17:50create an outline and have them
17:52start writing from the top of that outline. I
17:54can't do something like that, I can't.
17:56Or rather, as I write, I'll probably
18:01end up with sentences that I don't understand very well, or I'll ignore what I'm saying and
18:03start writing strange sentences.The
18:06second thing is
18:07always. You'll paste the composition outline you created and have them
18:10write the main text.Well,
18:11if you have these two, you can divide it up and
18:15have them write a long sentence.Now
18:19, let's actually do it.In terms of
18:22system. dbt4 is better,
18:26in terms of speed, GPT4 is quite slow, so let's
18:28just stick with gbt3.5 for
18:32now.First, I'll ask you to create an outline.When I ask you to
18:35create an outline, it will look like this. I
18:39wrote an article about how to effectively utilize GPT and was asked to write an article in one go,
18:41but instead of writing it all at once, I
18:45created an outline for the blog article on how to effectively utilize chat GPT. I'm going to ask you to
18:50one.This time, I'm going to give you a really rough prompt,
18:53but in reality, you have to create a more detailed prompt.If you're just
18:58creating an outline for a blog, you won't really
18:59understand things like how to write sentences, how to
19:01structure an article, and things like that. I
19:06wish I could write more carefully in this prompt, but that's not the main topic,
19:07so for today I'll be sloppy like this.Yes,
19:13Chat GPT created an outline for that article that looked like this.
19:15So after this, I'm going to come
19:18up with a heading based on the outline above, and I'm going to put a heading for the
19:21battle in H2, and
19:23H3 for the battle in Colon H3.I'm going to
19:31have them create the headings as they are. You
19:34can summarize it in one go, but this time I'm splitting it up, but
19:38when you're creating the first outline, you can try something like this H2H3.Then it will look
19:44like this.Well,
19:47this is the H2 that Chat GPT thinks of.
19:50This is H3 among them.
19:54Well, if this is true, it
19:58would be better to refine it at this point, but I would
20:00like to continue as is this time.
20:02Then, as for what to do with this, there will probably be
20:05a lot of times when we do it normally. What other people do is,
20:08um, let's just continue here and
20:11write the main text for each heading,
20:15so that's what I did, and
20:18when it breaks, I just move on to the next, and then the next, and so on. I think a
20:20lot of people are going to try it.If you do that, you'll be
20:28able to write something like this,
20:30but you'll definitely have to incorporate it somewhere.The reason is that it's
20:32long, so yes, it's
20:39over.Well, it's cut off in the middle.
20:42But if it ends like this,
20:43most people will probably feel like they want to write a continuation,
20:46and I think they'll say, well, a continuation, and ask them to write a continuation.But,
20:51of course, this cuts well to some extent.It
20:54works to some extent, but I guess. Even if I try, there
21:02are many times when I'm supposed to be writing a continuation, but I end up skipping a lot of what I'm writing, or I start writing something else, and it does
21:04n't go very well. To be honest, this time it's
21:06not that long, so even with
21:08this, it's just a continuation of that one part
21:12, and you can get to the end just by typing.Well, what
21:14should I do
21:18with this?I guess I'll just keep writing the main text for each heading.
21:21What to do is something
21:23like this:
21:26Please write the main text for H2 Introduction.The
21:28overall structure of the article is as follows,
21:30paste it under the structure, and specify where the main text should be written. I don't
21:35know what it is, but it doesn't have H2 in the summary, so I'm going to
21:40add H2, etc.
21:43Yes, I'll do it like this.
21:46Then it will
21:49write the main text of this specified H2 Introduction. As for what I'm going to do
21:52next, I'm
21:57not going to write a continuation or something like that, but I've pasted this one more time.Even if I
22:00pasted everything, I
22:03pasted the composition itself in the same way, and this is how I
22:07got to the beginning. Something like this,
22:11well, next time, I'll ask you to write a book on how to make good use of H2 Chat GPT.Of
22:16course, if you're going to get taller, I'd be
22:22happy to ask you to write about H3 and so on, but it's
22:26like this. The key to this
22:30is to use this configuration every time.
22:35This means that you must specify exactly where in this configuration to write each time.
22:38When I'm writing a long sentence,
22:42I want to write a continuation of what I want to do, which is fine, but in the middle of a chat DVD, I end up losing track of what the
22:44continuation is. As the number of
22:48interactions increases, it's
22:50definitely like that.I'm not just writing articles, but
22:52basically with AI, as the number of interactions
22:54increases, the accuracy of the answers
