Don't Use ChatGPT Until You Watch This Video

Summary

本视频提供了一系列实用建议，旨在帮助观众更有效地使用ChatGPT。从利用自定义设置以保存个人偏好，到教授ChatGPT以用户自己的写作风格，再到如何精简回复、使用自我批评和自我提示技巧，视频涵盖了多种技巧以提升与ChatGPT的互动效率。视频还探讨了如何指定输出格式，包括文本、表格或HTML，以适应不同的使用场景。通过这些策略，观众可以避免长篇大论的回复，直接获得所需信息，从而节省时间并提高生产力。

Takeaways

• 😀 利用自定义指令可以使ChatGPT更好地了解你的需求，从而提供更加个性化的回答。
• 📘 如果你是特定行业的专业人士，比如老师、学生、会计或律师，告知ChatGPT你的职业可以获得更相关的回答。
• 🌍 告诉ChatGPT你所在的地区，可以帮助它提供与你地理位置相关的信息。
• 📝 通过设定回答的语气（正式或非正式）、立场（有意见或中立）和长度（长或短），可以获得更符合你期望的回答。
• 🔧 对于像Excel公式这样的具体请求，可以指定只给出最有效的解决方案而不需要额外解释。
• 🔍 要求ChatGPT在提供包含事实的答案时给出信心水平，并附上来源URL，以便获取更准确的信息。
• ✍️ 可以通过提供写作样本来教ChatGPT以你的风格写作，提高文本的个性化程度。
• 📊 使用自我批评功能，让ChatGPT分析并改进其文本，提高内容的质量。
• 📧 利用自我提示功能，可以优化ChatGPT的提示，获得更加精准的输出结果。
• 📐 明确指定输出格式（如文本、表格、CSV、HTML等），可以使得ChatGPT生成的内容更加适合你的使用场景。

Q & A

ChatGPT的自定义指令有什么作用?

自定义指令可以让用户向ChatGPT提供有关自己的信息,以获得更加相关和有用的回复。例如可以告知职业、所在地区等,以便ChatGPT给出适合该背景的建议。

如何让ChatGPT的回复更加简洁?

可以在提示中指定文字长度上限,或者事后让ChatGPT用更简洁的语句重新表述已给出的回复。

ChatGPT的自我批评功能有何用处?

这个功能可以让ChatGPT审视并提出改进自己给出回复的建议,对于进一步优化回复质量很有帮助。

ChatGPT支持哪些输出格式?

ChatGPT支持多种格式输出,包括纯文本、表格、CSV、HTML、JSON、XML、Pandas等。可根据实际需求选择合适的输出格式。

如何教ChatGPT模仿自己的写作风格?

可以提供若干自己文章的示例,让ChatGPT分析撰写风格,并要求其后续内容使用学习到的风格。

什么是ChatGPT的自我提示技巧?

即让ChatGPT自己给出最优提示词,以获得某一特定领域的最佳回复。这可以大大提高ChatGPT回复的质量。

如何判断ChatGPT回复的可信度?

可以要求其声明回复中的事实性内容可信度,并给出支持性来源链接。对于推测性内容,也要求其明确声明。

视频中提到的使ChatGPT发挥最大效能的关键是什么?

关键在于输入质量,也就是“废料输入产生废料输出”。必须注意提供高质量的提示,才能获得高质量的回复。

ChatGPT的自我提示可以采取哪些方式?

例如让其提出问题以确保完全理解所需;让其给出多个候选提示并解释优劣;让其对自己给出的最佳提示进行解释等。

ChatGPT的回复过于冗长该如何处理?

可以事先或事后指定文字长度上限;可以让其用更简洁的话重述已给出的内容等。这都可以避免获得冗长或不必要的解释。

Outlines

00:00

😊 自定义设置让交互更顺畅

本段介绍如何利用 ChatGPT 的自定义设置功能,让对话更加顺畅高效。可以设置自己的身份背景、交互语气、回复长度等,避免不必要的重复和冗长。

05:04

😃 PROVIDE TEXT SAMPLES 让写作更具个性

通过提供文字样例,可以让 ChatGPT 学习并模拟你独特的写作风格。你可以指定一个别名来保存风格,之后让 ChatGPT 用该风格写作就可以了。

Keywords

💡自定义指令

您可以为 ChatGPT 设置自定义指令和偏好,以便它能够根据您的需求更好地回复。这可以让对话更加个性化,避免重复设置。视频中提到的例子包括告知职业、地区等信息。

