US-based startup launches Devin, which they are calling the ‘first AI software engineer’. This is a revolutionary advancement for the developer community as now you can have an entire source code generated from just one line of code prompt. Quite amazing isn’t it? Let’s find out more about all of its features!
Highlights:
- Cognition unveils Devin, the world’s first complete AI software engineer.
- Can generate several lines of code, perform debugging, and also deployment.
- Can also fix major repositories and work on generating real-time models based on existing algorithms.
Meet Devin AI Coding Assistant
Devin is a Generative AI Coding Assistant developed by Cognition that can write and deploy codes of up to hundreds of lines with just a single prompt.
Although there are some similar tools for the same purpose such as Microsoft’s Copilot, Devin is quite the advancement as it not only generates the source code for software or website but it debugs the end-to-end before the final execution.
The CEO of Cognition Labs, Scott Wu, demonstrates in one video how users can monitor the AI tool’s code editor, workflow, and command line as it goes through several stages to finish data research and coding projects:
Here is what they said about this major development in their official announcement:
“Meet Devin, the world’s first fully autonomous AI software engineer. Devin is a tireless, skilled teammate, equally ready to build alongside you or independently complete tasks for you to review. With Devin, engineers can focus on more interesting problems and engineering teams can strive for more ambitious goals.”
Devin’s never-seen-before features make the work of previous tools such as Github and Microsoft’s Copilot much easier and complete. When it gets a request, it can search the internet for tutorials to learn how to accomplish jobs and troubleshoot problems that arise on its own while engineering. However, users can step in if they so want.
Talking about the startup, Cognition Labs is currently a small team who were medalists in the International Olympiad in Informatics (IOI).
How to Get Access to Devin AI?
Cognition has still kept Devin in its early access period. Very few users and tech enthusiasts have had access to the state-of-the-art model via Cognition Labs developers.
However, you can request early access to Devin for engineering and development-related purposes by filling out this Google form. You can also contact Cognition regarding any updates or access-related issues via their e-mail: [email protected].
In the Google form, you can state or your designations and explain in detail what task you would like to give Devin. Some of the user requests are being evaluated by Cognition’s development team. Based on your type of request your waitlist for access will be queued.
While it is still in the testing phase, developers are sharing some early reactions about Devin AI who got to give it a try.
Recently, we also got StarCoder2 AI, which is also for code generation.
What Can Devin AI Do?
Devin’s chatbot-style interface just requires the human user to prompt what code they want, just like. From there, it takes over, creating a comprehensive, step-by-step strategy to solve the problem.
It starts the project by utilizing its developer tools in the same way that a human would, generating its code, resolving problems, testing, and providing real-time progress reports so that the user can monitor everything as it operates.
It is also possible for the user to enter the chat interface and instruct the AI to correct anything that appears incorrect to a human observer. This allows engineering teams to assign the AI part of their projects so they can concentrate on more innovative jobs that call for human expertise.
However, Devin also comes with several other capabilities. We have discussed each of them in detail further looking into the ways how developers can benefit from them.
Learning From Unfamiliar Technologies
Devin has been programmed to learn how to use unfamiliar technologies. All it needs to do is interact with blog posts or documentation regarding the technology to be learned. Look at the video where a user named Sara can be seen giving a prompt link to Devin regarding a blog post on generating images with hidden texts in them.
She further asks it to use the same technology to generate images with hidden texts and it does a wonderful task in using ControlNet on Modal to generate pictures for Sara that include hidden messages.
End-to-End App Deployment
Devin can deploy end-to-end applications as requested by the user. User can make necessary changes or recommendations based on their preferences. The AI will do an outstanding job of reflecting those changes with ease and even making adjustments for enhanced optimization.
In the video below Devin makes an interactive website that simulates the Game of Life! It incrementally adds features requested by the user and then deploys the app to Netlify:
As developers, you can take inspiration from any website or tool algorithm and have it reflected in your website or application of choice. This cutting-edge technology is doing wonders!
AI Model Fine-Tuning
If you are a developer who is always looking to fine-tune existing AI models with the latest algorithms and technologies for better-personalized models, then this AI engineer may just be the tool for you.
One of Devin’s outstanding features comes in the form of AI fine-tuning, where you can ask the tool to enhance or optimize any existing AI model with the help of some reference algorithm links or blog posts.
Take a look at the video below where a user can be seen asking Devin to fine-tune an existing 7B Llama model using a GitHub repository link on QLoRA, a method for enhancing Llama models:
Devin does great work in training the Llama model by creating a virtual environment and cloning the repository. It also monitors the fine-tuning process and informs the user about the current training status from time to time.
Debugging Major Repositories
Devin is also capable of fixing coding errors of various types on your repositories. The error can range from any general coding syntax error to even mathematical or reasoning errors. This will save a lot of time for developers who find a hard time rectifying mistakes from GitHub repositories for both their projects and deployment work.
Setting up Real-Time Models
Devin can be used for inferencing real-time models based on developer requests. This will allow you to deploy real-time open-source codes based on various aspects starting from Computer Vision to Deep Learning and much more.
Comparing the Benchmarks
Devin was tested on the SWE benchmark. It is an important benchmark for measuring coding and debugging capabilities as it requires agents to fix actual GitHub issues from open-source projects like scikit-learn and Django.
Devin got an amazing result when it solved 13.86% of the issues end-to-end and beat several AI giants including GPT-4, Claude 2, and SWE’s Llama 7B and 13B parameter models:
Each of these models even needed help, where they were instructed on which file needed to be corrected. On the other hand, Devin did quite amazing in demonstrating its coding capabilities.
Conclusion
Devin is a huge advancement in Coding space that few years ago none of us would have imagined could be possible. However, it also hampers the future job prospects of several software developers in the future. So, we need to see how good it is and how it affects the industry!
