According to Moneycontrol, Cognition, a US-based startup, has introduced Devin, an artificial intelligence (AI) software aiming to transform the coding process. This groundbreaking AI assistant can translate simple commands into operational websites or software programs.
Cognition, backed by the Founders Fund, asserts that Devin has not only successfully navigated practical engineering interviews at top AI firms but has also completed real tasks on platforms like Upwork. The software has set a new benchmark in coding proficiency on the SWE-bench coding benchmark.
SWE-bench is a dataset comprising over 2,000 software engineering problems extracted from GitHub issues and pull requests from 12 prominent Python repositories.
Devin, the software development assistant, exceeds traditional code completion tools. It manages the entire software development cycle independently, distinguishing itself from large language models (LLMs) such as Google’s Gemini or OpenAI’s ChatGPT. Equipped with its own code editor, command line, and browser, Devin autonomously executes software projects in manageable steps.
Capable of handling complex engineering projects and learning from its experiences, Devin can recall relevant contexts and rectify its own mistakes. In demonstrations, Devin swiftly addresses errors by adding print statements to code, aiding programmers in pinpointing issues. It also collaborates with users, offering real-time progress updates and design input.
Additionally, Devin quickly learns new technologies and can refine large language models based on research repositories.