• By quade
• 21 March 2024

Cognition AI Claims It Can Create Websites and Videos Games Using AI

With AI now capable of generating images and videos, there is plenty of buzz around tech companies regarding the next step of Artificial Intelligence. This comes at an important time as many experts question whether all this AI development is sustainable or if they have just found themselves in another major tech bubble. 

One way to ensure the success of AI is by continuing to innovate and improve the product. That is why so many companies have been set on showing off the capabilities of their products, whether through creative means like image and video generation or by automating tasks such as coding. 

Today, one of the newest AI companies, Cognition AI Inc. has come out with their product and claims that it can do something few others can, create video games and websites using AI.

Who Is Cognition AI Inc?

Compared to other players in the AI industry like Nvdia and OpenAI, Cognition is a relative newcomer to the field. The corporation was founded about two months ago with a staff of 10 who spend most of their time either in home offices or AirBnBs. with such a small company, it might seem surprising to hear they’re trying to take the spotlight now, but they have some huge backers.

So far, Cognition AI Inc. has raised about $21 million from its backers including Peter Thiel, the founder of PayPal, and Elad Gil, the former Twitter executive. 

The people in charge are Scott Wu, the CEO of Cognition, Steven Hao, the CTO, and Walden Yan, the CPO. Though young, these three men are ambitious and strongly believe in this project. Before starting this company, Steven was one of the top engineers in Scale AI, a business responsible for training these AI models. Walden meanwhile had given up his studies in Harvard to work in this company. Scott and his brother Neal are world-renowned coders, having won multiple international competitions. They now use this talent to make this AI dream a reality.

They aren’t even the only sports coders working on this project. The Cognition staff has won 10 gold medals in coding between them, a fact that they believe will give them an edge in their business. Scot talked about this recently.

Teaching AI to be a programmer is actually a very deep algorithmic problem that requires the system to make complex decisions and look a few steps into the future to decide what route it should pick. It’s almost like this game that we’ve all been playing in our minds for years, and now there’s this chance to code it into an AI system.”

-Scott Wu, CEO of Cognition AI Inc.

What Cognition AI Has to Offer

The flagship product of Cognition AI is a new AI tool called Devin. This program is meant to serve as a software development assistant, similar to Copilot, but more advanced. What this means for your business is that Devin will not just give you suggestions and autocorrects, but handle entire projects.

This is a highly ambitious goal as no other AI model can do this yet, but the Cognition team’s dream is for it to handle projects by itself. This means you can give the AI instructions such as “Create a website that maps all the Italian restaurants in Sydney,” and it will do it for your business. With those instructions, the AI will search all the restaurant’s addresses, build the website, and add it there.

Don’t worry about issues with this task as Devin will also show its work when you give it a task. This will allow your business to review the code and fix any bugs that might appear. 

The other big innovation they have boasted about is improving the AI’s ability to reason. Being able to reason means the AI can think and rationalize what you are saying. Most AIs have simplistic designs that cause them to view things literally, but the ability to reason is said to be the next major step in improving the system.

This can be seen in how Devin handles jobs. You can give it tasks natural language commands and it can accomplish them. While it works, it will also explain its plan and show the code. That setup allows you to add new prompts and fix the issue. Devin will then incorporate the feedback as it works. So far, Devin can go through hundreds of tasks without losing focus, something other AIs struggle with.

Some tests have shown that Devin can create simple websites within 5 to 10 minutes and even create simple games like Pong After that, you can adjust its appearance and performance by adding additional prompts. The tests showed that you can make cosmetic changes to the website and adjust the game’s physics. 

Other AI experts have claimed that Devin has taken a leap beyond just being an AI assistant and has become its worker with its capabilities. It can now prototype projects, fix bugs, and display data without getting derailed.

“This feels very different because it’s an autonomous system that can do something for you. Most of the other assistants derail after four or five steps, but this maintains its state almost effortlessly through the whole job,”

– Silas Alberti, computer scientist

The AI Coding Race

When asked to explain how Cognition AI made these innovations, Scott declined to reveal their secrets. All he said was that their team managed to find a way to combine e large language models (LLMs) such as OpenAI’s GPT-4 with reinforcement learning techniques.

“It’s obviously something that people in this space have thought about for a long time. It’s very dependent on the models and the approach and getting things to align just right.”

-Scott Wu, CEO of Cognition AI Inc.

Cognition isn’t the only one working to create an AI coder. Magic AI, another startup has already raised  $100 million to create what they describe as “a superhuman software engineer.”

They explain that their company is not using other LLMs for influence, but creating the model and all the technology from scratch. However, this model is still young, and the data hasn’t been released.

With the huge success of AI systems, many are wondering what it means for their livelihoods. Some believe coding assistants can help programmers from having to handle menial tasks or allow people with little coding experience to create their websites and apps. However, others claim this could destroy the coding workforce and leave thousands out of the job.

For AI companies, it is still too soon to give a definitive answer about how useful these systems can be. Right now, AI has just reached the point where it can operate somewhat independently, nobody knows how far the technology will go.