Anthropic Unveils Claude 2, Outperforming ChatGPT-4 in AI Arena

0:00What can 160 years of experience teach you about the future? When it comes to protecting what matters, Pacific Life provides life insurance, retirement income, and employee benefits for people and businesses building a more confident tomorrow. Strategies rooted in strength and backed by experience. Ask a financial professional how Pacific Life can help you today. Pacific Life Insurance Company, Omaha, Nebraska, and in New York. Pacific Life and Annuity, Phoenix, Arizona. major breaking news in AI Anthropic which has been kind of the second biggest AI model after chat GPT for quite a long time they'd raised a ton of money they recently have been passed by inflection so they're probably more closer to round or the third place but in any case they have officially launched their chat GPT competitor which is really really big news if you're in the AI space.

0:54So today on the podcast, we're going to be talking about what Anthropics' new Claude 2 model can do. You know, we've been able to see it for a little while on Ho.com, which is Quora's kind of AI model center where you can test out a bunch of different AI models. But that's been the only consumer way to actually access Claude up until this point. And I think they're doing that intentionally to kind of test it, but they have officially launched to the public. You can use Claude on their website and so I think this is a really big step I want to break down everything that Claude can do that ChatGPT can't but before I do that I do want to say coming on the heels of ChatGPT obviously they have a big they have a lot of big shoes to fill right like they can't just come and say hey look we can do the same thing as ChatGPT and hope to get any sort of market share obviously because everyone already uses ChatGPT we're already signed up and using it so they have to do they have to go above and beyond to do a lot more and I think that's been the challenge Perhaps that's one reason that has kind of pushed them to not actually be able to launch as soon.

1:54And recently, I think something that did help is that they were able to get a few hundred million dollars from Google's cloud division. So I believe that they are plugged right into Google Cloud. And interestingly enough, Anthropic does have a, you know, semi-controversial background. I mean, it's not as bad as, I don't know, Anthropic's theology or ideology of, you know, humans not being as important as animals. But Anthropic, of course, is famously funded in part by FTX and Sam Bankman Freed, who is, you know, been arrested for running essentially a massive Ponzi scheme. That's my opinion, not investment advice.

2:32Don't sue me. But of course, the FTX exchange collapsed, but they did lead around investing hundreds of millions of dollars into Anthropic before their exchange kind of collapsed. And so I think Anthropic is happy that they were able to close that and get that money before the whole thing went bust. So Anthropic was kind of funded off of that, which a lot of people have called controversial. But overall, I think the tech that they have is pretty powerful and very interesting. So I want to talk about what they can do. The first thing to know is that this new GPT competitor is only available in the US and the UK for public testing at the moment.

3:10so if you're in those two areas you can try it I will say something really impressive about Claude too is the fact that it can summarize up to 75 ,000 words this is a massive difference from ChadGPT that has a 3 ,000 word limit like right orders of magnitude this is more than 10x right and so it is almost you know this is almost this is more than 20x what ChadGPT4 can do so this is really really big and um and that actually has a lot of really powerful use cases that i want to talk about in a second but really quickly um something else that i think is really important is the fact that claude 2's knowledge cutoff is in the early 2023s so you know gpt4 has september 21 it doesn't know anything that happened after that this one seems a lot more fresh because it's you know this year relatively early this year it has kind of its knowledge cutoff and so you get just a little bit more updated information.

4:04Not perfect, of course, like there's things that have happened in the last handful of months, but it does have a little bit more accurate, more updated info. I wonder if ChatGPT or GPT-4 is going to try to feed some updated info into their model or if they're just going to keep that hard cut off date. In any case, Claude2 also outperforms GPT on the GRE writing on human eval coding and bar exam. So it, you know, they have a number of areas where they've obviously prior to launching have tested and really focused on making sure that they were doing better than chat gpt of course you know this platform can't be just as good as gpt4 it has to be better otherwise they're not going to be able to get users who are already you know on that platform so i think that's a big that's a big thing for them another really interesting thing it has you know just this week we had um chat gpt announced code interpreter which allows it to run code and kind of interpret code right on in GPT-4.

5:04And so Claude 2 does have very similar abilities to ChatGPT's code interpreter. And I'll talk a little bit about some of those use cases in a minute. But one thing that I think is absolutely massive is the ability for it to import an entire PDF document and have it like read the document and work on it. This is something that gpt4 does not do right it only does whatever you give it in the text prompt so this is you're able to upload entire documents very easily and you're able to upload multiple documents and ask questions about them which is really cool um and then the last thing i wanted to say as far as just features on this thing from a high level is that it's around five times cheaper than gpt4 for developers to use which is is pretty significant especially when you're a company that's kind of focusing on your margins and you do a lot of you're really heavy on ai these api calls these different companies can be pretty expensive it can be a big chunk of your revenue and so being able to make something five times cheaper definitely is a big draw if they have a lot of really cool uses that gpt4 doesn't do and then they're cheaper like that's that's kind of where they have to be in order to get people over to their platform but that is a pretty compelling draw for people to go over and actually try this new clod2 platform so i was looking at what clod2 was kind of capable of doing.

