Today, OpenAI announced that they are launching their new GPT-4. This is probably one of the most highly anticipated pieces of technology in tech. So we're going to break down exactly what that means, how it works, and what it can do better than the original ChatGPT. So just as a quick little refresher, when ChatGPT launched, originally it was on GPT-3.
They did a slight update to do GPT-3.5, some slight updates, and now they have officially launched GPT-4. So in order to use it, you have to be a pain premium user of ChatGPT, and then you can opt into it. And it's got a lot of cool new features, but, you know, I would say for anyone that thinks this is going to be like the next thing and AI and robots are going to take over everything, as I've been saying for a while on the podcast, probably a few weeks ago, maybe a month ago, Sam Wallman, CEO of OpenAI, came out and said, you know, for people that think this is the case with GPT-4, they're just getting their hopes up, and it's not as maybe crazy as some might think. So as far as the breakdown really quickly of exactly what is different with this, number one, it has a lot better reasoning capabilities.
Number two, it can take images as inputs. So you can feed it an image and say, based off this image, blah, blah, blah, blah, so like mixing an image and text. And the third thing it can do is that they increase the text it can generate and receive, and I'll explain what the difference is on that, from 3,000 words to 25,000 words. So it could technically generate a 25,000 word article off of a single prompt or piece of content, essentially, right, if you prompted it the right way and did what you needed.
So what's interesting about the word increase there is specifically for companies and software that are built on top of ChatGPT, the way it works is because there's so much computational power, the prompt that you give it and the text, the outputs, are combined together. So if you have, you know, a really sophisticated, for example, I have like a sophisticated life coach, and the way it works is we have like a full page prompt we've written instructing it how do you life coach and how that whole thing works. And so if you combine all the text there, whatever the output it gives you can only be previously 3,000 words with what you gave it plus what output. So now they've increased this, this gives the opportunity not just for it to create longer content, but for you to be able to input much, much bigger prompts, much more complex prompts.
And the fact that it can understand, right, like 25,000 words, you could probably put like a massive chunk of a book in here and have it write a summary versus just like a single page article. You know, you could get like whole chapter summaries on stuff. So it's really interesting to see what happens just alone with the increase in text that they've kind of bumped it up to. Another thing they say is that it's much more safe.
I think ChatGPT in their kind of like documentation on this said that if you, you know, it's like 82% less likely to respond to requests for what they call disallowed content, which to be honest pretty much just means they had more people manually go and put manual safeguards, which has been criticized a lot, right? A lot of it is politically motivated or essentially, you know, biased towards whatever they deem is socially acceptable or whatever is okay content. So, I mean, you know, that is taken with a grain of salt. A handful of people I've talked to are not thrilled by just the more, you know, like people, pretty much a lot of people are not excited when you say like, oh, I made my thing more safe.
It's like safe from who? Like, you know, if you're an adult, you can just handle what you get out of it. So anyways, that's a grain of salt. And maybe some people like to be kind of babied in that way in the content they see or whatnot.
And, you know, definitely for children and for kids using it, that might be useful for adults. You know, some adults just want to be treated like adults, but I know that's not for everyone. In any case, off of my spiel on that. One of the things ChatGPT did is, this kind of a funny thing they threw out there on their kind of like launch notes, is that ChatGPT took the bar exam and it was in the 10th percentile, so it's just like the OG ChatGPT.
And with the SnoopGPT4, it took the bar exam and it was in the 90th percentile. So obviously, this is kind of like reasoning capabilities that they're putting out there. Have actually made some big impacts with like academic stuff, which I think is important because they're really trying to push this thing into kind of a tool for educators and education. Overall, I would say I'm pretty glad to see that they're kind of pushing for this rapid incremental change.
Like it wasn't that long ago we had GPT3, GPT3.5, but they're iterating very fast, making a lot of updates. So I think this is a really solid update at a very fast pace. I really can't complain. But like that being said, this isn't the apocalypse.
This thing isn't, you know, super genius. It is interesting, though, a couple companies they've been integrating with. Number one, Duolingo. They have it to kind of, I guess they say, to deepen the conversation on Duolingo, really to kind of create a chatbot that helps you when you're learning different languages or whatnot.
The other thing is they're working with a company called Be My Eyes. And essentially, it's for blind people to take a picture of something. And then there's a bunch of volunteers that can go and help, you know, learn or help tell people what exactly the blind person took a picture of and make sure that, you know, they're buying the right thing or making the right choice or whatever. So it's a really awesome kind of, I think it's a non-profit organization.
In any case, they're working with them because now that you can do the image inputs, they're helping those people to know exactly what they're seeing, right? So you may not actually need volunteers to tell you exactly what you're taking a picture of because now opening iChatGPT could help people with visual and vision disabilities to see better. So that's kind of interesting. Another area is Stripe.
Stripe is leveraging the GPT-4 to pretty much streamline user experience and to combat fraud. So that is obviously a very powerful use case. We'll dive more into exactly what that means. Morgan Stanley, the bank, you know, they're using wealth management tools and they deploy GPT-4 to organize a lot of their knowledge base.
So that's very interesting. Khan Academy, right? Like I was talking about, they're kind of pushing into education. Khan Academy integrates GPT-4 as every student's customized tutor, right?
So now that this thing can, if you think about it, it's kind of interesting because they have a handful of companies that they obviously gave GPT-4 to ahead of time. Obviously well selected, right? Like we're talking a bank or Stripe or also the government of Iceland to help preserve its language. Khan Academy, which is a non-profit education company.
Be My Eyes, a non-profit for people with disabilities. Duolingo for learning new language, right? I feel like they've really specifically chosen A, like companies that kind of have goodwill, maybe less Stripe and Morgan Stanley, but like Khan Academy, preserving the Icelandic language, you know, learning new language, Duolingo, Be My Eyes, a non-profit, right? So I feel like they've really intentionally chosen a number of specific companies.
But what's interesting is it would appear that a lot of their new features they've added with GPT-4 also directly apply to these companies, right? Like Be My Eyes is probably one of the big motivators for them to allow image inputs. And then we had Dolly too, which classified images, so obviously the technology's not that far off, so like they invented something too crazy. But it'll be interesting to see what they continue to do, how this continues to change and improve in the future.
But all in all, a very solid technology and some serious progress they've made so far.