We know of a new GPT-4 speech model that can handle both text and image input or can play the roles the user creates for it. How it’s used was demonstrated a few days ago during a presentation by the president of OpenAI, the company behind the popular artificial intelligence. But since then, people have been racing to find other cases where GPT-4 could help. Below are the options. There is the ability to make recipes out of leftovers from the fridge, program computer games or help with homework.

In addition to the new ability to enter commands into models in both text and image form, GPT-4 can describe desired actions more comprehensively. According to OpenAI, it can handle up to 25,000 words, which is a marked change from around 1,500 words on the GPT-3.5 model. In order for the model to have much better capabilities and fewer errors, OpenAI uses the word hallucination. According to the company, it should be 40 percent more accurate than the previous model. However, it is true that his work needs to be examined and verified.

The important thing we don’t know about GPT-4 is its size. The predecessor works with about 175 billion parameters. OpenAI also hasn’t publicly explained how GPT-4 was built. Not a word about the data, computing power, or how to train the model. For direct inquiries from Will Heaven of MIT Technology Overview regarding the information, the GPT-4 development team replied: “We can’t comment on that right now. We have quite a few competitors.”

GPT-4 is currently available for free only after registering at waiting list, if you don’t want to wait, you can buy a ChatGPT Plus subscription where the new model is available. GPT-4 is also for the internet search engine Bing joined Microsoft. The public is now testing the capabilities of GPT-4 and finding more ways to use it, of which there are countless. Therefore we have selected a few of them and explained them clearly.

Programming

It’s not that earlier models from OpenAI or other similar ones couldn’t help with writing code, but the simplicity of the procedure offered by the new models thanks to the ability to upload images on login. The President of OpenAI pointed it out during a demo – you want to create a website design, so you sketch out on paper how it will look. You take a picture of the sketch, upload it to GPT-4 and ask it to write code.

In an instant, you have working code that you can instantly turn into the product in question. However, since the model is not error-resistant, it’s possible that the code won’t work for some reason. Even in that case, you don’t need to lose your head, even if you don’t understand programming. For GPT-4, you only need to write, for example, what errors the programming software reports to you, and the model itself will offer a solution. Moreover, if you instruct him to explain in simple terms how to solve the problem, he will actually give you such a procedure.

I recently saw GPT-4 turn a hand-drawn sketch into a functional website.

This is crazy. pic.twitter.com/P5nSjrk7Wn

— Rowan Cheung (@rowancheung) March 14, 2023

Although the website is relatively simple, GPT-4 can also handle, for example, computer game code. The classic game Pong is said to have been created by artificial intelligence designer Pietro Schirano with the help of GPT-4 in just one minute. You can see the results below.

So if you want to program something and don’t know how, now you have a place to start. Thanks to another feature of the new model, you can tell him how to behave before starting a conversation. So, for example, you could ask him to not only write code for you, but also explain what each part of it means and why the code exists.

I don’t care that it’s not AGI, GPT-4 is an amazing and transformative technology.

I recreated the Pong game in under 60 seconds.
It was my first try.

Things will never be the same. #gpt4 pic.twitter.com/8YMUK0UQmd

— Pietro Schirano (@skirano) March 14, 2023

Explanation of graphs and tables

Also, when you first look at an Excel spreadsheet or some charts, do you ever wonder what you’re actually looking at? Ask GPT-4. The procedure is similar to the previous case. Upload a photo of a graphic or table, type in and let the program explain the image to you, and you’re done.

But artificial intelligence can also cope with somewhat ridiculous tasks. If you ever didn’t understand a picture that was supposed to be funny and were afraid to ask your friends for an explanation, now you can simply upload the picture to the system and ask them to explain what the funny aspect is.

A teacher for every household

Thanks to the flexibility of the aforementioned GPT-4 speech model, you can also disable it from directly answering your questions, and instead instruct it to guide you to the correct answer through a series of questions in the style of the ancient philosopher Socrates’ teachings. . It is enough to describe the role intended for him in the so-called system messages before the conversation.

For example, the developers from OpenAI wanted the model to lead them to the correct solution of linear equations in the spirit of the famous thinker, and not tell them the result right away. The system then actually guides them and even rejects direct calls to simply reveal the answer.

Cook

Everyone has probably thought about what you can cook with leftovers in the fridge. Now GPT-4 will help you with this. All you have to do is take a picture of the food left in the fridge, place the model as the cook via text entry and ask what kind of food can be made from the ingredients.

Another Twitter user tried just that, to whom the program recommended five ingredients-based recipes — fruit salad, omelet with cheese, sandwich with ham and cheese, and fruit smoothie.

Here’s how I gave a GPT-4 a photo of his fridge and asked him to make a food recipe in under 60 seconds. pic.twitter.com/CLBCWTJEab

— Sudharshan (@sudu_cb) March 15, 2023