Diving into Devin - World's First AI Software Programmer

Introduction

The days of programmers scratching their heads over code and searching the internet for answers are long gone. The day when programs write their own programs appears to be here. Presenting DEVIN, the newest AI-powered software developer from Cognition Labs. Devin has risen to what some programmers might call the "Devil" of the software industry. But precisely what is DEVIN? Who are the geniuses that founded Cognition Labs? How does everything function? Don't worry, we'll tackle these important topics in this post along with a dash of fun and wit.

A Peek at Cognition Labs ????

Imagine a group of highly skilled and competitive programmers, such as Neal Wu, Scott Wu, Andrew He, Steven Hao, Walden Yan, Adhyyan Sekhsaria (check here).The goal of this dream team's US-based firm, Cognition , is to use clever artificial intelligence (AI) to transform the software industry. They have an impressive $21 million in funding, so they have the resources to support their goals. That being said, not much is known about them as of yet, much like a magician producing a rabbit out of a hat.

Who's DEVIN Anyway? ??

Cognition Labs recently introduced DEVIN, an advanced software engineer extraordinaire, in a blog post on X. They assert that DEVIN has excelled in several technical interviews and has even done freelance work on Upwork (although we have yet to receive the LinkedIn recommendation). Maybe, it is too early to tell if these claims are true, the intriguing features shown in their promotional film point to a bright future for DEVIN.

DEVIN's Bag of Tricks ???

Any programmer would be in a state of disbelief when they saw the wealth of capabilities that DEVIN offers. DEVIN is a comprehensive solution for all your coding requirements, featuring an integrated shell, code editor, and web browser. Imagine spending a few hours to setup VScode, Git, etc. while Devin completes your whole job in the meantime (Pretty scary, right). It uses the shell (similar to linux terminal) to create repos, work with directories, and do similar stuff. There are many of compilers included with the code editor, so you should not worry about that either. The web browser is the greatest power. It conducts an independent search of the entire web and turns up some pertinent content. Nothing remained, as far as I can tell. Who knows what could be added in the future if needed? (However, nothing appears to be omitted.)

DEVIN's Race to the Top ??

But can DEVIN walk the walk or just talk the talk? The SWE-bench provides a set of task instances, which include GitHub issues and requires the AI models to solve them. It's pretty hefty for Language models to solve GitHub issues, mainly due to restrictions of capabilities and data used to train them. Read the paper about SWE-Bench here. Devin actually was able to solve 13.86% of the issues unassisted, which is much higher than previous 1.96% (unassisted, by Claude 2 + BM25 Retrieval) and 4.8% (assisted, by Claude 2). This significant rise in %age is enough to provide claims about the powers by Devin.

Devin Tested Actually

In a daring feat of bravery (or perhaps madness), Andrew Kean takes DEVIN for a spin and lives to tell the tale. Je was able to gain early access on Devin and got a few conclusions. Here are the few tasks Andrew performed on Devn:

• Creating a website where you play chess against an LLM (GPT-4) - Devin got stuck somewhere in the middle, but asks for the API key to GPT-4 :).
• Creating a map of Antarctica sea temperatures in the last fifty years - Successfully deployed the whole Application on Netlify
• Make a chrome extension to convert GitHub repos to Claude prompt - Actually works and the extension is available on the X-thread.

The above samples tested by Andrew are very much evident of the things Devin can do right now.

Conclusion: A Brave New World ????

One thing is certain as we welcome the emergence of AI programmers and say adieu to the manual coding era: DEVIN is leading the way. DEVIN opens up a world of possibilities by bridging the gap between AI models and human developers with its flawless performance and amazing capacities. Thus, fasten your seatbelt and prepare for an incredible journey, for DEVIN is the programming of the future. And who knows, maybe DEVIN may DM you one day to say how much you like this article. ???

Thank you for reading this....