Transcription
I have long dreamed of being able to organize my prompts in GPT. For this, I stored them in various text files, in Notion, in Obsidian. But now GPT has made it beautiful; they have introduced the Projects functionality. And now I will show how it works. They are located in the right tab, Projects. Here you can create a new project and place different chats inside the project. This is very convenient.
In order to organize all your chats in a structured folder, you create a project. I will create a test project. Here it is, it is immediately displayed here in the right sidebar. What can be done here? Here you can add files for a knowledge base, so that it draws information from the knowledge base, and add your instructions. But you can work perfectly well without this too.
The most positive thing that I like is that here I can add my chats, and here will be exactly those chats that I need for the project. I will not have to search for them here in a huge list. Adding a chat is very simple. Here, for example, I have a chat. I click the three dots, add to project, and move it to a specific project. Here I move it. Everything, it is here. But if you don't know where this project is, it will be difficult to do.
For example, I have a chat where I copied Mash's style, but I don't know where it is. I just found it through search. Where is it? Well, for this, I just need to write something so that it appears in the left sidebar, and then I can move it. Here it appeared on the left, and I am now moving it to the project. Everything, it is here.
Now, regarding custom instructions. In GPT, there has been GPT customization from the very beginning. Here you can write instructions and how GPT should relate to you. I have never used this because I have many different roles, just like you probably do. For example, at work you have one project, at home you might have some of your own projects, maybe freelance, and you don't want this to overlap.
Now, in Projects, you can customize GPT, use this GPT customization setting for a group of chats, and it will only be used here. Here you can add instructions, and here you can add knowledge. This is just incredibly convenient for not mixing different types of information. The only thing is Memory. Memory will work everywhere, so I recommend turning off this Memory so that it doesn't interfere with chats, if this is relevant for you.
Now, for example, I will create a new project, upload a specific knowledge base there, and show how to use it. So, I am creating a new project. This is a project for creating responses to reviews on marketplaces. For example, you have such a task to form responses to reviews. What will I add here? I Googled how to respond to reviews on marketplaces. I opened several articles that are more or less trustworthy. I copied articles, for example, this article with all the examples. Here I copied, here I copied from this page, I copied. I got several text files.
Here is a text file with different examples, good and bad, on how to respond. Here is an article on how to make responses. Here is a transcription of a YouTube video on how to deal with negative comments. And here is a special file, relevant products. That is, here I have generated relevant products, just collected articles, names, and descriptions. At the end of my review, according to my idea, a relevant product should be added. Here they are. I click "Add Files".
So, if you select all files and drag them here, they will just be added to the chat. It doesn't work like that. You need to click "Add Files" and upload files only to this area. I am uploading files for the project. All files are uploaded, and I click the cross. Here are all seven files. Now I am adding instructions. I created such an instruction: "Your task is to write reviews for buyers of marketplace products. I will give you the text of the review, and you should write a response to the review, relying on the knowledge in Project Files, in order to form a response to the review." Thus, I am making a reference so that it uses its knowledge base that I uploaded to it.
Further, at the end of the review, select a relevant product from "relevant.txt". That is, here is this "relevant.txt". I thus referred to it so that it takes relevant products from here, so that it understands that only this file relates to relevant products. Well, and here I write that at the end of the review, select a relevant product. The product should be as relevant as possible, and explain why this is necessary, based on the description from the relevant text. Here it has a complete instruction on how to do this.
Note that I am not writing here how or in what style responses should be made, that is, it should rely entirely on the knowledge base, on all the information found in these smart articles. They know better how to give correct responses to reviews. Now I will go to some products and let it suggest responses to reviews.
For example, a hat. We go to reviews. We take some review. Here, for example, a four-star review. I take a screenshot. You can also copy the text, but a screenshot is faster. I paste it, and well, I think I need to insert the name of the hat so that it understands what it's all about. Well, it formed some kind of review, some kind of response to the review. "Thank you for your purchase. I'm glad you liked the hat." It understood that it liked it and was pleased with the comfort. It clung to the word "comfort" and gave a relevant response to this review. Well, this suggestion that it will cope with the cold and become a reliable companion is excellent. And here it suggests thermal underwear. Yes, now I'll look. I have in relevant. Yes, this is the product. Everything, it has made it relevant.
Let's look at something else. Let's find some nasty review. For example, these boots. So, I'll find a bad review. Excellent. I paste it, and I need to copy the product name. It understood everything that the boots did not meet expectations, that they are very slippery. Well, honestly, for the user, this is not particularly important. It writes here that "we will pass on your feedback to the development team to improve quality in the future." That is, you will still have slippery boots, and someone will develop them, and it will be good. That's not great to write. And here we are, so to speak, offering the most polite option: "If you really want good ones with improved grip, then take other boots with this article number. They have a reinforced sole, blah blah blah." And by the way, what I like is that it integrated this relevant product very well. That is, it has excellent stability on ice, snow, and for walking, and for driving a car. And the woman complained that it was slipping off the pedal.
The prompt, of course, can be improved. I don't know what is listed in the knowledge base. Surely there is something like "we will pass on your feedback to the team." To whom will they pass it on, because it's just an ordinary reseller? That is, well, such things simply cannot be written, and this needs to be in the prompt further. That is, I have all the instructions. You will ask me, how does this differ from GPTs? That is, here, GPTs can be made, and here, similarly, you can add instructions and a knowledge base. The actual response will be the same as what GPT did. But GPT is not as convenient to use, and plus, chats are not saved here, and you cannot add other chats to GPT. You simply cannot group chats.
Often, in fact, I don't necessarily need to use these custom instructions to combine chats. I just need chats to access them, so that they are with me. I don't necessarily need to write custom instructions. And here you cannot group them. Also, custom GPTs are a more complex construction. Here you can make additional settings. Here you can get data from external URLs. This is a more complex system. Also, these GPTs can be shared on the internet, and someone can use them. But a project cannot be shared.
In my opinion, this is a big step for the user. This is a new feature. I haven't had time to upload everything I need here yet, but I plan to create projects with prompts here, chats, and use these chats as prompt storage. This is one example of what I need.
There is also a disadvantage in Projects. When you create a project, you can only use GPT. Nothing else can be used here. You can add files here, you can also add files from Google Drive and use the canvas mode. But you cannot use GPT-4o, GPT-4o Mini, and some older models. Well, often, for example, I don't like that I can't use GPT-4o because, well, I wanted to structure projects for development too. GPT-4o creates code much better. That is, I cannot organize this yet, it will just be scattered everywhere.
Also, as you can see, in regular GPT-4 there is search, but here there is no search yet. Look, this chat. There is no search here. You can only attach Web Tools, images, canvas, files. That's it, no searches here. And that's good. Most prompts I can simply systematize and then just transfer to GPT-4o when I need it. This may seem like a trifle, but in my opinion, it significantly improves the user experience. It is now much more convenient to work with GPT and systematize your projects.
Personally, I need this very much because I am active in different directions, in different languages. Due to the chaotic arrangement of files, I had to spend extra time to find the right prompt, the right chat, and work with it. And now I organize everything by shelves, and I will have quick access to everything I need.
Subscribe to my YouTube channel so as not to miss information about the release of new such features. Follow the link below to Telegram, subscribe to my Telegram channel, where I talk about various GPT features that are not here on YouTube, trifles, prompt engineering. In general, I have a lot of additional content on Telegram. That's all for now. Bye.