The other day OpenAI, the creator of ChatGPT, announced the GPT-4 model. The most advanced artificial intelligence model so far, GPT-4’s work is “jaw-dropping.
Table of Contents
The new artificial intelligence model, which has an arrow-mode structure, has received some very important improvements compared to its previous version, GPT-3.5.
The AI, which has been cleaned of its existing bugs, now accepts visual inputs, making the usage scenario almost limitless.
OpenAI, which owns AI tools such as ChatGPT and Dall-E, has already released GPT-4. As we mentioned in our previous news, GPT-4 has been included in Microsoft’s new Bing for a while. Now it is available to everyone via a waiting list in the ChatGPT Plus version.
OpenAI prepared a demo event to show what the new GPT-4 can do and opened the door to unlimited usage scenarios.
It’s up to you whether you want to walk through that door or not, but it’s clear that GPT-4 can be of great benefit to you no matter what you do.
On the other hand, warnings about the accuracy of the answers continue.
What Can GPT-4 Do?
GPT-4 is a model of artificial intelligence, and it is still far from being the “artificial general intelligence” we define as AGI.
But this is not a shortcoming, GPT-4 is an incredible “technology” in its own right. For example, it takes only 60 seconds to make Pong, one of the oldest arcade games made in 1972, with GPT-4.
From the model you can just ask “I want to create a Pong-like game, what’s the best language to do it in?
I also want to use Replit to build it” which is enough to make a game. But the possibilities are not limited to this.
It Can Be a Lawyer
It only takes a few minutes to build a case file with GPT-4.
In one example, GPT-4 powered ChatGPT was asked to prepare a federal lawsuit against a spam caller.
The damages requested in the lawsuit are 1500 dollars. In a few seconds, ChatGPT gives you a case file.
From Image to Website
If that’s not impressive enough, here’s a use case OpenAI demonstrated at its own demo event.
GPT-4 can turn a hand-drawn sketch on a small piece of paper into a functional website.
Because as we said above, GPT-4 can accept visual input. So you can add an image to the query and ask the model to do something similar or to describe the image. In this scenario, a small joke website was made.
Can Make Drug Discoveries
GPT-4 is also much more “exploratory”. For example, it is possible to ask
GPT-4 to make a drug discovery. It is possible to give an existing drug to
GPT-4 and ask it to find compounds with similar properties, and also to ensure that they are not patented. It can even create an order form for the supply of the found compounds.
This feature is also a bad feature, depending on how you look at it. It can be used maliciously, but if it is used ethically, great things can happen.
GPT-4 can be your savior if you’re not good with contracts.
In one example, you ask for the shortcomings of the Ethereum contract, and in a few seconds, the model highlights the shortcomings, areas that can be exploited, and possible ways to take advantage of them.
Can Improve Dating Apps
With GPT-4, it seems that more personalized, targeted ads will enter our lives.
We actually see a small section of this in the example of a dating app. It is ensured that the most accurate matching can be made by analyzing the data input of the users.
In conclusion, I cannot say that GPT-4 is a radical innovation, but I can say that tons of things have been put on top of the GPT-3.5 model.
Whether you want to make a Pong game, or a snake game, create a case file or turn a paper sketch into a real website. The options are very wide.
On the other hand, some important vulnerabilities in previous versions have been fixed. GPT-4 is more attentive to sensitive queries, freer of bias, and less violent rhetoric.
OpenAI emphasizes that the new model is not perfect, so it is still possible to see the harmful speech.
In addition, while there are improvements in the “accuracy of information” part, it is important to investigate the accuracy of the output and keep in mind that it may be inaccurate.
For more details, watch the demo video below or check out the GPT-4 white paper in the source section.