# Claude 2.1 is Here - A Real Threat to ChatGPT

## Метаданные

- **Канал:** Skill Leap AI
- **YouTube:** https://www.youtube.com/watch?v=wACvSPggAJA
- **Дата:** 21.11.2023
- **Длительность:** 10:17
- **Просмотры:** 19,856
- **Источник:** https://ekstraktznaniy.ru/video/12889

## Описание

Claude 2.1, the latest version of the AI chatbot developed by Anthropic, marks a significant development in the field of large language models. This new version emerges as a direct competitor to ChatGPT, offering unique features that distinguish it from its predecessors and rivals.

Enhancements in Claude 2.1
The new iteration of Claude includes several notable updates:

Expanded Context Window:

Claude 2.1 offers a context window of 200,000 tokens, equivalent to approximately 150,000 words or 500 pages of text. This allows for the analysis or summarization of extensive documents such as financial statements, code bases, or entire books.
This feature is exclusive to Claude Pro, which requires a subscription.
Reduced Hallucination Rates:

The model significantly reduces the rate of generating incorrect or fabricated information, a common issue in AI chatbots.
Improved Accuracy:

The update has led to a 30% reduction in incorrect answers, indicating enhanced intelligence and reliability.

## Транскрипт

### Segment 1 (00:00 - 05:00) []

we have a brand new AI chatbot called claw 2. 1 now claw 2. 0 if you've ever used it before is a large language model it's a chat GPT competitor they raised a ton of money from Amazon and it's become a real player it's by a company called anthropic but 2. 1 has things we've never seen before in this AI chatbot World in this large language models and let me go ahead and give you some of the key points here and then we'll take Claud here for the test drive because if you go to Claud. a it is now available to test out this new model 2. 1 so there's three big updates I'm going to cover before we take it to the test drive here the first one a 200k context window so that is 200,000 tokens just to give you some perspective on non technical terms that is 150,000 WS now you could upload as a document to analyze or to summarize or to interact with that's 500 pages of material so that could be an entire financial statement like an S1 document it could be an entire codebase book and it says you could then use that content or that data and Claude could summarize it for you could do Q& A you could forecast Trends with financial data and things like that and you could do this with multiple documents compare and contrast uploading multi multiple documents and count towards your context window of 150,000 words now I should note this one update everything else is actually included in the free version of Claud except this one requires Claud Pro which is $20 a month so I'm actually going to upgrade for this video because I've been using Cloud for free but with the 200k token context window I have to upgrade there's nothing that comes even remotely close now just to give you some context of where we were before claw 2. 0 actually had the largest context window was 12,000 tokens but now this 200,000 tokens if you compare it with some chat GPT models like GPT 4 that's only 8,000 this is 200,000 these are not even in the same league now here's another key point that is very useful two times decrease in hallucination rates so if you've ever used any AI chatbot like chat GPT or Claude you notice they just sometimes make stuff up in fact I've seen Claude 2. 0 make some crazy things up in one time created an entire company created an entire link to how much money they raised and none of that was true it kind of blew my mind it was so sure and I was so sure that it was giving me the right information but I basically tested it on Google and it was all wrong so that Hallucination is a huge problem and basically cutting in half very useful update as well and the third big update here is it's actually smarter right so each time they update these they make them smarter so here they ran a test and it says 30% reduction in incorrect answers and here with this graph and I'll link this below here if you want to read the full post with this graph you could kind of see that the improvements it's made now when we take it for a test drive we'll do some testing on the context window and so on and right now it's not only available inside of claw. a so the regular chapot that's free to use you could go there and test it out and it's also available inside of the API so if you use the CLA API to build your apps on a lot of people are using the chat GPT API from open AI some people are using Cloud instead and this now is a lot more useful with the 2. 1 inside of the API and inside of the free chatbot okay let's take this for a test drive so go to cloud. and if you haven't used CLA before and maybe you're only using chat GPT or maybe you're using Bing or Bard this is a worth a try I probably split my week between Claud and chat GPT recently I've been using chat GPT a lot more but I'm really excited now with this context window and with all these improvements to try 2. 1 even more so right here this is where your message will go and then right here this is where you could upload files so it says Files 5 Max 10 megabytes each and typically I get better results with CSV file but it says you could do PDFs and text files as well so let me go ahead and upload a document I'm going to do a txt because with a word doc I have all kinds of problems you usually gives me an error message so I usually convert any word document into a txd chpd does it much better job with file formats that are Excel and W and things like that this is better with txt and CSV files okay so I uploaded this document this is somewhere between 50 and 70,000 words this is the biggest document I have right now and usually I was breaking this up when I was using trat GPT into much smaller files right the context window is very small here right now let me go ahead and see if it could give me a one paragraph summary and I'm going to just see how long this takes I'll let you know if I need to just cut this out but right now I just

