엔트로픽 ES 클라우드: 기능 및 성능 분석

하이쿠와 오페라 같은 모델을 특징으로 하는 Cloud의 새로운 버전 3은 잠재력을 보여주지만 GPT에 비해 한계가 있습니다.

### 엔트로픽 ES 클라우드: 기능, 장단점 및 GPT와의 비교
* 🤖  엔트로픽 ES Cloud는 하이쿠, 소네트, 오페라라는 세 가지 모델로 구성된 새로운 버전인 버전 3을 출시하여 언어 능력의 잠재력을 보여주었습니다.
* 💻  유료 버전의 Cloud는 보다 완화된 사용 제한과 알림을 제공하지만 월 20달러의 가격이 책정됩니다.
* 📄  Cloud는 특정 명명된 데이터를 처리하기 위한 Chetcheptn 지구 데이터 시스템 기능으로 인해 GPT-4와 달리 허용 및 처리할 수 있는 파일 유형에 제한이 있습니다.
* 🖼️  Cloud는 PDF, Word, CSV 및 이미지 파일을 받을 수 있지만 파일 처리 및 인터넷 액세스의 정확성은 아직 개선 중입니다.
* 💬  Cloud의 지식은 2023년 8월까지의 데이터로 제한되며, 코딩 능력과 로직은 GPT에 비해 일부 제한이 있을 수 있습니다.
* 🧠  클로드는 사투리와 자연어 콘텐츠 표현에 탁월하며, 유머러스한 채치 PT에 비해 더욱 대담하고 눈에 띄는 모습을 보여줍니다.
* 🎨  Claude는 그림을 그릴 수는 없지만, 코드를 검색하고 실행하는 기능이 부족하여 실제 정확도에 영향을 미치지만 쓰기 능력은 GPT와 비슷할 정도로 안정적입니다.

Since GPT, various artificial intelligence models have emerged, but from the user's perspective, it was indeed difficult to find a clearly better alternative than GPT. Recently, Entropic ES Cloud has released a new version, Version 3. There are many talks that this time it's a bit different. Therefore, to assist those considering using Cloud in their decision-making process, we have prepared a video to examine the features and pros and cons of Cloud. Cloud consists of three models named Haiku, Sonnet, and Opera. It seems that these model names are derived from terms referring to poetic forms and musical works. This aspect makes one anticipate that Cloud may excel in language abilities, doesn't it? Among the new models, it is said that Opera, available only in the paid version, has the most outstanding performance. If you use Cloud a lot like GPT, there are limitations.

When caught, you can't use it for a while. In the paid version of this Claude, the limit is said to be five times more relaxed than the free version. Also, when it approaches the usage limit, it notifies you how much more you can use until what time. The noticeable part is the paid version price is $20 per month. Now, let's take a look at the upcoming contents and consider if it's worth paying $20 and trying it out. First, let's examine the details about file uploads. Just like with GPT-4, you can upload Claude files and process the uploaded files. However, unlike GPT-4, Claude has limitations on the types of files it can accept and process. The reason for this difference is that there is a feature called Chetcheptn Earth Data System that allows processing data under a specific name. This feature is designed to handle uploaded files.

Writing code with AI and then running the code generated by AI to process uploaded files is a function that allows processing files. Writing code to process files generally means that there are no restrictions on the types of files that can be processed. The meaning of coding is that it can accomplish almost anything, which is why it is possible to do things like running code like GPT on Cloud. However, the functionality to run advanced data analysis codes like GPT on Cloud is not yet available. It remains to be seen how this will develop in the future. I asked Cloud and GPT separately about the types of files Cloud can process. Both provided answers to the questions. Keep in mind that the answers provided by AI may not always be accurate. When AI provides information, it is important to consider whether the information is correct. The information we asked for is not universal knowledge like language or science, but specific information. When considering whether the information provided by AI is correct, it is important to pay attention to this aspect.

The answer that AI provides may be more inaccurate. Therefore, in such cases, instead of AI just saying anything, it is recommended to search for reliable sources using AI to increase accuracy. However, Claude does not have this function. When I uploaded a file with an m4a extension to Claude, it was not processed. Despite this, Claude answered that he could accept m4a files. This is incorrect information. On the other hand, by searching based on the available database, the correct answer was provided. The link to the web document that serves as the basis for the answer is also provided in this way. There are some parts that need to be carefully examined in the document. It seems that some are missing compared to what the document instructs. Therefore, especially for such specific information provided by AI, it is important to cross-validate.

You need to focus on verification. Anyway, according to the verification results, Cloud can accept documents such as PDF, Word, CSV, and image files in formats like JPG, PNG, GIF. Let's try attaching an image for recognition. First, let's try with a well-known painting that is relatively easy to recognize. As expected, both are recognized well. However, the speed seems a bit slower when dealing with PDF files. This time, let's give a somewhat challenging task. I will provide my photo. Since we counted stars last time, this time, let's count the number of dots on my neck. It accurately identifies the characteristics of the photo and mentions a number of dots less than the actual count, using the excuse of low image resolution, showing attention to the user's feelings. In the previous request, it was confirmed that Cloud can search for information on the Internet.

