People are Already Doing Amazing Things With GPT-4

GPT-4 is taking the world by storm. The latest iteration of OpenAI’s large language model looks to have unleashed a furry of creative adventures. So far it has been a little over a week, and it is incredible what some people have done with GPT-4. Many have taken to Twitter to show off what the new LLM is capable of and the results are quite interesting. So let’s go through a few examples of what people are doing with GPT-4.

Video Games

A user by the name of Pietro Schirano decided to show off his love for classic video games. He was able to recreate the game of Pong in under 60 seconds with the aid of GPT-4. According to his post, this was done with a single try. As you can see, this is an interesting use of GPT-4’s multi-model abilities:

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

I recreated the game of Pong 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

But it wasn’t just retro games such as Pong. Another user decided to take it to the next level with the classic first-person shooter, Doom. In a thread, Javi Lopez was able to fully recreate the 3D game with the aid of GPT-4. Not only that, but also documented the process and provided detailed instructions on how he utilized GPT-4 to accomplish this feat:

Create a basic 3D game prototype like Doom using GPT-4.

Let GPT do all the maths, raycasting and hard work!

🧵 A thread pic.twitter.com/IZgSlj4QTO

— Javi Lopez ⛩️ (@javilopen) March 15, 2023

There are other examples of similar users of GPT-4. A user by the name of Ammaar Reshi recreated the game Snake. According to his post, in twenty minutes and with no experience in JavaScript, he was able to create the game that runs a person’s browser. But it’s not just classic games, as some even went so far as to create their original games. In one instance a user, in under thirty minutes was able to create a new platform game “Ethereum in Bitcoin land.”

Finally, another user, Haofei Yu, created a GitHub repository where he and his team collected prompts that allow for instant game generation on GPT-4.

Web development & Coding

I just watched GPT-4 turn a hand-drawn sketch into a functional website.

This is insane. pic.twitter.com/P5nSjrk7Wn

— Rowan Cheung (@rowancheung) March 14, 2023

GPT-4 isn’t just creating websites from scratch. According to user BioBootloader, it can take your Python scripting to the next level with a new feature inspired by X-Men comic hero Wolverine. Aptly named Wolverine, the program “gives your python scripts regenerative healing abilities.” Though that might not sound like much, what it basically does is every time you run a script and it crashes, GPT-4-Powered Wolverine, both edits and explains what went wrong.

For those who are either intermediate or beginners learning Python, this could be a game changer as it breaks down what went wrong. Even for experienced programmers, this could help you debug at a rate unimaginable. Check out the post below:

Today I used GPT-4 to make "Wolverine" – it gives your python scripts regenerative healing abilities!

Run your scripts with it and when they crash, GPT-4 edits them and explains what went wrong. Even if you have many bugs it'll repeatedly rerun until everything is fixed pic.twitter.com/gN0X7pA2M2

— BioBootloader (@bio_bootloader) March 18, 2023

In another thread, the host of the YouTube channel On Chain Gaming, Jake Browatzke, was able to have GPT-4 help him create his own Google Chrome extension. According to his post, he had zero coding experience and GPT-4 was walk him through the entire coding process from start to finish.  Similar to Wolverine above, it fixed any error that came up while Jake was writing the code and made sure it worked. So, you might ask, what does this extension do? Well, it’s called “Pirate Summarizer” and it translates the text of any webpage into so-called “pirate speech.”

GPT-4 Still has limits

However, it’s important to note that GPT-4 is not without its limitations and challenges. As some have seen with the prior iteration, GPT-3, 4 can still produce unreliable output with biased information and “hallucinations” according to OpenAI’s own paper. This is due to issues such as a lack of information about events after September 2021, the cutoff data for the majority of its pre-training data.

Then there is the issue of “Jailbreaks.” GPT-4 is still vulnerable to adversarial prompts which create outputs that OpenAI had intended not to be generated. According to the GPT-4 Technical Report, GPT-4 users should take “great care” when using GPT-4 outputs based on this information. But with all that said, it’s clear that many users are being quite creative with OpenAI’s LLM.