Duolingo is a leading education technology company specializing in application-based learning. It’s also a leader in educational artificial intelligence (AI), and recently partnered with OpenAI to become one of the first companies to offer customers GPT-4, its latest language model.
OpenAI’s GPT technology has recently become familiar to millions, thanks to its ChatGPT application, which is said to have had the fastest growing user base ever. GPT-4 technology has only recently been available in a handful of applications, including the new paid version of ChatGPT, the latest version of Microsoft’s Bing search engine, and, starting this month, the latest version of Duolingo’s language platform, Duolingo Max.
The company isn’t making its first foray into artificial intelligence. Since its inception in 2011, its mission has been to create a “virtual language tutor” that makes language learning available to anyone, anywhere, anytime.
So let’s take a look at how the company has used AI over the years, as well as some of its latest AI-powered services that support GPT-4.
Personalized learning, automated feedback and English tests
Duolingo’s mission is to create the best virtual language tutor, and to achieve that, it must be able to take care of each of its 500+ million students as an individual. This is one of the first applications of artificial intelligence within the organization. Data from each of these exercises is captured and analyzed using machine learning algorithms to understand how people learn and how the app can be used to improve how they absorb new information.
“These exercises allow us to better understand what our learners know,” explains Clinton Bicknell, Head of AI at Duolingo.
“We can then use this data using machine learning models to determine the best way to teach and keep learners engaged.”
Another key use case for Duolingo revolves around creating interactive learning experiences. One example is the automatic comments on the pronunciation and grammar exercises that students are asked to complete. App users receive real-time feedback on how they are using the language, as well as tips on how to improve it.
Artificial intelligence is also used extensively in the company’s English language testing system, which is used by more than 4,000 institutions, including universities, colleges and government organizations, to assess the skills of non-native speakers. During exams conducted remotely via webcam, artificial intelligence is used to detect different cheating methods that candidates may attempt. It also automatically creates test items, rates the difficulty level of its tests, automatically scores test takers, and reviews overall performance before releasing test results.
How is GPT-4 used?
Through its partnership with OpenAI, Duolingo became one of the first third-party organizations to officially license and release a GPT-4-based product.
GPT-4 is much more powerful and complex than GPT-3. It would have been trained on a trillion parameters (a piece of information), as opposed to 175 billion for GPT-3. This means that he is able to provide more detailed, accurate and fluent answers to questions asked of him.
“It’s very exciting,” says Mr. Bicknell. “We were blown away by the power of the model, and we immediately asked ourselves the question: With a strong language large model (LLM), what kind of things did we want to do in the past why the technology wasn’t there?”
Bicknell’s team has narrowed down two initial use cases, which are now available to users of the app.
The first is a role-playing game that allows users to interact with an AI-powered character and perform various learning-oriented tasks. An example of this is chatting with an AI barista in a Parisian café. Users can order drinks, select how they are made, and chat with the barista on a number of topics. In order to make the experience more engaging, the characters are designed as complete personalities, with their own personality and story, which learners can discover through conversation and interaction.
The second GPT-4 compatible function already provided is called ” Explain my answer (Detailed explanation). It allows users to get an explanation as to why the answer they gave to a question might have been marked as correct or incorrect and to receive a detailed and personalized explanation, as if they were sitting down and talking to a human guardian.
“It’s a really useful feature,” says Mr Bicknell. “We’re really bringing Duolingo closer to our ideal of being a one-to-one tutor in a way we haven’t been able to do before – we’re really excited about the idea that these features will help people learn their language.” »
Working with OpenAI
Developing these features requires a close working relationship with AI and data engineers and OpenAI software developers. In fact, a kind of symbiotic relationship quickly emerged from the business practices, as both companies provided feedback on use cases and processes that were used to mutually improve operations and user experience.
“It was a long process…the basic prototyping didn’t take long…the ‘Explanation of My Answer’ was completed in a day,” says Mr Bicknell. »
“The hardest part, and the most important part of the job is that the underlying prototype is often broken in different ways. So a lot of the work has been figuring out how to measure when things go wrong, and how to reduce the possibility of errors.”
This involved creating and categorizing massive amounts of data, which were then used to improve questions submitted to the Lifelong Learning Program. This data was then used to improve the initial version of the GPT-4 model that the Duolingo team was working with.
Uses for Duolingo’s existing AI algorithms have also been found – mainly to ensure conversations with GPT-4-powered bots, like AI Barista, stay on track.
As we saw with Microsoft’s initial Bing launch, which gave us our first taste of the power – and in some cases unpredictability – of GPT-4, users quickly found ways to generate unexpected and, in some cases, unwanted results.
To prevent his characters from being ruined in this way, AI routines are used to ensure that conversation remains, in general, centered around relevant topics. If not, it is possible to gently return them to the task in question.
“We have a lot of AI models to determine if things are going as planned,” he explains.
The goal is to provide an immersive conversational experience while keeping users safe and engaged in learning activities.
The future of AI at Duolingo
The success of Duolingo’s AI projects to date, including its most recent achievement with GPT-4, means that AI will continue to be a top priority when it comes to growing the business, creating new experiences for users and fulfilling the mission of becoming a personal language tutor around. the world.
Upcoming new applications include AI implementation and course content creation. “People always want more courses and want them to cover more advanced topics,” Bicknell says.
“We want to make sure they reach the highest standards of learning, which is a very slow and very expensive process.”
Creating new course content is one task in particular as the current generation of language models has proven to be very effective.
“They’re creative and can help generate many more trainings. So we’re thinking about incorporating them into our workflow,” he explains.
The integration of GPT-4 into the company’s English language testing system is also progressing. Specifically, research is being done on ways to determine a test taker’s language skills from a small sample of their writing.
Of course, deepening the customization possibilities remains a top priority. The end goal is to become increasingly good at understanding what can be learned from user responses and how we can leverage that to keep that particular user engaged and convince them that they are benefiting from a valuable learning experience.
Overall, this strategy appears to provide a very valuable insight into how AI will become an important aspect of education systems around the world, both online and in the classroom. In all contexts, understanding how personalization, engagement, and experience influence our ability to teach and learn can be truly transformative, making learning more effective, rewarding, and more accessible to future generations.
If your English allows, click here to watch the webinar with Clinton Bicknell, Head of AI at Duolingo, in which he, along with Bernard Marr, author of this article, discusses the potential implications of AI for education in the future.
Translated article from the American magazine Forbes – Author: Bernard Marr
<< اقرأ أيضًا: ChatGPT ونماذج اللغات الكبيرة (LLM): تم الإعلان عن ثورة في Enterprise Search >>>