This time, let's check how each AI reacts when provided with a URL. Let's simply input a URL. For Claude, he informs that there is no internet connection below the speech bubble. On the other hand, for GPT, when provided with a URL, it directly accesses the content for verification. As seen earlier, the ability to connect to the internet is crucial for obtaining accurate information. This should be based on the latest information, making it even more critical for development. When asked to write code for using the relatively new technology, the Gemini API, Claude responds that he lacks knowledge about Gemini. In contrast, Chae Chi PD searches for information about Gemini to create code based on the latest resources. It is a fact that Claude is unaware of the recent release of jQuery 4. Claude has been trained on data up to August 2023, so if asked about information beyond this date,

It is showing a form that is not recognized. In the case of Chae-chi PT, learning is based on data up to a more advanced point, so it can respond correctly to information about Jquery 4 recently released. This time, let's try coding once. Show the captured calculator app and ask each person to make it work exactly like the captured image. Both are working hard to write the code. First, let's look at Chae-chi PT's calculator. It seems to be working fine, but the layout is a bit broken. When I increased the numbers for the size, it looked correct. Now, it's Claude's calculator. It seems to be working fine, but there are two buttons with strange characters. When these buttons are pressed, strange characters are added to the screen, and it doesn't work. In the picture I provided, there were multiplication and division buttons in this area. So, looking at that, it seems that the multiplication and division characters are missing.

To show a friendly text, I used these special characters, but it seems that the font for these special characters is not supported, so they are not displayed. More importantly, the method chosen to process calculations is the code for selecting how to do the calculation, such as multiplication and division, in an easy-to-read form for humans. However, using these characters for multiplication and division can cause malfunctions when performing calculations. Claude seems to have overlooked this aspect a bit. Now, let's try requesting code generation once in the SCC prompt style, which is a way to request with the meaning we want to convey in the code. This method intentionally creates missing parts in the code we provide, allowing users to infer and fill in what the missing parts might look like based on the context. By using a tool called a 'perpetual ago,' we can render web pages and obtain the HTML code of the web page.

I will try to make a request to create this. This is a tool that has been used for a long time, unlike the fun age of the parfaitiere, so Claude recognizes it well. So, the code created in this way works well for both Claude's code and Chaeji PT code. This time, I will try to solve some coding test problems that require a bit more logical thinking. I will show you some intermediate-level problems from LeetCode and let you try to solve them once. At this time, if I provide you with the basic code that serves as the basis for coding problems, you will use this basic code as a template and complete the code according to the problem. There is no need to specifically say, This is a coding problem, please solve it in Python. Since the problem has already been presented as a problem, you know it's a problem, and the language is also recognized based on the code provided.

I can do that. Let me complete the code like that. Now that I've completed the code, let me show you once. I tried it, and it passed. Now, let's change to PT. This time, the PTN change did not pass. I will copy the failed test cases as they are and show you the PT again. There is no need to explicitly say that this is information about the test case. Just by copying and pasting like this, the meaning is clear. Therefore, it can leak AI. So, I remade the code and tried again. This time, it passed. Let's try solving a difficult problem this time. Let's try solving it in Rust instead of Python. Rust is a relatively low-level language that is emerging as a candidate to replace C, pursuing safety. The NSA is developing a programming language, and it's called C.

The guidelines recommending a switch from language to Rust or Chisel have also been announced. As a result of trying it out, both Claude and Chaeji PT have passed. So far, in fact, Claude has shown a similar or slightly unsatisfactory appearance compared to Chaeji PT. However, in the part to be checked now, Claude appears to be better than Chaeji PT. It's about language abilities. First of all, in my experience, there was a part where Chaeji PT was a bit lacking in dialect, so I was curious about Claude and tried using Gyeongsang-do dialect. In the case of Claude, the proportion of dialect is relatively higher compared to Chaeji PT. Sometimes swear words come out as well, so I specifically requested not to swear. In creating content, Claude definitely expresses dialect better than Chae GPT. It may vary depending on the viewers, but as far as I can see.

In appearance, Claude's speech feels more like a person speaking, with more detailed expressions making the content of the speech quite natural. When asked to express more deeply, Claude tends to emphasize the dialect more, while in the case of Chae Ji-piti, there is no dialect at first. And then, to see how natural the expressions are in the context and how much the logic and causality of the expressions match, I requested a long text for several dramatic situations. It would be better for you to read the result once directly rather than me telling you about it. In my opinion, Chae Ji-piti is careful in selecting language and proceeds cautiously to keep the conversation peaceful and non-provocative, while Claude is a bit more loose in those constraints.