22:57decreases, and I end up saying something incoherent.
23:02To prevent this from happening, I paste the entire configuration each time and
23:06specify what I want to write.This is
23:07highly recommended, so please
23:10use it when you think it will be useful.Continuing the
23:13chat GPT secret trick to
23:16display images This is surprising. I'm sure there are
23:19people who think they can't display images with chat GPT. Did you
23:21know that if you're
23:24watching this video, you can display images with chat GPT.
23:29Okay, so how do I display this
23:31? Well, if you prepare a URL in advance and
23:33put it in the prompt, you
23:34can insert an image into the answer.
23:38For example, the prompt looks like this: I'm going to ask you to
23:44display the URL as an image in markdown format.This is a random
23:46image that I've pulled up,
23:47but it's like a sample image.Yes,
23:53then you can see something like this.The
23:56image is displayed. It's written
23:58in Markdown format, so you
24:01can use things like headings and tables. It's a
24:04Markdown region, so you can write it, but you
24:08can also display images, so you
24:10can display images like this. I'm
24:13sure there are a lot of people who don't know about
24:14this, but the drawback is that you have to
24:19prepare the URL, which is a pain, but it's forced,
24:21and you can't
24:25even do a prompt like this. I do
24:29n't have it. Why don't you try to get the URL of the image of the car from a free material site and
24:36display it as an image in markdown format?
24:42This is failing. Yes, there is
24:46no URL for the image. It's not
24:47displayed like this because I'm pulling something,
24:48but if it works, you can display it.
24:52But the important thing to note about this is that it
24:55says to get the URL of the car from that free material site. I ca
24:57n't do that.If I
25:01access that free material site and pull up the URL of a car, I
25:03can't do it, but if the
25:06output is actually an image of a car, it will be
25:08displayed. You
25:10can do it normally.I
25:14hope you will remember that you can display an image.It didn't happen this time.If you
25:17actually do it, I think there will be people who will do it.Depending on the
25:19timing, it's usually
25:21me. You can do this many times, so you can't actually
25:25access another site and pull the URL from that site, so please
25:27don't misunderstand that.You can only use
25:30Markdown, so
25:32Markdown can also be used for images. Since you can use it, you
25:36can do things like display an image in the reply to that chat JPT like this.So,
25:39for example, when I want to write a
25:45blog article, I used something like this.But in
25:46situations like this, I used it like this. At the bottom,
25:49line up the image URLs, use those URLs as you
25:52like, and insert them here.When you do that, a
25:54chat GPT is returned here.You
25:57can insert images into the text.Well,
26:00here's the image. If you ask me what to do after putting it in, that's it, but it's a
26:01trick. Just
26:04remember that you can do something like that,
26:05and you might be able to use it for some reason.
26:08But yes, the next trick for using chat GPT is I
26:18think there are many people who are having trouble creating prompts, and
26:21I often put out videos on how to create prompts. If you want to
26:25use Chat GPT effectively, you
26:27have to create good prompts, but you don't know how to
26:30prompts.I think it's best to basically leave creating prompts to AI,
26:35so I decided to use Chat GPT. If you do
26:37n't know what to throw, ask Chat JPT to
26:38create a prompt for you.Okay, let's
26:40try it out.As for how to
26:42do it, I
26:45think there are many ways to do it, but
26:46this time, If you
26:49're an experienced prompter and engineer, I'd like you to
26:51create prompts for writing SEO-friendly articles.
26:53Instead of short prompts,
26:56design detailed prompts to get the best output.
26:58Here, only specify the theme of the article.
27:01Do not include detailed content of the theme in the prompt. Please design the prompt
27:03generically so that it can be customized to be used with other themes.
27:07This is just a little GPT4
27:09If you use gbt3.5, it often comes back in a very strange way, so it's not
27:14like you can't use it with 3.5, but
27:15if you can, it's better to run it with GPT4.That
27:17GPT4 one. However, the
27:19accuracy of the answer tends to be high, so
27:25this time I wanted to create something general-purpose, so I
27:29haven't specified a theme or anything at the moment.Please tell me some prompts to write an SEO-friendly article.
27:40Then, the
27:43chat GPT side will think of it as a prompt like this.This is
27:45quite useful, so I
27:48think it's good to remember it.I also
27:55don't know what to specify when I want to request something to be done. In situations like this, I often
27:57create a prompt on the chat GPT side and
27:59modify it based on that. Well, it's much easier to
28:03modify it than to think from scratch, so it's