💡语气和风格

在自定义指令中,您可以指定想要 ChatGPT 使用的语气和风格。比如友好、正式或是简洁。这可以让回复更符合您的要求。视频举了一些设置语气和风格的例子。

💡置信度

让 ChatGPT 总是提供答案的置信度。这对检查事实性答案是否可靠很有用。视频中演示了如何要求 ChatGPT 提供答案的置信度和引用来源。

💡引用来源

如果 ChatGPT 的答案中包含事实性内容,可以要求它提供可靠的引用链接。这可以验证答案的准确性。视频中给出了 FIFA 世界杯获胜队伍的例子。

💡编写样式

可以提供文字示例,教 ChatGPT 学习并模仿您的写作风格。之后让它用这种风格书写文章。视频中演示了这种方式来编写个人博客文章。

💡自我批评

让 ChatGPT 分析和批评它自己的文字,指出可以改进的地方。然后修改文字。视频中 ChatGPT 对SUMMARY作了自我分析和提高。

💡自我催生

让 ChatGPT 为自己设计最优提示,以获取最好的答复。视频中给出了让它写 5 个 Email 提示的例子。之后选择最优提示来获得定制化的 Email。

💡长度限制

可以事先或后期要求 ChatGPT 限制答复的长度,使其更简洁。视频中要求答复控制在 500 个词以内。也可让它逐步减少已给出答复的长度。

💡输出格式

ChatGPT 的答复不限于纯文字,可以指定表格、CSV、HTML 等格式。之后直接使用这些格式。视频中演示了设定并应用这些格式的方法。

Highlights

建立自定义说明可针对性地提供有用信息

明确说明ChatGPT的回复风格可以减少不必要的来回

提供自身背景信息让ChatGPT提供更相关的回复

指定回复的语言风格可以得到更符合预期的内容

要求ChatGPT提供回复的信心水平可以提高准确性

要求提供引用来源可验证ChatGPT回复的真实性

提供写作样例能让ChatGPT模拟自身写作风格

让ChatGPT自我批评可以改进创作内容

自我提示可以产生更优化、定制化的指示

询问最佳提示的原因可以理解提示的有效性

指定最大字数可以得到更加简洁的回复

指定不同的输出格式可以得到定制的内容

HTML格式输出可以在网站上展示内容

优化提示可以得到更完善的指定格式输出

输入决定输出,明确高质量的提示十分重要

Transcripts

By now, you've probably started using ChatGPT,  

right? It's not rocket science. Anyone can  ask questions, and it's going to give you  

answers. In school, we learned that there  are no bad questions. In ChatGPT world,  

there are. It's not going to tell you that, but it's  going to give you poor quality answers,  

and you're just going to waste a lot of time going  back and forth. I've been there—super annoyed—so  

I decided to spend some time to find the best  prompts to get the most out of ChatGPT.

My first tip for you is to take advantage of  custom instructions, which you're going to find  

by clicking on your profile icon. So, when you go  to custom instructions, you have the ability to  

hand over a note to ChatGPT that explains who you  are, what you want, and how you want ChatGPT 

to respond. This way, you won't have to repeat  your preferences in every single conversation.

So, the first question is: what would  you like ChatGPT to know about you  

to provide better responses? It helps if you  can provide some context about yourself. So,  

for example, if you're a teacher,  let it know. If you're a student,  

an accountant, a lawyer, let it know. This  way, you can get responses that are more  

relevant to your line of work. If you want  responses that are relevant to your region,  

tell it where you live. Not the exact address,  but just the area where you are. So, for example,  

you could say, "I'm a software developer based in  Chicago working with Python." Or you could say,  

"I'm a marketing professional in New York writing  advertising copy." Right, you get the idea.

The next question is: how would you like  ChatGPT to respond? We get some thought  

starters, like how formal or casual should  ChatGPT be? For example, you could put in,  

"Language and tone should be friendly and  casual." You can decide if ChatGPT can have  

opinions on topics or remain neutral, and if  responses should be long or short. And this  

can really make a difference because you can  get rid of a lot of frustrating back and forth.

