When I asked Claude what is its purpose and capabilities are:
My purpose is to have natural conversations and be helpful, harmless, and honest. I’m an AI assistant created by Anthropic to be useful, safe, and transparent.
Some of my key capabilities include:
- Conversation – I can engage in open-ended dialogue and answer followup questions. I don’t have a specific domain, I’m designed for general conversation.
- Reasoning – I can make logical inferences and provide reasoned explanations for my responses. This helps me have more natural and contextual conversations.
- Knowledge – I have access to a broad set of facts and general world knowledge that I can draw on to answer questions. I don’t have personal experiences though.
- Language understanding – I can comprehend complex language and nuanced meanings in questions to discern intent and context.
- Limitations awareness – I’m transparent about the limits of my knowledge and capabilities. I’ll let you know if I don’t have enough information to answer well.
- Harmlessness – I was developed carefully to avoid biased, harmful or unethical responses.
My goal is to provide the most helpful, harmless and honest assistance possible through natural dialogue. Please let me know if you have any other questions!
Starting with number one, which is Claude from Anthropic. Now, this is similar to chat GPT in that you chat with it. You can ask questions, you can get advice from it, but what I really, really love Claude for is for summarizing long content. And the reason for that is when you’re using GPT4, you have a context window of roughly 16,000 tokens, which comes out to about 12,000 words or so combined between the amount of text you can input and the amount of text it will give back to you. So if you try to plug in a really, really, really long document that’s longer than, say, 12,000 words, it’s not going to work very well in chat GPT. However, Anthropic has a 100,000 token context window which gets you roughly 75,000 words in and out. So when it comes to summarizing really long research papers or PDFs or online articles, nothing beats Claude. And here’s a research paper I came across called iterative multi-granular image editing using diffusion models. And as you can see, it’s a 15 page document with a lot of words, probably more words than what would fit into the chat GPT context window. So let’s go ahead and download this PDF. If I jump back to Claude here, come to my folder where I download this PDF and just drag and drop it right here into Claude. I actually get this error that says you may not upload files larger than 10 megabytes. That’s not a big deal, because check this out. Let’s go ahead and open the PDF here and I’m going to control A. And I’m just going to select the entire PDF. So I have all of the text and I’m going to do a control C or command C on Mac to copy everything. Then if I come back over here to Claude and I just command V, paste that in, you can see it just pastes it as a text file. So I get the same result out of it. Now I can type something like please summarize this research paper. Give me bullet points and explain it in simple terms so that even non-techy people can understand it. Here’s a simple summary of the key points from this research paper. The paper introduces a new problem setting called iterative multi-granular image editing that involves iteratively editing and image according to a sequence of instructions while controlling the spatial extent of each edit. So then it goes on to explain it here. Still a little more tacky than I would have liked. Why should people care about this paper explain it like on five? And there we go. Here’s an explanation of why this paper matters in simple terms. Imagine you’re driving a picture, you start with a blank page and what to draw a cat. First you sketch the outline of the cat, it looks good. Now you want to add a hat on the cat’s head. You’re able to erase just the head part and draw the hat without messing up the rest of the cat. You already drew. Then you decide to also give the cat a balloon in its head. You can erase just the hand part and add to the balloon. This is like iterative and multi-granular editing. And it goes on to explain it in the simplest of terms. So now I have that analogy, that sort of better explanation that simplifies what this paper actually does. Claude is amazing for this. Claude can do a lot more. This is mostly what I use it for. It’s for summarization of long articles. But you can do a lot of similar stuff with this that you can do with chat GPT. And as of right now, it’s still 100% free to use.
My Machine Learning Blog may not own some of the content presented.
Copyright Disclaimer under section 107 of the Copyright Act of 1976, allowance is made for “fair use” for purposes such as criticism, comment, news reporting, teaching, scholarship, education, and research. Fair use is a use permitted by copyright statute that might otherwise be infringing.
All posts on this video blog are my personal opinion and they don’t in any way reflect the opinions of my employer. All materials, posts, and advice from this site are informational, researching, and for testing purposes only. You can use them at your own responsibility. I’m not in any way responsible for any damage done by following posts, advice, tutorials, and articles from this video blog.