How many of you are currently using GPT? To be honest, I would like to tell you that these days, I use a lot of GPT called Claude. In my case, because I have a lot of writing and content production, but recently, Claude announced version 3.5, which has become a hot topic. When I tried using it, it was really good. The feature that shows a preview really improved productivity, allowing you to see the appearance of data charts immediately. But today, there was another amazing announcement. If you watch this video, you might not want to use GPT in the future. Personally, I have two GPT subscriptions, but I am thinking of canceling one and adding another Claude subscription instead. Or should I switch to a team plan and use it with my team members?

There was an important update. It's the project feature. What was the most disappointing part while using the cloud so far, and why did you use GPT despite the goodness of the cloud? It was because of custom GPT. I pre-entered prompts for tasks that needed to be repeated or uploaded data files to create custom GPTC and used them there. Although the cloud is good, I used GPT in parallel. Now, through the project feature in the cloud, similar functions have been implemented. In today's video, I will briefly introduce the project and show how to create and implement it. Secondly, when you upload about 500 pages of text as context to the project in this cloud.

It is said that they recognize them, so I will actually see if that is true. For example, I will upload a 500-page book once. Secondly, I will test whether it can find well when it is done in multiple files, and how long it takes. I will also test these aspects. I will also take the time to compare it with GPT under the same conditions. So, in today's video, I will focus on how well Claude's project function can recognize and implement a 500-page book and how helpful it will be in writing. Now, let's start the video. First, I have appeared on the screen, so let's start with Claude's presentation. It was presented on June 25th. I presented about Claude's project. The content is available at the URL I posted in the pinned comment of the video.

I will let you see it, and I will only talk about important parts. In simple terms, we are now providing Pro Planning Housing. In the project function provided by Cloud ES, it is said to include a contact window of 200k corresponding to a book of about 500 pages. So, by uploading the content equivalent to a book of about 500 pages, you can continue to have consistent conversations within it. There used to be parts where repetitive tasks had to be repeatedly done for Cloud, but now they are made like pieces and can be easily found. In other words, to put it simply, it's like creating your own custom GPT project. If you look here, you can upload files like this. There is a limit to the file size when uploading this file. We will look at that later. And now, after uploading it, if you utilize it as a team plan.

You can share it with your team and discuss the story together. Let me show you a simple video as well. Rud's project, let's briefly understand it. I will test 500 pages in real life. Now, let me switch the screen and show you. Let's test with Claude Lorang. First, let's see how to create a project in Claude. If you go to the top left here in Claude, you can see the projects I have created like this. These two are not made by me, but these two are the ones I made. For example, in this case, I probably uploaded a commission policy PDF and always asked for a commission from here. It's simple. First, click on create a project. I named the project 500 Page Confirmation and then you can write what you want to do in this part or you don't have to write anything.

Let's proceed with the test. If you look to the right, you can see the upload button. We will try uploading the actual file. It's about 500 pages long. Let's try uploading this PDF file by drag and drop. This is the first test. After uploading, you will see the loading process. In many cases, it's not easy to test because there are not many documents that are 500 pages long. I will do the test instead. Here, you can see that this file is currently 10 MB when uploading. Once the upload starts, it shows that 55% of the capacity is used. You can't keep uploading infinitely; there is a limit to how much you can upload. Here, you can see there is one file uploaded. So, this is how the upload was done with a very large file.

And when I uploaded it, there was a file size limit. I remember it shouldn't exceed 30 megabytes. So, shall we ask about this now? For example, there are 23 pages of content here, and I will inquire about the part called 스라. Let's ask about the content 스라 on page 23 of a 500-page document. It took a long time with GPTs, or there were some issues. Let's test it out. Here it is: BNS Creative Expert is a platform where designers can exhibit their work and discover others' works. This part is about designers. Here it is. I was asked to provide the source of the original text. Sometimes, the original source may be fabricated, so I will inquire about the source of the original text. This content was released after January, specifically in the Marketing Insights section.

The explanation of BNS is provided in the marketing insight session. So, let's take a look at this session as it is mentioned in the marketing insight session. If you scroll up, yes, it was indeed the marketing insight session. Therefore, in this marketing insight session, they precisely talked about this part. Now, let's ask about something else. Seller Kingdom Conference. Let's ask about the Seller Kingdom Conference. This is on page 376, and if you look here, it says it was held in Seoul on October 22, 2023. Yes, if you see here, it ended on October 21st, with 150 to 200 people. Yes, and Hill Getty and other companies. So, this part is also being discussed accurately. Now, let's randomly ask about the content around pages 100 to 150. On page 150, it says, Sweat doesn't make everything happen.

There is a part that is ready, so let me ask about it. When you can't remember a small sentence, let's ask now. I asked here about the content on page 150, such as leverage. Yes, so it accurately reflects social boundaries. Therefore, it doesn't take too long, and it only took a little longer when I asked the initial question. You can see it accurately. Now, the second thing I want to test is, now, when there is one file, I understand the content of page 500 of one file. However, when uploaded as multiple files, for example, not all 500 pages in one, but maybe 70 pages, 60 pages, 50 pages, 40 pages, like this when uploaded as multiple files, I will test whether it can respond in the same way by finding the interaction between the files and looking well between the files. This is about testing the interaction between multiple files. Let's start again and there are a total of six files. Six.