6:23And one really interesting use case I saw in their demo, they are demoing it and they essentially said this to it. They said, I am a scientist or a science enthusiast. Can you explain the importance of the first included paper, describe its exciting new results to me in simple terms. For the second paper, please make a two column markdown table for me of the section headings in the paper and a corresponding detailed highlight from each section. Then they upload two separate PDF documents into Claude 2, and then it is able to essentially generate exactly what it's looking for. It's able to generate, you know, something describing the results from the first scientific paper.

7:05It's able to do the two-column markdown table based off of the second paper. And so I think what's amazing here is the fact that you can not only put PDFs in there, because that alone is really cool, but you can put multiple in there and ask it to compare them contrast them see the differences pull something out of one pull something out of another you're able to get like a lot of really rich detail like this is absolutely an incredible use case for so many people scientists people that work with data are going to love this because you can put spreadsheets um you can put pdfs you can put all sorts of data into this thing and it will just and then ask it questions about that data and I think that people have been making use cases with GPT-4 that can do some sort of similar versions of this but this is really taking it to the next level especially with the fact that you can put 75 ,000 words and have it you know review them into this thing so I've seen it I've seen people post in time and does it very fast so I've seen people post the entire text of the Great Gatsby into Anthropic.

8:09And then they changed one word in it, republished it, put it in there and said, hey, you know, find the one word difference or whatever, and is able to locate where something had been changed. And I did a very, very quickly. So it's really good with these like kind of massive document sizes. So if you're doing really big studies that have a ton of data that's or, you know, papers that are super dense, this is an awesome way to handle that. That's one amazing use case another really amazing use case i'm seeing out of claude 2 um is working with code so similar to you know something you could do with gpt4 with their code interpreter you can work with code in claude 2 and this is what something interesting i saw someone recently do on their demo they essentially said can you help okay so pretty much they said what does this file do they uploaded a file um which was a js file and they said you know what does it do and then anthropic said or claude said this map.js file renders an interactive world map visualization using d3.js some key things it does and then it listed out some key things that this thing did but essentially the code that they gave it was a giant you can kind of imagine like a giant interactive map of the globe with all the different countries in it, right?

9:28And after it kind of gave a description of what the file did, they next asked Claude to write code that added a mouseover to their interactive map. So they said, can you help add a mouseover tooltip that will display the value of a similar score column with the title similarity score? It then did that very quickly, and it pulled off the whole thing. They were able to copy and paste the code, and they were able to run the whole thing. Now, that being said, they did copy and paste it somewhere and run it on, you know, a local host. So I'd be curious to see if eventually they'll allow you to run code on their platform.

10:05That would be something kind of interesting. But it was, you know, they were able to do it. They ran the code and it worked. It was fully functional. They had the full, you know, they had the full map. And when they hovered over every single country, they had a similarity score, which is based off of something in their data. And you were able to hover over pretty much every country and and you know see this kind of interactive map which is really cool so it's really fun when you can like upload your code and then ask them what it does to make sure it's doing the right thing and then ask it to do you know make some sort of change to it they make some sort of change and they ran the thing and it actually worked so overall i think this is really really exciting um the my my big takeaway right here is you know some people are going to be like oh well i don't know they're going to say there's pros and cons of gbt4 and claude and of course that's the way it should be but as is claude 2 does a lot of things that gpt4 cannot do it does a lot of really impressive really cool things and i think the overall takeaway here is that this is um this is a big win for ai this is a big win for the ai industry and it's consumers who are going to win at the end of the day here because the more competition we have in this space the faster ai development will happen um and the better tools and products will have so overall i I think this is a really big win for everybody.

11:20I'm really excited. I'm going to do a more of a deep dive on this in the future, similar to things I've done with tools like Inflections Pi, where I kind of tested some of its ethical theories. And we found out that Inflections Pi model values a bee's life just as the same value of a human life, which is very concerning, especially if we're going to integrate that into any health care or military applications. and so you know we've been trying to raise awareness around that i've been reaching out on tweeting at people from the team over there talking to them over on linkedin trying to get them to fix that in their ai model we'll do something similar to that with claude and kind of run it through the ringer and test it as well but in the meantime i think from a performance perspective overall anthropic does a lot of really powerful things with the new claude 2 model that like frankly speaking this is what you would expect to see out of like a first tier ai model um when you're looking at like inflection and their goal is to be like their goal is to be everyone's personal assistant but when you try it it's kind of just a chatbot that is like someone could argue it's more empathetic but also it's terrifying because it doesn't value your life more than an animal's so aside from that fact it's really just a chatbot it doesn't do a lot of these cool features with big text uploads you know data analysis it doesn't do some of these code implications and use cases so I would have expected with the amount of money that and through the inflections pi has raised they'd be able to do something more like this out of anthropics cloud 2 and overall big kudos to the team at anthropic I think you did a really good job pulling off something that is more more powerful use cases for some things than GPT-4 that's not an easy thing to do obviously chat gpt is a massive product that is that is really successful and so being able to compete with them on so many different levels and come up with so many different new features that they're currently not offering are able to do is a really big win so kudos to the team over anthropic and you definitely should test out claude 2 and see how this might be able to be used in your business