So, for example, let's say you're good in  Excel and use ChatGPT to help you out when  

you get stuck, but you don't like seeing all  those explanations about the solutions. So,  

what you could write is, "When I ask for Excel  formulas, just provide the most efficient formula  

without any explanation." If you were a programmer  and you didn't want to see all those programming  

explanations, you could type in, "When I ask  you for code, just give me the most efficient  

code with code snippets without additional  explanation." Right, so that's really helpful.  

In this case, I'm picky about Excel formulas.  I'm going to leave this in and click on Save.

Now, let's go and start a new chat, and I'm  going to ask it for a formula just to make  

sure this works. Let's say I'm in Excel, and I'm  struggling with updating this formula. Currently,  

it's returning everything that's greater than  12,000, but I want to change this to be between  

12,000 and 15,000. And I have no idea  how to tell it to do it between these  

values. I'm going to go back to  ChatGPT and type in my formula,  

tell it to update this Excel formula so it returns  values in the B column that are between these two.  

And when I send this, it just provides  the solution without any explanation.

Now, without that custom setting, this is what  I would get. And sometimes you would just  

end up with a lot longer explanations  like this one. These are great if you  

don't know your way around, but if  you do, they can be pretty annoying.

Another thing that could be quite helpful is  to tell ChatGPT to always inform us about the  

confidence level of its answer. This could  be quite helpful for factual topics. Now,  

we can also expand on this and say, "When  your answer includes facts, always provide  

a valid URL with the source for your answer.  And if you speculate or predict something,  

inform me." Okay, so let's test  this out. I'm going to save this,  

go ahead and start a new chat, and let's ask  it over the FIFA World Cup winners of the '90s.

When I run this, I get: 1990: West Germany, Brazil, France.  

Confidence level is high, and I get a list  of valid URLs directly from FIFA.com. Right,  

so these settings can be really helpful.  I'm sure you're going to find them handy.

Now let's move on to prompts. The  first one is to write like you. So,  

if you ask ChatGPT to write some text for you,  the results will probably sound a bit generic,  

right? Even if you're emphasizing  the custom instructions and the  

tone and the style that you want, it might not  properly reflect your style of writing. Now,  

the good news is that you can teach it to write  in your own style by giving it some examples. So,  

first, we're going to explain to ChatGPT  what we're going to do with this prompt.

"I'd like you to help me write articles from  my productivity blog." Just replace this with  

whatever type of blog or article you need.  "First, I want you to understand my writing  

style based on examples that I give you. You'll  save my writing style under 'LG_STYLE.'" Now,  

this makes it easier to refer to later. "After  that, you'll ask me what the topic of my specific  

content is. You'll then write the article  using LG_STYLE." Okay, so let's give it a try.

Okay, so it understands what we're trying to do  and it's ready for some examples. I'm just going  

to go and grab some copy from my website. Let's  copy this paragraph from the About page and paste  

it in as example one of LG_STYLE. Now I'll give  it a second example of LG_STYLE. I'll just go  

and grab the other copy from here, and paste it  in. So, it summarized my style as informative,  

personable, and aims to establish a connection.  And now it's ready to write our content. So,  

I want to write an article about the  importance of daily coffee for productivity.  

Cool, start writing it in my personal writing  style, the one that it previously saved. Now,  

you can, of course, continue working  on this and make it better. You can  

also come back to this chat and ask  it to write other related articles.

And by the way, if you love your daily coffee,  subscribe to this channel because we all love  

coffee around here. Next up is self-critic.  So, another great option is to ask ChatGPT to  

review its own text and provide feedback. Now,  it sounds funny, but it really works well. So,  

let's say I asked ChatGPT to provide me with a summary  of why Python and Excel can work well together and  

it comes up with this reply which I'm not really  happy with. So, I'm going to ask it to act as a  

critic. Be ruthless, analyze the text, and tell me  where it can be better. It will go over the reply  

and provide step-by-step feedback on potential  issues. For example, with clarity and how adding  

specific examples could make the summary better,  how we could highlight specific capabilities.

It also went ahead and revised this original  reply, giving examples of Python's strengths  

and Excel's strengths. From there, you can further  improve on this reply by asking it to specifically  

emphasize and include some of the previous  pointers that it gave us. It's pretty cool, right?

Next up is self-prompting. So, how about using  ChatGPT to self-prompt to optimize its own prompt?  

Here's how you can do that. So, let's say I want  to send an email to my team encouraging them to  

participate in our team-building event. To get the  perfect prompt, I could ask ChatGPT, "Write  

five perfect ChatGPT prompts that will really show  off the power of ChatGPT. Focus the prompts  

around writing an email to my team encouraging  them to participate in our team-building event.  

