Code Automation: This Startup Tried AI, and It Failed

Kite, an early mover in the autocompletion development tools industry, has decided to go out of business, saying there’s still a long way to go and that the state of the art in ML (machine learning) material on code is not good enough.

The San Francisco start-up has been working on Python and JavaScript code autocompletion tools for developers since 2014. Tools that aim to reduce repetitive tasks in programming. Kite CEO and Founder Adam Smith announced Kite’s demise in a blog post, noting that even though his product had reached 500,000 monthly active developers, the company had failed to monetize it.

“We failed to realize our vision of AI-assisted programming because we were more than 10 years ahead of the market, i.e. the technology is not ready yet “, he wrote. “We failed to build a business because our product was not monetized, and it took too long to realize that. »

A competitor to IntelliSense, from Microsoft

Adam Smith also suggests that Kite might have been able to solve the code synthesis problem reliably, but it would have required $100 million and more engineers.

“The biggest problem is that state-of-the-art models don’t understand code structure, such as non-local context. We’ve made some progress towards better models for the code, but the problem requires a lot of engineering. Building a production tool that can reliably synthesize code could cost upwards of $100 million, and no one has tried it yet,” Smith points out.

Kite was, however, a promising alternative to IntelliSense, Microsoft’s code completion tool. In 2019, while still only supporting code completion in Python, the company raised $19 million in a Series A funding round led by Trinity Ventures, with personal participation from Nat Friedman, then new CEO of GitHub, as TechCrunch reported at the time.

Tabnine, the new competitor

But in 2020, with new competitors like Tabnine using OpenAI’s GPT-2 Wide Language Model (LLM) for language-neutral code completion, Kite chose to redesign their product using GPT-2 to take supports autocompletion for 11 additional languages. Subsequently, Kite also built integration with Microsoft’s VS Code editor.

Kite has opened most of its code on GitHub. It includes Kite’s data-driven Python-like inference engine, Python public package analyzer, desktop software, editor integrations, GitHub crawler and analytics bot, and more.

In June 2021, Microsoft released a preview of a VS Code extension leveraging OpenAI Codex to create GitHub Copilot. In June 2022, GitHub Copilot became available for $10 per developer per month, or $100 per user per year.

“Copilot is very promising, but still has a long way to go”

However, Adam Smith points out that cutting-edge machine learning is still not good enough to do what it needs to do.

“We built the most advanced AI to help developers at the time, but it didn’t achieve the 10x improvement needed to break through, because the state of the art for ML on code is not good enough. You can see this in GitHub Copilot, which is built by GitHub in collaboration with Open AI. At the end of 2022, Copilot is very promising, but still has a long way to go,” he wrote.

However, according to Adam Smith, the future is still bright for these tools: “We can’t wait for AI to revolutionize programming. Computers are so magical; it will be amazing to see a step-by-step increase in what they can do for us. »


Similar Posts

Leave a Reply

Your email address will not be published. Required fields are marked *