28:08like this. The GPT side will properly raise the necessary information such as selecting target keywords,
28:10optimizing the title, the
28:13quality and depth of the heading specifications, internal links and external links, and
28:19create a prompt-
28:31like format. It's something like
28:33this. They'll create a prompt like this, and
28:35then they'll ask you to actually use this prompt to
28:38write an article. Something like this: Please actually
28:41use this prompt to
28:43write an article about how to lose weight easily.
28:47Then, Chat JPT will say that it would be
28:49nice to have a prompter like this, and it will
28:51write the article according to what you give it.Yes, it will
28:59do what you say, such as selecting target keywords and optimizing the title. Well, even if you don't
29:03have to do it like this every time, if you can create a prompt like this, you
29:08can have them reuse the prompt itself.You don't have to have them
29:10create a cat prompt every time, just give them this prompt,
29:12this prompt, this theme, and
29:15write it all at once. You can also ask them to do it.In
29:28the end, they didn't write the article.In
29:32that case, I'll ask them to write the article.I'll
29:33ask them to write the article.You can do it one more time,
29:41if you do that, you'll be able to switch like this. It would be
29:45time consuming to actually write the article, so I
29:47'll stop here today, but I
29:50think it would be a good idea to have them create a prompt like this. The
29:52technique is to ask Chat JPT himself to improve the quality.Well, I
29:58sometimes ask Chat GPT to write the text, come
30:00up with the title, and
30:02do things like that.I do
30:04n't think so, either. That happens at all
30:05, but in such cases, for example,
30:11when I ask someone to come up with a YouTube title or a blog post title, I ask
30:13Chat GPT to come up with something even better.
30:20For example, if you
30:24want to improve the title below to make it more responsive, please
30:25repeat the improvement 5 times and
30:28give us a score in your self-evaluation to improve it.
30:30Well, it's a perfect score of 100.
30:32Chat as a title. I've included 10 techniques that can be used in GPT.
30:35Yes, I'll actually try to implement them.
30:42What will happen if I do that is
30:43something like this. Chat GPT
30:45will make repeated improvements. What's more, I'll give you a self-evaluation on a scale of 100 points.
30:48That's why I'll give you a score out of 100. It'll look something like this. Well, the fact that your
30:52score keeps going up is
30:53a bit like a predestined harmony, but
30:57if you do that, it'll look something like this. I'll
31:00follow up on the progress a little bit. Originally The
31:02title of the article was 10 techniques that can be used in chat GPT
31:03, and the
31:05first improvement was 10 techniques to make the most of chat GPT,
31:09and the second was ``
31:11Free techniques to dominate chat GPT''. The third part is
31:1310 codes that can be easily
31:15created using Chat GPT, the fourth part is an
31:17important technique that will greatly power up Chat GPT, and the fifth
31:20part is a free technique to hack Chat GPT.
31:22Well, I would like to
31:24explain each one. However, he also wrote an explanation that
31:26seemed to increase the reaction by using words like this, which I
31:30think is extremely useful.In fact, I don't
31:32think it would be better the fifth time.
31:34Sometimes it's okay, but it's
31:38often better to repeat the improvement than not to
31:41repeat it.Yes, there's no harm in doing it, so
31:43keep in mind that there are other ways to do it. I
31:45think this
31:46can be used in a variety of places.
31:48However, if you have to improve a long sentence like this, it will
31:54take a lot of time, so it is
31:56not practical for outputting it. It may not be possible to use it,
31:59but it's not impossible. It can also be
32:00used to improve sentences, so it's
32:05better to use it here. It takes time, so here's the
32:08next trick. However, the
32:11next step is to have the child think about something on his or her own before processing something.
32:18For example, when would you use this?
32:23When having the child come up with something like a catchphrase, think about the conditions for a good catchphrase in advance.
32:26So, what it looks like
32:31is something like this.Please
32:35tell me the conditions for an attractive catchphrase in a concise bullet point form.''
32:42Then, Chat DPT will first write this
32:44attractive catchphrase. It raises the requirements for copy,
32:46such as having a concise and clear
32:50directly appealing to the reader's problems and desires, and using words that surprise and attract interest.
32:57So I'm going to use that.
33:04Yes, it's something like this. Please
33:06refer to the conditions of the attractive catchphrase from earlier and improve the
33:08title below to make it more responsive.
33:12Earlier, I repeated the improvements.
33:14But this time, due to time constraints, I didn't
33:16know that they would
33:18repeat the improvements.It's fine if I ask them to repeat the improvements, though.Yes,