Before you write anything, ask me questions until  you're sure you can create the optimal prompts."

It will usually come back with some questions  about the topic, in this case, about the purpose,  

dates, and locations and other useful information.  Now, after I provide my answers, so, for example,  

for any particular incentives or benefits, I've  put "nice dinner, nice breakfast, no work,"  

then I send this off, and it creates five possible  prompts for me. Now, these are prompts that will  

help me get the best reply. So, one is, "Draft an  email to your creative team encouraging them to  

join a special team-building event. Mention the  importance of taking a break from work." Another  

one is, "Compose an email inviting your team to  a casual and fun team-building event. Stress the  

importance of building strong relationships within  the group and the opportunity for a nice dinner."  

Okay, so I like number three better. I'm just  going to tell it to use it. Now it just provides  

me with a prompt, but I actually want it to run  it. So can I do it? It starts writing the email.

Now, we can see it has a nice subject, the tone  is soft and casual, and it's highlighting the  

bonding and relationships. Right, so we  can see it sprinkled in different places,  

like we can share stories, laughter,  and delicious food. We'll have a great  

breakfast. There's no work, no deadlines.  So, it's a very customized email.

Now, compare this to a case where  I don't optimize the prompt. So,  

I tell it to write an email to my team to  participate in our team-building event. It's  

no surprise that I get a very generic email about  team-building and what generally team-building  

events are designed to do. So, as you can see,  self-prompting can give you a much better output.

But you can also take this self-prompting a step  further. So, if you go back to the step where we  

got the different prompts, we could ask it why  a prompt would work well. It gives me detailed  

information about the reasons it thinks this  prompt is effective, but I'm not so sure. So,  

I'm going to ask it, "which prompt do you find the  best?" It tells me that prompt three effectively  

combines the key elements I provided, so it  could work well in my case, which it actually did.

Often when you ask ChatGPT to write something,  the reply can be rather long. So, one way to  

avoid this is to tell it from the beginning to stick to a specific word count. Now, if this is something that you want  

in every single reply, you can add it to your  custom settings. If not, you can add it to your  

prompt. So, for example, we want to know the  advantages of using XLOOKUP over VLOOKUP in  

Excel. I'm going to add that the maximum length of  the text should be 500 words.

A few moments later.

Okay, so let's double-check. I'll copy this, go to a new page, type in "word.new" to open a blank Word document in the  

browser, and paste in the text. We can see  down here that we have 469 words. Now,  

another alternative is to tell it to reduce the  length of a text that it already gave you. So,  

a good prompt for that is, "Now, say the same  thing more concise and prefer using only 60%  

as many words or whatever percentage you need. You  could try cutting it down even more, step by step,  

until you get the crispiness that you like.  Your audience is going to be thankful for it."

Specify the output format. ChatGPT has many  different output formats, not just plain text. So,  

for example, you can tell it to give you  the response in a table format. Let's say,  

"Create a table with the winners of the FIFA  World Cups between 1990 and 2018. The headings  

should be year, winner, and runners-up." So, we  get a table. We can just highlight the content,  

use the shortcut Ctrl+C to copy it, move over to  an Excel sheet, and paste the table with Ctrl+V.  

Now, alternatively, you can tell it to output  the table as CSV. Then all you need to do is to  

copy the code from here and paste it into a text  editor, and then save it as a CSV file extension.

Now, if you need this, let's say for  your website, you could output as HTML.  

Moments later...

I'm just going to test this out. So, I'm  going to grab this code and paste it on  

our site in an HTML editor. Then, when  I go to preview, I can see the table,  

but it doesn't look really nice. I'd  rather have borders and spacing. So,  

let's improve the prompt and ask it to add  the borders and spacing.

Later that night.

Now, when I replace the code,

I have a table that's a lot easier to read. Right? So, you can also ouput in different format

like JSON or XML. Or if you need it as a Pandas data frame,  

you can do that as well. Just give it a try.

SO, these are some tips to get the most out of  ChatGPT or any large language model AI, actually.  

It really comes down to the old principle:  garbage in, garbage out. Keep this in mind,  

and you'll definitely make better use of these  new tools. I hope you found this helpful. Do  

subscribe if you aren't subscribed yet, and  I'm going to catch you in the next video.