Submit a comment
An AI-generated software engineer named Devin can program himself and work independently when assigned tasks, helping human engineers reduce their working time.
On March 13, Cognition Labs, a US-based artificial intelligence technology startup, announced the world's first AI software engineer. "Devin is a talented and tireless teammate who is willing to work alongside you or complete tasks himself for you to evaluate. With Devin, engineers can focus on more interesting problems and engineering teams can work on more ambitious goals," Cognition said.
This is the product of a group of engineers working at many large companies such as Google DeepMind, Waymo, Scale AI, led by engineers including Scott Wu, Steven Hao and Walden Yan.
Though it has been around for a while, Cognition Labs was only officially launched earlier this year and raised $21 million in funding. In its introduction, it describes itself as a “small but talented team,” with founders who have won 10 gold medals at the International Olympiad in Informatics.
Before Devin, the world also had AI assistants to support programming such as Github Copilot or chatbots ChatGPT, Google Bard. However, Devin is different because it can handle the entire development process from start to finish.
In an introductory video, Scott Wu, CEO of Cognition Labs, demonstrated how Devin users can write initial code, monitor code, test and debug, and even manage projects on platforms like Upwork.
Devin can search the internet to learn how to complete tasks, giving engineers a fully capable AI agent to work on projects instead of an assistant that can only suggest code snippets.
To use, users type a natural language command into Devin's chatbot-like interface, and the AI engineer takes over. It then creates a detailed plan to solve the problem step by step.
Next, Devin will launch the project using developer tools. It will write code, debug, test, and report progress in real time, allowing users to keep track of everything and prevent unexpected things from happening.
If they see a mistake, users can go into a chat interface and command the AI to make changes. This allows engineering teams to delegate some of their tasks to AI and focus on more creative projects that require human intelligence.
In Bloomberg's test, Devin was able to program a basic Pong game and create a website in less than 20 minutes.
According to the developer's announcement, when evaluated based on the SWE-Bench standard for solving issues on GitHub, Devin correctly solved 13.86% of issues without assistance. While with the previous model, the performance was only 1.96% and the highest was 4.80% with assistance.
However, Devin is not yet widely available, only to a select number of pre-registered users. The developer says this product is just the first step in developing its AI engineers.
“We are developing AI teammates with capabilities that go beyond what current AI tools can do. By solving the reasoning problem, we can open up new possibilities in many areas – code is just the beginning. We want to help people around the world turn their ideas into reality,” Cognition explains.
Cogniton Labs’ product is considered the clearest illustration of Nvidia founder Jensen Huang’s previous statement. At an event in late February, Mr. Huang shocked the world by advising young people not to learn to code.
According to him, humans are at the beginning of the AI revolution and programming is no longer an important skill. He believes that artificial intelligence is getting better at this job, so humans should focus on more valuable expertise such as biology, education, manufacturing or agriculture.
TB (according to VnExpress)