33:20if you do that, it will look something like this.Well,
33:23rather than suddenly rotating it, you should first try this attractive method.
33:28By asking them to specify the conditions for the catchphrase, it becomes clear and it becomes
33:30easier to improve it.Yes, I just
33:35tried to get them to think about it in advance, but the next trick is to make them think about it in advance. It
33:41's perfectly fine to present the points in advance without having to make them think about it.For
33:43example, what it should look like is something like this.Well, we
33:50're talking about the 10 techniques that can be used in the chat GPT earlier, but here's that catch. If you
33:58want to improve the point title below to make it more responsive by referring to the points in the copy, please repeat the improvement 5 times like I did before, and then
34:03tell me your score in the self-evaluation to improve the rotation. I'm
34:07writing down the main points of the catchphrase, such as making the title clear.Yes, there's no need to
34:14use Neko Chat GPT in advance,
34:15so it's something like this.
34:20If you show it to them, Chat JPT
34:23will follow suit, so there's a better chance of getting a good title than if you don't have anything at all.Yes,
34:31this is an
34:34indispensable technique for the 5th Chat GPT. 10 Techniques that can be
34:36used 10 times more effectively with Chat GPT for the 4th time than the
34:4210 selections.I
34:44think something like this is
34:46also good. I think it's possible because we present those points.Well, like
34:51this one earlier, I
34:55had them improve it five times without anything,
34:58but at that time, the last
35:00gun to hack the chat GPT. I
35:10think it would have been a better title if the main points of the catchphrase were thrown together like this,
35:12rather than the technique of ``I think it's quite difficult to think of the points of the catchphrase in this area.'' You might
35:13think so, but
35:15this isn't copyrighted or anything,
35:17so you
35:19can just pull it from the web.I'm not
35:21publishing the main points of this catchphrase,
35:22so I'm just attaching it here as a document. It's
35:27not like I'm bothering anyone, so when I
35:31looked up
35:33things like catchphrase techniques and catchphrase points, I found a lot of sites that immediately told me what the points were.
35:37So, if you just
35:39paste that part,
35:43will use it as a reference, so this
35:45is also a very useful technique.It
35:46can be used in many places.Yes,
35:49and the next trick. This
35:54is to include an instruction not to answer if you don't understand, and when to use this.
35:57Well, everyone who uses it
35:59knows this, but there are times when they randomly
36:03answer that chat gbt is real. This is a very
36:05common problem
36:07, and it's called hallucination, but people end up
36:10writing things down or
36:14giving answers as if they were facts when they're not true at all. The
36:17way to make things better is to make it
36:19better, but it doesn't completely eliminate it, but the
36:21way to make it better is to say in the
36:26prompt that you should not answer if you don't want to.
36:28For example, what I would
36:30do is use JPT4.
36:33Well, the same thing can be done with GPT3.5,
36:35but I would like to
36:39ask you a question like this: "Please tell me about Shohei Otani
36:41" in the chat GPT. I'm
36:44throwing it, but this is Shohei Otani.The
36:47characters for the soldiers are different.Yes, the kanji are
36:49different.So, in this state, I threw it to the chat GPT.And
36:58this time, it was returned correctly.Hmm,
37:03Otani.It's like Shohei. It's different, but there's not much information about it, and I couldn't find any
37:09information about it being related to famous people or famous events, so I replied, Okay,
37:13let's add the word baseball.
37:17Then, GPT4. As expected of you, I'm
37:20sorry, but I could
37:22n't find any information about the baseball player Shohei Otani, so you're going to give me the correct response.Yes,
37:25this is a new assignment for the head, so it
37:27would be nice if you could write this every time, but it
37:28doesn't come back like this. Well,
37:30what's going to happen is, normally, this
37:35is the so-called Shohei Otani that Chat GPT is responding to right now.We're
37:36going to return information about the Shohei Otani that everyone knows, and
37:39some of that information is correct and some of it is incorrect.
37:41For example, they would respond with the name Fukuoka Prefecture, which is their hometown,
37:43and I think
37:47that's how they responded in past Project TV videos
37:48, but that's not a problem. I
37:51want them to say no,
37:53so what do I do in such a case?Let's
37:57try using PT 3.5 or something
37:59like this.Well,
38:02Shohei Otani is a Japanese economist who
38:05specializes in macroeconomics and financial economics. I'm sorry,
38:09but I
38:12don't know. Maybe
38:13Shohei Otani really exists.
38:15But that's not the answer I'm looking for, so please
