How the Duolingo of STEM uses AI to teach tech skills to users

The interview and assessment platform CodeSignal has a new AI-powered, online learning platform that aims to build professionals’ technical skills, as companies everywhere struggle to keep up with rapidly changing technology.

On Thursday, the company debuted CodeSignal Learn, which features about 100 technical skills courses ranging from intro to programming to full stack engineering. CodeSignal Learn will offer a free tier and a paid tier, which will cost $24.99 per month. The platform was announced at CodeSignal’s annual Beyond event, which included conversations with tech leaders like Apple cofounder Steve Wozniak and Khan Academy founder Sal Khan.

The platform is built around Cosmo, CodeSignal’s AI, which combines several large language models (including Open AI’s GPT-4) and the company’s proprietary skills data. Cosmo, a space-suit-wearing Corgi, is designed to offer encouragement and hints like a one-on-one tutor, and answer questions within the context of a user’s familiarity with each subject.

CodeSignal Learn comes as AI continues to shape the global economy. For tech workers, AI’s potential for professional skill learning couldn’t come at a better time. Nearly 32,000 have been laid off since the start of the year; many will seek out skill-building opportunities to maintain their value to companies.

“Our mission has always been to discover and develop the skills that will shape the future,” says CodeSignal cofounder and CEO Tigran Sloyan.

Sloyan started CodeSignal in 2016. Since then, the company has raised over $87 million and now counts Netflix and Meta among its clientele.

Sloyan sees the platform as a natural next step, allowing users to actually learn the tech skills for which CodeSignal screens and evaluates candidates. “We have been hearing from our customers that hiring can’t be the only solution,” he says, because there are not enough people to hire and skills are changing too fast. Now, these customers want to develop the skill set of their existing teams—and Sloyan hopes his platform can close that skills gap.

The platform interface itself feels a lot like Duolingo. An animal mascot shepherds users through a path of activities, which are organized by unit and illuminate in color once completed. And like Duolingo, CodeSignal Learn employs a streak system, motivating users’ daily practice via gamification.

Additionally, CodeSignal Learn aims to bypass traditional learning structures with a practice-first approach. “People learn through practice. We all instinctively know this, but the entire education system doesn’t emphasize this at all,” Sloyan says, pointing out that so much of digital learning now includes videos of other people practicing skills, rather than activities for learners to complete themselves. CodeSignal Learn is about 90% activities and just 10% instructions, Sloyan says.

With all that practice, Sloyan says it’s natural some learners will get stuck. That’s where Cosmo—the platform’s built-in, AI-generated mentorship—hopes to curb any discouragement. Sloyan says he designed Cosmo based on research from the late educational psychologist Benjamin Bloom, who showed one-on-one tutoring is the most effective way to learn. While personalized tutors are largely inaccessible in traditional education due to high costs, Sloyan sees AI as the key to making one-on-one tutors accessible at scale.

But some remain doubtful of AI-powered tutors. Satya Nitta, one of the creators of IBM’s Watson, quit his five-year effort to build a personal, AI-powered digital tutor a few years back, before the introduction of ChatGPT. “We’ll have flying cars before we will have AI tutors,” he told EdSurge last month. “It is a deeply human process that AI is hopelessly incapable of meeting in a meaningful way.”

Other adult education platforms like Coursera use AI-generated assistants, too. Khan Academy, a learning platform popular for K-12 education, is currently piloting its ChatGPT-based personal tutor in schools, with one warning: the chatbot may “hallucinate,” or fabricate information when it doesn’t know an answer. Hallucinations are common among chatbots, and pose a real issue for students who depend on them to answer their questions. But Michael Feldstein, an edtech consultant, points out that human tutors can be wrong, too. “They’re tools. They’re strange tools. They misbehave in strange ways—as do people,” he told EdSurge.

Sloyan believes CodeSignal Learn’s combination of large language models—including Open AI’s GPT-4, GPT-4-Turbo, Claude 2, and LLAMA 2—and proprietary skills data will help Cosmo succeed where other personalized tutors have failed.

Beyond building technical skills, Sloyan hopes the platform will inspire people to add new skills long after their schooling is done. “Most people get told from a young age that they’re bad at math. They start to believe it,” he says. “In reality, talent is not a gift. It’s built through practice.”