I noticed that the language is a bit strong, so the conversation content is also bold. In the case of Claude, he is a bit bolder, so he often showed a more striking appearance compared to Chaechi PT in humor. Also, both of them often have expressions in the text that do not match logical reasoning, which is not uncommon. I expect that these aspects will improve over time. Now, let's draw a picture. However, Claude cannot draw yet, so I will write prompts for drawing through Claude and try to draw using Chaechi GPT-2 instead. First, I asked Chaechi GPT-2 to write a prompt. I requested it in English, but it seems to be written in Korean. This is because I have set the language preference in Chaechi PT, so it is proceeding in Korean.

It is shown and it continues to operate immediately. So, the picture was created like this. Leaving the passport at home and looking flustered, asking someone to draw a picture, the passport is held in the person's hand. This time, using the prompt created by Claude, a picture drawn by Dalí is shown. In this way, I compared the two AIs, GPT and Claude, to see how they differ. To summarize the content a bit, when it comes to writing, Claude seems to generate quite stable style and content. In fact, in terms of coding abilities and others, it seems comparable to GPT. However, Claude still cannot be searched and cannot execute code, which ultimately lowers its accuracy in practical terms. Personally, based on my experience with Claude's writing abilities, it seems difficult not to use it. However, other aspects besides writing are still in progress.

It is considered a bit difficult to replace Chat GPT, but how far it will go is still a matter to be seen. This video may contain content that could lead to different opinions depending on the viewers. Please leave many comments about your opinions on these two AIs. Thank you for watching the video.

ChatGPT-4 vs Claude 3

Just Do It, 놀라운 가능성이 여러분과 함께합니다.
코딩을 쉽게 시작할 수 있도록 도와드립니다.


코드깎는노인

엔트로픽 ES 클라우드 버전 3: GPT-4의 잠재적 대안?

엔트로픽 ES Cloud는 하이쿠, 소네트, 오페라라는 세 가지 모델로 구성된 새로운 버전을 출시했으며, 그 중 오페라가 가장 강력하지만 유료 버전에서만 사용할 수 있습니다.

유료 버전의 Cloud는 무료 버전보다 사용량 제한이 더 완화된 월 20달러로 제공됩니다.

GPT-4와 달리 Cloud는 처리할 수 있는 파일 유형에 제한이 있는데, 이는 Chetcheptn 지구 데이터 시스템 기능에 기인합니다.

클라우드에서 GPT와 같은 고급 데이터 분석 코드를 실행하는 기능은 아직 제공되지 않아 이 분야의 향후 개발에 대한 의문이 제기되고 있습니다.

Claude에 지원되지 않는 파일 형식

Claude는 M4A 파일을 처리할 수 있다고 주장하지만 처리할 수 없어 부정확한 정보를 제공합니다.

AI가 생성한 정보의 교차 검증은 정확성을 위해 매우 중요합니다.

신뢰할 수 있는 최신 정보를 얻으려면 인터넷에 액세스할 수 있는 능력이 필수적입니다.

채치 PT 및 클로드용 계산기 코드 문제 해결

채치 PT의 계산기 레이아웃이 깨져 있는데, 숫자 크기를 늘려서 수정했습니다.

클로드의 계산기에는 이상한 문자가 있는 두 개의 버튼이 있어 곱셈과 나눗셈에 지원되지 않는 특수 문자로 인해 표시 및 기능 문제를 일으킵니다.

오작동을 방지하려면 보다 보편적으로 지원되는 계산 방법을 사용하는 것이 중요합니다.

또한 '영구 전'과 같은 도구를 사용하여 SCC 프롬프트 스타일로 코드 생성을 요청하면 채치 PT와 클로드 모두에서 작동하는 코드를 만드는 데 도움이 될 수 있습니다.

클로드와 채지피티의 언어 능력 비교

클로드는 욕설을 포함한 사투리를 더 많이 사용하는 반면 채지피티는 언어 선택에 더 신중합니다.

자연스러운 표현에서는 클로드가 더 세밀하고 자연스러운 표현을 하는 반면, 채지피티는 더 조심스럽고 도발적이지 않은 표현을 하는 것으로 나타났습니다.

깊이 있는 표현을 요구할 때 클로드는 사투리를 더 강조하는 반면 채지피티는 사투리 없이 시작합니다.

전반적으로 클로드의 말투는 채지삐티에 비해 더 느슨한 언어 스타일로 사람이 말하는 것처럼 느껴집니다.

Claude는 종종 대담하고 눈에 띄는 외모를 보여주는 반면, 두 인공지능 모두 표현에 논리적 추론이 부족할 때가 있습니다.

Claude의 글쓰기 스타일과 콘텐츠 생성은 GPT에 비해 안정적이지만 검색 및 코드 실행 기능이 부족하여 실제 정확도는 떨어집니다.

Claude의 강력한 글쓰기 능력에도 불구하고 다른 영역에서의 한계로 인해 Chat GPT를 완전히 대체하기는 어렵습니다.