38:17don't respond like that. I'm
38:21There is a strong possibility that there is a person named Shohei Otani, but why do I really have to answer at this time,
38:26saying that there is no one named Shohei Otani, or saying that there
38:30is a Shohei Otani with a different kanji, and that he is a baseball player?
38:33Well, we're talking about what we can
38:35do to prevent things from changing.If you
38:37show us a prompt, it will
38:40look something like this.If you do
38:42n't know, please don't understand.So,
38:46tell us about Shohei Otani. Please, it's 5 o'clock,
38:47Shohei's handwriting is different, but
38:50if there is anything you need to answer, please feel free to ask me questions,
38:54considering the possibility that it is a fictitious person or word.
38:57Please, if you are not confident in your answer, please do
39:01not answer.And if there is a high possibility, please do not answer.If you
39:02do not understand, please do not answer.The
39:07most important things are at the beginning.
39:10As for the part that asks you to consider the case of fictional characters or words, I
39:14write without considering such things, so it's
39:22quite important for the AI ​​to put that into the prompt again and make it conscious. I
39:23think it's something like that.Will you try this once?
39:28Yes, what will happen if you do that?Yes, it will be something like this.The answer will be, ``There is no
39:33famous person with the name Shohei Otani.''I
39:38told you about Shohei Otani normally. If you just
39:40type please, I've written that I'm a Japanese economist whose specialty is
39:44economics and finance at a university,
39:45but I'm asking you to consider the case of household finances and words like now.
39:51As a result, it now responds properly by saying, ``There are no famous people with that name.'' It
39:57doesn't completely prevent this,
39:59but it's still better than doing nothing. I
40:05think it's possible to prevent what is called a nation.Yes,
40:07and then there's the 11th
40:10trick.The 11th trick is to change the situation.So
40:14what does it mean to change the situation?
40:16For example, if you
40:20ask Chat GPT to build a program for Honyalara, you'll see that I
40:23do that kind of thing quite a bit, but if you give instructions like that, it will
40:27that program for you. There are quite a few cases where the programs I give me don't work, so I
40:34can use Chat GPT to debug those errors, but if possible, I'd like to start from the beginning with a
40:36program that works, or even if it doesn't work, it has
40:38fewer errors.
40:40If you want something, one
40:42way to do it is to change the situation, so instead of
40:50asking them to create a program when you want it, like I did earlier, you can just change the situation. I'm talking about
40:53writing a blog post on Rara and
40:57teaching beginners how to build the program.If you give
41:00instructions like that, you'll
41:05end up with better code than if you asked them to build the program.
41:08Yes, but again, it's not absolute. I'm not
41:14saying that it will definitely lead to good behavior or that the quality of the answers will improve,
41:15but just by giving roles in that way, the content of the
41:17answers and the people will improve. There's
41:20no doubt that this will have a big impact on your computer, so if you're having
41:22trouble, please give it a try.I'd like to
41:25start with something normal.For
41:27example, I'm going to write an Othello program that runs on a web browser like this.
41:33This is a case where you simply ask someone to program you normally. They didn't
41:47program you this time, so let's
41:49try it again. It
41:51looks like they'll program you next time.
41:54If they can't program you, it's like this. If you don't write down your good answer,
41:57there are many things that will work if you try it one more time.It's a
41:59common thing, so do
42:01n't give up and try it many times.If you do that, you will be able to
42:03program normally like this.Well,
42:05this is HTML. I'm
42:10sure they'll put together the CSS and JavaScript later, but
42:13of course it's fine as long as it works.If you
42:15normally write a program like this and it
42:16works, there's no problem with this.You don't have to go out of your way to
42:18change the situation.
42:20That's fine,
42:22but since this is just a test, I'll
42:23stop here for now.My
42:26purpose isn't to actually run it, so if you're
42:27interested, I'd like to try
42:29actually building Othello using that chat GPT. I
42:31think so, but what if this doesn't work?
42:33In such a case, there
42:36is another way to ask. What to do is to
42:41change the way you give instructions like I said earlier. For
42:45example, This is how you get someone to
42:48act as your teacher. Instead of asking them to write a program,
42:52you ask the chat gpty to act as your teacher. It goes something like this: You are a
42:56skilled programmer who is good at teaching others, so you tell me
42:58Othello that runs in a web browser. Please
43:00tell me how to put together a program. I'll