The interview and assessment platform CodeSignal has a new AI-powered, online learning platform that aims to build professionals’ technical skills, as companies everywhere struggle to keep up with rapidly changing technology.

On Thursday, the company debuted CodeSignal Learn, which features about 100 technical skills courses ranging from intro to programming to full stack engineering. CodeSignal Learn will offer a free tier and a paid tier, which will cost $24.99 per month. The platform was announced at CodeSignal’s annual Beyond event, which included conversations with tech leaders like Apple cofounder Steve Wozniak and Khan Academy founder Sal Khan.

The platform is built around Cosmo, CodeSignal’s AI, which combines several large language models (including Open AI’s GPT-4) and the company’s proprietary skills data. Cosmo, a space-suit-wearing Corgi, is designed to offer encouragement and hints like a one-on-one tutor, and answer questions within the context of a user’s familiarity with each subject.

CodeSignal Learn comes as AI continues to shape the global economy. For tech workers, AI’s potential for professional skill learning couldn’t come at a better time. Nearly 32,000 have been laid off since the start of the year; many will seek out skill-building opportunities to maintain their value to companies.

“Our mission has always been to discover and develop the skills that will shape the future,” says CodeSignal cofounder and CEO Tigran Sloyan.

Sloyan started CodeSignal in 2016. Since then, the company has raised over $87 million and now counts Netflix and Meta among its clientele.

Sloyan sees the platform as a natural next step, allowing users to actually learn the tech skills for which CodeSignal screens and evaluates candidates. “We have been hearing from our customers that hiring can’t be the only solution,” he says, because there are not enough people to hire and skills are changing too fast. Now, these customers want to develop the skill set of their existing teams—and Sloyan hopes his platform can close that skills gap.

The platform interface itself feels a lot like Duolingo. An animal mascot shepherds users through a path of activities, which are organized by unit and illuminate in color once completed. And like Duolingo, CodeSignal Learn employs a streak system, motivating users’ daily practice via gamification.

Additionally, CodeSignal Learn aims to bypass traditional learning structures with a practice-first approach. “People learn through practice. We all instinctively know this, but the entire education system doesn’t emphasize this at all,” Sloyan says, pointing out that so much of digital learning now includes videos of other people practicing skills, rather than activities for learners to complete themselves. CodeSignal Learn is about 90% activities and just 10% instructions, Sloyan says.

With all that practice, Sloyan says it’s natural some learners will get stuck. That’s where Cosmo—the platform’s built-in, AI-generated mentorship—hopes to curb any discouragement. Sloyan says he designed Cosmo based on research from the late educational psychologist Benjamin Bloom, who showed one-on-one tutoring is the most effective way to learn. While personalized tutors are largely inaccessible in traditional education due to high costs, Sloyan sees AI as the key to making one-on-one tutors accessible at scale.

But some remain doubtful of AI-powered tutors. Satya Nitta, one of the creators of IBM’s Watson, quit his five-year effort to build a personal, AI-powered digital tutor a few years back, before the introduction of ChatGPT. “We’ll have flying cars before we will have AI tutors,” he told EdSurge last month. “It is a deeply human process that AI is hopelessly incapable of meeting in a meaningful way.”

Other adult education platforms like Coursera use AI-generated assistants, too. Khan Academy, a learning platform popular for K-12 education, is currently piloting its ChatGPT-based personal tutor in schools, with one warning: the chatbot may “hallucinate,” or fabricate information when it doesn’t know an answer. Hallucinations are common among chatbots, and pose a real issue for students who depend on them to answer their questions. But Michael Feldstein, an edtech consultant, points out that human tutors can be wrong, too. “They’re tools. They’re strange tools. They misbehave in strange ways—as do people,” he told EdSurge.

Sloyan believes CodeSignal Learn’s combination of large language models—including Open AI’s GPT-4, GPT-4-Turbo, Claude 2, and LLAMA 2—and proprietary skills data will help Cosmo succeed where other personalized tutors have failed.

Beyond building technical skills, Sloyan hopes the platform will inspire people to add new skills long after their schooling is done. “Most people get told from a young age that they’re bad at math. They start to believe it,” he says. “In reality, talent is not a gift. It’s built through practice.”