GPT 6 + ChatGPT Work: 3套AI自動化工作流，直接拿去用！

OpenAI's recently released GPT-4o is one of their most powerful models to date, and the previously launched ChatGPT Team is the workplace tool built for non-technical users. When we combine the two, a single prompt can get it to write in your own voice and tone, create interactive visualization dashboards so you never have to look at boring Excel data again, and even have a morning briefing ready for you before you start your workday. So, in today's video, I'll combine GPT-4o and ChatGPT Team to share three use cases closely related to your daily life and work. Once you master these workflows, they can help you save several hours every single week. Besides that, I'll also explain when it's appropriate to use ChatGPT Team versus when to use the recently launched GPT-4o. Without further ado, let's get started right away. First, if you're not familiar with ChatGPT Team yet, when you open ChatGPT, there is a "Team" button at the top; just click it to get started. It functions similarly to the regular ChatGPT chat; there's an input box in the middle where you can tell ChatGPT exactly what you need it to do for you. The biggest difference between ChatGPT Team and a regular chat is that while you might use a regular chat for brainstorming or generating slide content, in ChatGPT Team, you can feed it your notes and have it create the actual slides for you. Therefore, ChatGPT Team is better suited to help you complete a task from start to finish. Additionally, if you frequently use Claude, you can simply think of ChatGPT Team as Claude Team; you can use both to help you work, but we will introduce some features that Claude Team doesn't have. The cases we demonstrate today will all be in the ChatGPT Team desktop app, so if you haven't downloaded and installed the desktop version yet, be sure to do so, as some of the features I'll show next are only available on the desktop app. When you enter ChatGPT Team, if you want to use their latest model, you can select GPT-4o right here, and you can even choose the "Thinking" level. The " Thinking" level determines how much time and effort the model spends thinking before generating a response. A higher level usually results in more comprehensive, accurate, and better-structured answers, while a lower level provides faster and more concise responses. However, keep in mind that a higher "Thinking" level consumes more tokens and at a faster rate. I should mention that since GPT-6 is inherently very smart, even if you set it to the lowest level, it remains highly intelligent. OpenAI has previously stated that GPT-6’s performance at its lowest reasoning level already surpasses the top level of previous models. Based on my recent experience, the medium setting is the best choice, balancing both performance and cost-effectiveness. In the first example, we want to have ChatGPT help us with writing; more precisely, we want it to write in our own voice. I believe GPT-6 is currently the best model for professionals, but to truly leverage it, we need to create a "Skill " that understands our writing style and can write just like us. If you're not familiar with Skills, simply put, they are a set of reusable instructions that detail the specific process for how to do something. You just tell GPT you want to create a Skill and describe the process you want to automate, and it will package it into a reusable skill for you. This way, you won't have to repeat your instructions or explain your writing style to it every single time. Whenever you want it to write like you, just enter the name of that skill in the chat, and it's ready to use. You can click the plugin icon in the top-left corner and select Skills to display all the skills you have already installed. Creating a new skill is also simple: just open a new chat window, tell it what kind of skill you want to create, and it will generate it for you. I'll give you an example: we want to create a skill called "Write Like Me." When setting up this skill, I strongly recommend adjusting its reasoning level; you can set it to high or very high. Note that you only need to make this adjustment while creating the skill, because a higher reasoning level helps the AI better understand your writing style for better results. Once the skill is created, you can switch it back when performing tasks. Next, find a few of your previous writing samples and attach them to this task. You can click the plus icon to select files or folders and upload them from your local device. Now, if you have an entire folder containing a lot of your own writing—like past emails to clients or work progress reports— and you want the AI to reference that whole folder, just create a project here and grant it access to the entire folder. Here, I’ve uploaded all the content I usually use to make videos for you, and then we use this prompt. By the way, all the prompts mentioned in today's video will be given to you for free, and you can find the download link in the description and comment section. Here, we ask GPT-W to create a skill called "Write Like Me," referencing only the samples I’ve uploaded here. Now you can see it has created this writing skill. When you want to use it, just type a slash in the dialog box followed by the skill's name, like "Write Like Me" or the English command "writelikeme," and either one will trigger it. For example , if you are using AI to help you do research, you can say, "Now, help me organize this content into a report." This way, GPT knows you don't just want a report, but a report written in your specific writing style. For our second case study, let's look at how to use GPT-4 and GPT-W to create visualizations. Interactive dashboards. The GPT-4 model is particularly well-suited for dashboards because it excels at making sense of complex data, and its designs are great. Plus, since we can use plugins like Sheets in GPT, creating these interactive, shareable dashboards is extremely convenient. For instance, here, I want GPT to take a holistic look at my YouTube data and video performance. Here is the prompt I gave it: "I want to create a dashboard specifically to showcase my YouTube data analytics." Of course, you can provide it with any data. Then, we need to tell it where this data comes from, so I told it here: "The attachment is the YouTube backend data I uploaded from May to August." Of course, you can also use plugins to pull this data directly. For example, if we search for "vidIQ" in the plugins, this