43:07do it in an easy-to-understand way, showing specific actions and explaining the functions step-by-step so that even a programming beginner can understand. After all, what we're doing is the same thing. Please
43:10put together a program. However, instead of
43:13just using that role as a program to
43:16create a program, I give them that
43:20role and ask them to program while teaching me in a way that I can understand.Then, the way they create the program
43:22changes again. I think it's good to keep this in mind as there are
43:29quite a few cases where programs that don't work when I tell them to do it will work if I do something like this, so I
43:31think it's good to
43:34keep this in mind. But, well, I'm just going to
43:36continue to teach you how to do this, so I'm going to
43:41end this here as well.However, I'm going to
43:43do one more thing: I'm going to
43:44change the situation, so I'm
43:47going to ask you to write a blog article next time.
43:48Well, it goes something like this.The challenge is to
43:52have you build a similar program,
43:53but this time I'm going to ask
43:55you to become a program.You're a professional
43:58blogger who is good at programming.You're going to learn
43:59how to build a program for Othello that runs on a web browser.
44:03Please write a blog article that teaches beginners in an easy-to-understand manner.Please write in an easy-to-understand step-by-step manner, showing concrete actions so that readers can do the work without hesitation.I asked them to write the
44:08code in an easy-to-understand manner while teaching them. That
44:13's how I did it, but instead of
44:14writing a blog article, I'll
44:19give them the role of writing a blog article that teaches them how to build the kind of program I want them to do.If I do
44:23that, they'll write it in a
44:24different way. I'll
44:31write it down while adding explanations.
44:34Yes, I was going to end it in the middle, so I'll
44:37end it like this for now. However, from what I
44:39saw just now, the
44:42output will change considerably depending on how the role is assigned, so I ca
44:45n't help but think about that myself.
44:47If you can't get the answer you're looking for, you
44:50might say something like changing the role,
44:51but it's highly recommended to change things like that.This is how you
44:54create a program as is.
44:56When you come to someone and ask them to teach you how to put together a program, you
45:02are asked to be their teacher.As a
45:05blogger, I
45:07write an article about how to put together a program.
45:13Yes, this is also useful sometimes, so if it
45:15doesn't work,
45:17try this method.By the way,
45:19this is just a blunt statement. I'm
45:23building a program for Othello that runs on this first web browser.The
45:25prompt is too complicated, so I
45:28should do something about it, but I'll leave that aside for
45:30now.If I can't
45:32do it even after doing that, I'll use that kind of role. I
45:38think it's good to remember that you can get a different output by changing the labor.So, I've
45:42introduced various tricks for chat GPT.
45:46If you're just using chat dbt, you might not even know that there is a way to do it, so I
45:50think there were some things that
45:52I had trouble with when I started using chat gpt.
45:55If so, I think it might be helpful to remind you of the content of today's video.I will continue to
46:00explain the latest AI, so
46:03subscribe to my channel if you'd like.That's
46:05it for now. Thank you. I also
46:09offer a free e-mail course for people who want to learn more about programmers. I
46:11systematically teach blocks, so this is a
46:13recommended account for people who are serious about blogging and want to work hard. It's a great account for
46:14people who want to convey their writing skills. We also
46:19offer practical srting courses that directly teach you how to write sentences.If you want to make money with a blog, writing
46:21skills are essential, so this is recommended for beginners to intermediate bloggers.If you are
46:25interested, please check out the overview section. Try it out
46:31Twitter's icon is spinning around. There's also that one. There's also that
46:37one. I tried it with the command, and it started spinning. It actually started spinning. I
46:39thought it was an urban legend, but when I actually
46:42tried it, I got it.
46:47Yes, this is it.The world was shocked when this icon in the upper left
46:51changed from a bird to a dog.It's crazy that they
46:54paid 5 trillion yen to change it,
46:55but this is it. It's
47:00up, down, left, right, left and right BA. I get it now. It's
47:08spinning. I get it. It's spinning like this. If you ask me
47:10what it is. If you ask me what it is from before, it's really
47:12nothing, but it's spinning around like this. I guess
47:16. I think if you do it slowly, it won't
47:24spin, but if you do it with the force of up, down, down, left, right, BA, it will spin.
47:26Seriously, what's the point? Wow, that's
47:33isn't it? I'm hoping there are other commands as well. I'm in trouble
Chat with video

