Cognition, a newly established AI firm, has introduced “Devin,” a fully autonomous AI software engineer. The company is supported by tech industry heavyweights including Tony Xu, co-founder of Doordash, and former Twitter CEO Elad Gil through the Founders Fund.
Although several coding assistants are available, such as the well-known Github Copilot, Devin is reputed to distinguish itself from the competition by managing full development projects from the beginning to the conclusion, from writing the code to resolving related issues to carrying them out to completion. According to the business, this is the first product of its sort and it can even handle assignments on Upwork.
What exactly can Devin do?
Cognition CEO Scott Wu shows out how users may see the model in action in a video that attached to a blog post introducing Devin. As it works step-by-step to finish extensive coding projects and data research tasks handed to it, they may observe its command line, code editor, and workflow.
According to the startup, human users only need to provide a basic description of the project via chatbot-style questions, and Devin takes care of everything. Like a human coder, but much more quickly, it starts by drafting a thorough, step-by-step plan for finishing the task at hand before utilizing its development tools. It can build its code, troubleshoot, test, and report in real-time, allowing users to always be aware of its progress.
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 user. According to Cognition, 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.
With its promise of offering engineers their own automated worker that can complete projects instead of a copilot that would just aid human coders with snippets or ideas, Devin’s launch represents a significant shift in the emerging AI-powered coding business. Devin is currently only accessible through private preview.
Capable of carrying out a range of developement tasks
Demos provided by Wu indicate that Devin in its current state is capable of handling a variety of jobs. This ranges from simpler tasks like setting up fine-tuning for a large language model using the link to a research repository on GitHub or learning how to use new technologies, to more complicated ones like deploying and improving apps/websites end-to-end and locating and fixing bugs in codebases.
In one instance, it discovered how to run the code to generate images with hidden messages from a blog post. In the meantime, it wrote and debugged the code for an Upwork project to run a computer vision model.
Can debug and test
In open-source repositories, Devin is an expert at testing and debugging code. It writes thorough test cases, moves around the codebase with ease, and uses sophisticated debugging techniques to find and fix problems when a specific defect is encountered. The AI software engineer guarantees that patches are effective and no new issues are introduced by rerunning tests and utilizing print statements, which saves developers a significant amount of time and effort.
Can fine tune large language models
Devin makes fine-tuning big language models, like the 7B llama model, a simple task. It automates the process of fitting models to particular tasks by cloning repositories, configuring dependencies, and executing training jobs. Devin ensures smooth training progress and provides timely status updates while troubleshooting problems such as CUDA issues by inspecting the environment and reinstalling programs.
The adage “learn to code” was once proposed as a substitute for those who lose their employment to AI, but it appears more out of date than ever as the AI era progresses. Devin’s developers think that in the future, it will be able to carry out a lot of simple coding tasks far more swiftly than human programmers.
With the closure of a $21 million Series A financing led by Founders Fund, Cognition now has capital as well. Companies interested in learning more about its capabilities are encouraged to apply via email as it is now trying to grow capacity and grant early access to more select users.
Leave a Reply
You must be logged in to post a comment.