### Segment 2 (05:00 - 10:00) [5:00]

pressed enter and it says conversation with long prompts or large files may take a few moments so I'll let you know exactly how long this took as soon as it's ready so it took about 12 seconds only to go through this again this is 50 60,000 W document here it says this is a comprehensive guide on generative AI focuses on introducing tools like chat GPT mid Journey Dolly and so on so yeah very accurate here on exactly a one paragraph summary of this and let me see if I could just follow up and see if this takes 12 seconds every time or it's going to have more of a hard time let me pull up my document here for prompts I created this document before when I made a claw 2 video so I'll go ahead and Link this in the description if you want to get this as well but this has basically depending on the category bunch of different prompts there's 100 prompts I put together here to do all kinds of analysis on different types of documents which is really the best use case for cloud over any other large language model so right here I'm going to take this who's the intended audience for this document so this is kind of an interesting question because now it has to actually figure something out that is not just pulling direct information out it has to analyze the whole document here to figure out who the target audience is let's see if it gets this right okay this time it took about 20 seconds here to give me this answer it says based on the content and the tone it seems to be intended for everyday people who are new to generative V want to learn how to use this tool for Pur Prof projects that's perfect and then it gave me some key bullet points and I read through this and it's very accurate it did a really nice job here now I'm going to show you one more data analysis with numbers here to see how it does with that but I want to show you this chart here it says CLA 2. 1 open in the conversation accuracy so it says right here 2. 0 declines to answer and 2. 1 the decline to answer rate has actually increased which is one of my biggest frustration that I had with claw 2. 0 so right now I'm going to ask it how do you compare to GPT 4 okay this is a very simple question right chat GPT B and Bing all are going to give me really good answers they typically create a table kind of format for me it says I do not have access to GPT 4 and I can't make accurate comparisons lots of times it just says AI as created by anthropic to be helpful harmless and honest and it just keeps repeating itself that way so just a decline to answer part of it is a very huge downside for me this was the same problem with 2. 1 or 2. 0 and it's now the incorrectness of answers has declined right but if it's refusing to answer more often that's kind of a problem so what I found is Claude is extremely useful when it comes to analyzing data better than anything else especially with this insane huge context window of 150,000 words right that's not going to be beatable by anyone not even close but when you want to get just answers to questions not very useful Bard does a better job and chat GPT but this does a better job sometimes in summarizing any type of text you paste into it too you don't always have to upload and it's done a good job writing email copy for me okay next let's look at some financial documents let's see how it works with numbers so I have this document here this is just the S& P 500 and I'm going to just ask you some questions and if you want to test it out there is a website called kaggle. com this is where I got it you have bunch of different document types here if you go to data sets you could download all kinds of different data sets that are available here for download and it's completely free to use let me go ahead and upload this and I'm going to refer to my prompt book here I have things based on doing analysis on financial data or analyzing just general data here and right now I just said give me the top 10 companies based on market cap specific to this doc doents let's see what it comes up with I'm checking here for accuracy and I'm checking for Speed as well okay and this took about 10 seconds and he got a right for the most part he got the top 10 but he put alphabet twice but he wouldn't know that because the data set also showed alphabet twice because it has two different types of stocks or class of stocks here and the numbers here this is supposed to be the market cap but I thought he made a mistake but then I look back at the documentation here so the Apple stock here it shows something like 8 trillion here as the market cap which is not true which is closer to three so you got some things wrong so you could see the market cap category just had the wrong numbers in it so there was something wrong with the shorting of the regular data set that I downloaded very quickly from that website but as long as you have accurate data set and I've tested this out with 2. 0 and it also did a really good job with financial data and any type of p& l and personal business data too that you have anything from QuickBooks that you could upload to it again this is not private so make sure you don't give it

### Segment 3 (10:00 - 10:00) [10:00]

something very personal but as far as doing a quick research for me off these type of CSV files it did a really good job and I'll do a deeper dive this was more a first look this just came out so I've only had a couple hours here so stay tuned for that subscribe thanks for watching I'll see you next time