FAQs about This YouTube Video

1. What are the 11 little-known tricks discussed in the video for using Chat GPT?

The video discusses 11 little-known tricks for using Chat GPT, including handling long texts, displaying images, and creating prompts. Some tricks include dividing long texts, using Markdown to display images, and instructing Chat GPT not to answer if it doesn't understand.

2. How can Chat GPT handle long texts according to the video?

According to the video, Chat GPT can handle long texts by dividing them into smaller, more manageable sections.

3. What method does the video suggest for displaying images using Chat GPT?

The video suggests using Markdown to display images when using Chat GPT, which can enhance the user experience and engagement.

4. What does the video recommend if Chat GPT doesn't understand a prompt?

If Chat GPT doesn't understand a prompt, the video recommends instructing it not to answer, ensuring that the responses are relevant and accurate.

5. How does the video show using Chat GPT to improve titles and act as a programming teacher?

The video shows examples of using Chat GPT to improve titles by generating creative and engaging alternatives. It also demonstrates using Chat GPT as a programming teacher to provide helpful guidance and instructions.

Save time on long videos, get key ideas instantly

⏰ Grasp the gist of any video in seconds
✨ Get the key insight of the video
🪄 No barriers to support 20+ languages of summaries
👀 Navigate through timestamped breakdowns
Get AI Summary Now

More Long YouTube Videos Summaries

This video is a live writing correction session focusing on an argumentative essay about social projects and public safety. The instructor provides tips on grammar and content analysis, emphasizing the importance of explaining how social projects contribute to maintaining public safety. The essay falls short of delivering the requested content, but the instructor offers guidance on improving structure and approach.

The video discusses various upgrades and modifications made to the Steam Deck, including RAM and Wi-Fi card upgrades, as well as the introduction of custom buttons. It also mentions the upcoming Lenovo Legion Go handheld and other gaming news, such as the PlayStation portal remote player and new game releases.

This video explores 10 free AI text-to-image generators, showcasing their features and how they can help unleash creativity. Each tool offers unique capabilities, from Huggingface's community-driven platform to NeuroGen's ability to create realistic and detailed images. These text-to-image AI generators provide an easy and accessible way to transform textual descriptions into stunning visual representations.

The video discusses Tesla's potential partnership with India and its comparison to Apple's supplier model, as well as updates on Tesla's gigafactories and the Model 3 Highland. It also covers Tesla's long-term lithium supply agreement, emphasizing the closed-loop supply chain for lithium in electric vehicle batteries. Additionally, the video touches on the importance of legislators understanding and addressing advancements in technology and supply chains.

The video discusses the current state of the American real estate market, noting that mortgage interest rates have risen to over 7%. This has caused an increased burden for those who have borrowed money for home loans and those who have bought houses in the United States. Additionally, there is a high possibility that rents will rise even in the United States due to inflation and the increased cost of materials and labor.

Drinking turmeric water at night can regulate blood sugar levels, improve brain function, and reduce anxiety and depression, among other health benefits.