Has Codeium Cracked the Code for AI Assistants?

In terms of AI-powered coding assistants, Microsoft’s Copilot has the identify and the numbers. However a competitor referred to as Codeium is rising rapidly, and in line with its co-founder and CEO Varun Mohan, the sky is the restrict for AI assistants.

Codeium began life in 2021 as Exafunction, an infrastructure startup that offered large compute for different corporations growing deep studying techniques. Mohan and his enterprise associate, Douglas Chen, had acquired 10,000 GPUs they usually primarily rented the processors to autonomous automobile growth corporations, an business they beforehand labored in.

However by late 2022, ChatGPT had exploded onto the scene, and Mohan and Chen realized that the transformer mannequin–the Google-developed mannequin powering the big langauge fashions (LLM) breakthrough–was going to be an enormous game-changer. Autonomous driving finally would come to fruition, however transformers are, uh, “reworking” the world proper now.

“Very hardly ever do you see one thing that drops that fully modifications the world multi functional shot,” Mohan tells Datanami. “That is extraordinarily, extraordinarily unusual. And that’s what makes it so cool.”

Mohan and Chen pivoted from supporting autonomous driving startups and launched Codeium in early 2023 with an AI coding assistant. The product, which it phrases an “clever AI code era software,” is powered by a customized LLM that clients run on their very own gear or within the cloud (clients can go for GPT-4 working within the cloud in the event that they like). Codeium plugs into greater than 40 built-in growth environments (IDEs), together with main ones like JetBrains, VS Code, Eclipse, and many others. and works with greater than 70 languages, together with large ones like Java, Python, and SQL.

Microsoft has the wherewithal so as to add a key to launch its Copilot product to Home windows 11 keyboards

When Codeium launched, the marketplace for AI-powered help was dominated by GitHub Copilot, the product of a collaboration between Microsoft and OpenAI that debuted in June 2021. That gave GitHub Copilot a giant head begin, which Microsoft is constructing upon by transitioning Copilot right into a company-wide growth effort. (Microsoft even added a Copilot key to the keyboard of Home windows PCs, simply to indicate the world that it’s enjoying with Monopoly cash.)

Regardless of the large head begin for GitHub Copilot, the market has proven its open to different “copilots”–particularly ones which can be extra open and work with a wider ecosystem of instruments than the one from Microsoft. And as Mohan factors out, a lot of the largest corporations don’t really use GitHub. As an alternative, they use different instruments like Bitbucket, Gitlab, Mercurial, Subversion, and CVS.

“There’s a number of totally different instruments that individuals use to retailer their supply code,” Mohan says. “We give individuals customized experiences, so we make it possible for the code that will get generated is definitely tied to the non-public code that an organization has. And we really made certain that the fashions are tuned and skilled on permissively licensed knowledge. So not knowledge that’s GPL-licensed.”

Codeium capabilities like a junior programmer that’s there to assist the human programmer working on the IDE, in line with Mohan. Its autocomplete perform will end the road of code began by the human, whereas its AI chat perform permits the human programmer to ask questions of all of the code within the repo.

“Codium helps you write a number of software program, however writing software program isn’t the purpose for a developer,” Mohan says. “The purpose for a developer is fixing a job and writing software program is one a part of fixing the duty.”

Varun Mohan, the co-founder and CEO of Codeium

Due to how Codeium robotically creates an index for every code base it’s uncovered to, the product is best in a position to reply questions the developer may need, and likewise supply higher ideas, Mohan says. That interprets into time-savings for the developer.

“One of many large issues Codeium has really achieved is shrink the time it takes to onboard a brand new code base from three to 6 months to a few to 6 weeks as a result of we all know what the code base is basically doing,” he says.

As a result of Codeium understands the context of the code its working with, it lends itself to code reuse. That helps to reduce code bloat, Mohan says.

There are restrictions to what Codeium can do. You’ll be able to’t simply inform Codeium to go create new software program for you, take a look at it, combine it with the code base, after which deploy it. The tendency for copilots to hallucinate signifies that people want to take care of strict oversight, Mohan says.

“It’s for the constructing of code, for producing concepts, and for extra rapidly reviewing software program,” he says. “However the core basic rules of the software program growth lifecycle are nonetheless the identical. That you must take a look at your code, debug your code, assessment your code, and deploy your code.”

Quickly after the corporate began early final 12 months, it had garnered about 1,000 customers. However Codeium has grown considerably since then, and immediately, greater than 600,000 builders use the product, in line with Mohan.

“We course of over 100 billion tokens of code day-after-day, which is over 10 billion strains of code day-after-day,” he says. “We’re one of many prime 5 largest generative AI apps on the planet by way of quantity of textual content processed day-after-day for the product.”

And it’s not simply empty digital keystrokes, both. In line with Mohan, about 45% of all software program written by Codeium is dedicated to the shoppers’ code base. That’s significanitly above the business common.

One of many early adopters of Codeium is Dell, the Texas-based pc firm. In line with Mohan, Dell builders are in a position to get extra work achieved as a result of they’re in a position to concentrate on the work within the IDE, eliminating the necessity to context-switch, and stay in “movement state” longer.

“Writing software program isn’t the one factor that developer does, however the motive why it nonetheless offers a number of worth is Codium is ready to allow the developer to navigate software program far more rapidly,” he says. “If there’s context switching overhead, if you happen to make them have a look at a Net web page, the place they will’t take a look at the software program, they will’t compile the software program, after which after that, they should deliver it again to the IDE,” it decreases productiveness.

It’s fairly clear that there’s a robust case that copilots and AI-coding assistants present actual advantages to builders now. There are nonetheless limitations, such because the tendency of LLMs to hallucinate, which suggests they want strict oversight, like junior programmers usually do. And a number of the harder coding issues, like migrating the billions of strains of previous COBOL code to extra fashionable languages like Java or .NET, aren’t going to be solved by copilots anytime quickly.

However in the long term, Mohan, who has a Grasp’s of Engineering from MIT, is bullish on the potential for AI to considerably affect the world of IT.

“Giant-scale autonomous brokers changing the way in which during which software program growth works within the subsequent 12 months, regardless of the present hysteria, in all probability isn’t going occur,” he says. “However is AI going to generate and do higher- and higher-level duties? Sure. I feel the subsequent 5 years are going to be very loopy. There’s going to be a number of innovation.”

Associated Gadgets:

Knowledge High quality Is A Mess, However GenAI Can Assist

Why A Unhealthy LLM Is Worse Than No LLM At All

Microsoft’s New Copilot Professional Presents AI in Workplace Apps and Precedence Entry