plugin is specifically designed to retrieve YouTube backend data. Just click to install the plugin, log in, and you'll be able to find it directly in your plugins. Next, I’ll explain the reasoning behind this request, because when the model understands the "why," its performance improves significantly. So here, I’ll say, "The reason I’m creating this dashboard is to gain a clear understanding of my channel’s overall situation and the performance of each video, so I can better decide which types of content to prioritize." And since GPT excels at creating designs based on reference images, I suggest you find a dashboard style you like. For example, you can use a free website like Dribbble, where you just need to search, and it will help you find matching dashboard styles. Of course, I’ve found that if you search in English, you’ll get far more options than in Chinese. So, let’s type in "analytics dashboard." You can open one, copy the image, paste it into GPT, and then say, "I’ve added a screenshot of a dashboard design I like ; I hope you can follow this style and layout." Finally, before we send the prompt, there is one more thing to do, which is to add two built-in plugins. The first is the Data Analysis plugin. You can activate it by typing "data" directly in the plugin area, or by using the "@" symbol to call it up. The Data Analysis plugin allows GPT to perform better when handling data because it has built-in specialized skills for interpreting and processing information. The second one you want to use is the Synth plugin, which is mainly used to generate links we can share with others, so you can share the dashboard you’ve created. I will also be releasing dedicated tutorials on Synth in the future. Now we’ll send the prompt. We’ll use the standard GPT-4o mode here. I feel that if your prompt is good and comprehensive enough , you don't really need to use any more advanced modes. Now, as you can see, GPT has built the dashboard exactly according to the key metrics I specified earlier. It clearly presents our channel’s view counts, watch time , and the performance of each video, and the dashboard style it created perfectly matches the reference photo we uploaded. Beyond just telling us about the content, direction, and subscription rates of the videos, it also handled the data processing for us , providing both overall statistics and individual monthly performance. Honestly, if you only gave me those boring Excel spreadsheets, I definitely couldn't have pulled these metrics in such a short amount of time. Given that we only gave it a single prompt, I think the result is fantastic. Plus, because we used the Si plugin, we can easily share it with the team or just copy the link and send it to anyone else. For the final case, we can use GPT-6 to create daily morning briefings for you or your entire team. Here, we're using GPT-6's capabilities combined with a feature called " Scheduled Tasks." We can have GPT-6 automatically check the web or your inbox every morning and prepare a briefing before you and your team start the workday. For example, here we have GPT-6 create an HTML file for my " Factory Manager AI Weekly" every morning, covering everything that happened in the AI field over the past 24 hours. It quickly updates me on new content, as well as videos and articles that are performing exceptionally well. GPT-6 browses the entire web, filtering through hundreds of X and YouTube accounts to find the most relevant content, ensuring I stay up to date with the latest developments. Of course , in your prompt, you can list the information sources you want to track daily and specify the filtering criteria based on your background. The more valid information you provide, the higher the quality of the feedback you receive. You can also adjust it to your needs; for instance, if you're in sales , you can have it automatically check your emails and research your CRM system, then prepare a daily briefing for your workday every morning. Here, GPT-6 generated the entire briefing in HTML format, with both the content and brand colors matching exactly what I requested in my prompt. By giving the AI one more instruction, we can have GPT-6 create a scheduled task to automatically email the updates to me every day. You can see this task in the scheduled tasks tab here, where the model has already written out the instructions for what it needs to do. Once the scheduled task is set, it will run automatically at the designated time every day. I plan to have the AI automatically scrape content from the web and organize it for me every day from now on, so I don't have to refresh and check every single account myself. Finally, I’d like to talk about how we can find that balance between performance and price. We all know that GPT-6 is incredibly powerful, but once you start using it, you'll find that if you're not careful, your usage costs can rack up very quickly. Since its release, I’ve spent the past few weeks trying to figure out how to balance its powerful capabilities with the cost of consumption. So, that is the underlying reason why I made this video. Because GPT-6 is quite expensive , we should really only use it for tasks where it can truly shine and deliver value. The use cases I demonstrated in today's video are exactly the types of tasks where GPT-6 excels. For everything else, I basically use the standard GPT-6 model. For example, when checking my emails, would there really be a difference if I switched to the GPT-6 model for the same task? Yes, there would be, but considering the cost of using GPT-6, is that difference really worth it? I don’t think it’s that worth it, so you can try giving GPT a task. The first time you do it, run the same task using both the GPT-6-RA and the regular GPT-6 models, then compare their results. If you don’t feel there’s a significant difference, just stick with the standard model, because that way, you can use it longer and it’s more cost-effective. Only when you encounter those truly critical tasks should you switch to the GPT-6-RA model . OK, that covers how to integrate GPT- 6 into ChatGPT to help us implement automated workflows. The prompts mentioned in the video are all in the description and comment section, so be sure to try them out yourself. And if you want to systematically understand both ChatGPT and GPT-6, these two videos of mine provide a complete introduction. If you found today’s video helpful, please like and subscribe to the channel. See you in the next video, bye.