Meet Devin, the world’s first AI software engineer who writes, debugs, and deploys code.

US-based startup Cognition has announced Devin, an artificial intelligence (AI) software engineer that promises to revolutionize the way code is written, debugged, and deployed.

The AI ​​assistant is being hailed as the world’s first AI of its kind, as it can take simple prompts and turn them into a functioning website or software program.

Founders Fund-backed startup claims its AI assistant is the new cutting edge in SWE bench coding benchmarks, passed practical engineering interviews from leading AI companies, and even completed real work on Upwork doing.

SWE-bench is an evaluation framework that encompasses 2,294 software engineering problems extracted from authentic GitHub problems and corresponding pull requests in 12 popular Python repositories.

Let’s take a closer look at the capabilities of this modern AI model.

What is Devin?

Devin is a software development assistant that goes beyond just code completion and suggestions. Unlike large-scale language models (LLMs) such as OpenAI’s ChatGPT or Google’s Gemini, Devin can handle the entire process of building and launching full-fledged software applications autonomously.

Also read: What powers ChatGPT and Bard? Overview of LLM or Large Language Models

With its own command line, code editor, and browser, Devin works independently to systematically implement software by breaking down tasks into stages.

What can Devin do?

Devin can plan and execute complex engineering projects involving thousands of decisions. Learn from experience, recall relevant context, and even correct your own mistakes.

In a demonstration shared by the company, Devin showed off the ability to add a print statement when an error occurs to quickly fix the problem.

A print statement is a line of code that is used to print a message to the console, helping the programmer identify where a problem occurs in the code. Devin also actively collaborates with users. Report progress in real time, accept feedback, and collaborate with users on design choices as needed.

AI assistants can also quickly learn unfamiliar technologies after reading articles. You can also train and fine-tune language models at scale by pulling instructions from research repositories.

What about performance?

Although the company has not released a detailed technical report yet, it claims that Devin has excellent coding skills and can solve coding problems at a much higher rate compared to other AI models. Challenging His SWE Bench His Coding Evaluated by his benchmarks, Devin succeeded in solving real problems found in his G itHub in open source projects.

It had an impressive problem resolution rate of 13.86%, beating out competing products such as Anthropic’s Claude 2 at 4.8% and GPT-4 at 1.8%.

Who is supporting the startup?

The AI ​​startup founded by Scott Wu, Steven Hao, and Walden Yan has raised $21 million in a Series A funding round led by Peter Thiel’s venture capital firm Founders Fund. Other investors include former Twitter executive Elad Gil and Sara Guo, founder of venture capital firm Conviction.

The founding team of 10 members, all International Information Olympiad gold medal holders, have worked at the cutting edge of applied AI with companies such as Cursor, Scale AI, Lunchclub, Modal, Google DeepMind, Waymo, and Nuro. I did.

Event alert: Moneycontrol and CNBC TV18 will host the ultimate event on Artificial Intelligence, bringing together entrepreneurs, ecosystem enablers, policy makers, industry leaders and innovators in Pune on March 27. Click here to register and access AI Alliance Pune Chapter.