It's a total of six files. Let's take a look. The first one is Marketing Insights, 89 pages. The second one is 114 pages, around 200 pages. The third one is from 57 to 250 pages, around 300 pages. The fourth and fifth files are 330 pages. The sixth file is about 50 pages. So, it's roughly about 400 pages in total. Let's ask if it's possible to search through these multi-files effectively. Writing reviews on Chess PD is a risky business. Let's inquire about this. Will it be able to find and discuss accurately? It might not be a good idea to make up things. Also, another thing I'm curious about is whether it can correctly identify the source. For example, if we search for certain content, we need to know in which file it is located, right? Let's verify if it can search properly. Amazon considers AI-generated reviews as fake reviews.

They are monitoring and it is coming out correctly. So, first of all, you did a great job in accuracy. This is AI Insight, the sixth file of AI Insight. Yes, that's correct. I will ask about a different file once. This time, it's the fifth file, a successful seller story. If you have any information about personal saunas, please share. It's coming out accurately about personal saunas as well. Yes, it says 500 units were sold, and it's accurately stated here. So, now it's not generating fake information but actually finding the information well. I will ask about the source. You found it accurately. Now, the Cloud Project seems quite reliable. I've tried using it. Now, let's compare GPT under the same conditions. I've uploaded files from 1 to 6 with GPT. Now, let's try to generate it once again to see if it finds it accurately.

Let's check the multi-file first. Currently, it is taking more time than expected with GPT. Let's see if it finds it correctly. For now, it is taking more time. This is not a frozen screen. I am still waiting, and it hasn't been completed yet. Let's wait a little longer. GPT is currently gaining popularity, mentioning that a product called Personal Sauna has sold about 510 units on Amazon recently. The rest of the information seems similar. So, the accurate information matches. The fees are also the same, with FBA fees at 20.56 points and referral fees accurately found. When searching for the first content, it took much longer than with the cloud. Now that we found the first content, I believe it will be faster when asking the second question. Let's try the second question. For instance, let's ask the same question as before. Let's try the second question. In the case of Cloud,

In the first part, it was very fast, and now in the second part, there seems to be a slight pause. Let's pause for a moment and now, I am searching for examples within the multi-file. GPT is currently doing well in finding examples, and now, even in loading, these parts have been almost followed, so I think the value of utilizing Claude has greatly increased. In terms of writing blogs, Claude is much better in writing, so I think I will use Claude more in the future. The only advantage of GPT was gpts, but now Claude is being brought in, and in loading, it's the same. I didn't explain this before, but when creating a project here, you can do custom instructions and pre-put the same prompt, just like GPT. Therefore, in the future, these aspects can still be comparable to Claude and GPT.

It seems to be at a level that can be endured, and therefore, it seems that more value can be shown. So, in my summary, I have used GPT for a while. However, it did not surpass Claude in writing. So, even though I used Claude, there was a part where GPT's function was missing. Nevertheless, I should have utilized both GPT and Claude. But now, with the introduction of the feature called Claude's Project, we can use similar functions like GPT. Therefore, I think I will use Claude more in the future. However, I will not immediately cancel my GPT subscription. So, if possible, I recommend that you also try using Claude, GPT, and maybe something fun like Minyi to feel and learn the differences. I believe it's important to learn from these differences. Today's video will end here. How was it for you all? Today, we talked about Claude like this.

I checked out the new features of the ES project that was recently announced. I'm not sure how others feel about it, but personally, it was one of the features I was most looking forward to, so I was very pleased. These days, I am using GPT, a cloud service, more often. After the release of 3.5 Solnet, it was once again a piece of news that made me happy. In the future, I will continue to share insights that I experience while using the cloud and AI and doing business in New York through this channel. So, for those of you who haven't subscribed to our channel yet, please subscribe. By doing so, you can stay updated on the latest AI information and practical cases and learn from them. Also, we have a news layer that we operate. By subscribing to the news for free, you can receive such news directly on your phone, so please subscribe for free. We also operate a community.

Since we are engaging in not just simple aggro but want to have in-depth communication with those who want to utilize AI, we would appreciate it if you could also join the community. Finally, we will be holding a conference at the Seoul AT Center on October 4th this year. So, if you are interested in AI, please make sure to come then, listen to the lectures, and also take the networking time. I hope it will be a valuable time for you. I will also be there, so you will have the opportunity to meet me as well. Now, let's move on to the next video. Always watch the AI Korea Community YouTube videos, leave comments, and express my gratitude to all of you. I will see you next time. Thank you.

클로드 대박! 커스텀 지피티 대체!? Claude 3.5 Sonnet 프로젝트 기능 출시! 커스텀 지피티의 대항마! (GPTs 기능이 이제 클로드로 가능!)

뉴욕에서 사업하면서 알게되는 AI 지식을 이 채널 구독자분들에게 나눕니다.
Business Contact: Bopyo@aikoreacommunity.com


클로드 대박! 커스텀 지피티 대체!? Claude 3.5 Sonnet 프로젝트 기능 출시! 커스텀 지피티의 대항마! (GPTs 기능이 이제 클로드로 가능!)

클로드 대박! 커스텀 지피티 대체!? Claude 3.5 Sonnet 프로젝트 기능 출시! 커스텀 지피티의 대항마! (GPTs 기능이 이제 클로드로 가능!)

Scripts