OpenAI’s GPT technology has recently become familiar to millions thanks to the ChatGPT app. fastest growing Best user base ever. GPT-4 is the recently released paid version of ChatGPT, the latest version of Microsoft’s Bing search engine, from this monththe latest version of Duolingo’s own language platform – Duolingo Max.
However, this is far from the beginning of the company’s AI efforts. Since its founding in 2011, its mission has been to create a “virtual language tutor” that enables language learning for anyone, anywhere, anytime.
So let’s take a look at some of the ways the company has leveraged AI over the years and some of the latest AI-powered services enabled by GPT-4.
Personalized Learning, Automated Feedback, English Tests
Duolingo’s mission is to create the ultimate virtual language tutor, and to achieve this, it must be able to serve each of its over 500 million students as individuals. This was one of the first applications of AI within the organization. Every day, the company’s students collectively complete approximately 1.5 billion exercises on the platform, and use them to understand how people learn and how apps are used. All data in has been acquired and analyzed using machine learning (ML) algorithms. Improve how you absorb new information.
Klinton Bicknell, Duolingo’s head of AI, said, “These exercises allow us to explore more closely what our learners know.
“We can then use that data through machine learning models to find ways to better teach and engage people.”
Another important use case for Duolingo revolves around creating interactive learning experiences. An example of this would be automated feedback for speech and grammar exercises that students are asked to complete. This gives app platform users real-time feedback on how they use the language and tips on how to improve it.
AI is also widely used throughout the company. English test systemis used by over 4,000 institutions, including universities, colleges and government agencies, to assess the performance of non-native speakers. Here, AI is used during remotely administered tests administered via webcam to detect various methods of cheating that candidates may attempt. It also automatically creates test items, assesses test difficulty, automatically grades candidates, and reviews overall performance before test scores are released.
How is GPT-4 used?
Partnering with OpenAI makes Duolingo one of the first third-party organizations to officially license and release GPT-4 powered products.
GPT-4 is said to be significantly more powerful and sophisticated than GPT-3, which many reading this are familiar with thanks to ChatGPT.that is report It is believed to have been trained with 1 trillion parameters (information), compared to 175 billion for GPT-3. This means that we can provide more detailed, accurate and fluent answers to the questions asked.
Bicknell told me “It’s very exciting…we were blown away by the power of the model and immediately sat down and said what we wanted to do with such a powerful Large Language Model (LLM). In the old days…did we just not have technology?”
Bicknell’s team settled on two initial use cases that are now available to app users.
The first is a role-playing feature that allows users to interact with AI-powered personas and perform a variety of learning-focused tasks. One example is talking to an AI barista in a cafe in Paris. Users can order drinks, specify how they’re made, and have conversations with baristas about various subjects. To make the experience more engaging, the characters are designed as fully fleshed out personas with their own personalities and backstories that learners can get to know through conversation and interaction.
A second GPT-4 capable feature already available is called Explain My Answer. This allows the user to get an explanation as to why the answer given to the question was rated correct or incorrect, as if they were sitting and talking with a human tutor. You will receive a detailed and personalized explanation.
Bicknell said: “This is a very useful feature…We are really getting closer to the ideal of making Duolingo a personalized tutor for everyone in a way that has never been possible before. I’m really looking forward to “people learning languages.” “
Developing these features involves working closely with OpenAI’s AI and data engineers and software developers. In fact, something of a symbiotic relationship quickly emerged from their working practices, with both companies providing feedback on use cases and operations that were used to mutually improve the operation and user experience of their products.
“It was a long process … very basic prototyping … took no time at all …” says Bicknell.[Explain My Answer] Assembled in one day.
“The harder part of the job, and a much bigger part, is that the basic prototype often fails in many ways. It took a lot of work to think about how sex could be reduced.”
This involved generating and labeling a huge amount of data, which was used to tailor the prompts (questions) sent to the LLM. This data was used to refine a pre-release version of his GPT-4 model that the Duolino team was working on.
We also found uses for Duolingo’s existing first-party AI algorithms. Mainly to keep conversations going smoothly with his GPT-4 powered bots such as AI baristas.
As we saw when Microsoft’s Bing first launched, we get our first glimpse of the power, and sometimes unpredictability, of GPT-4, and users quickly find ways to create the unexpected. I was. In some cases, undesirable result.
To prevent this unique character destruction, we use AI routines to, broadly speaking, keep the dialogue focused on relevant topics. If not, you can gently switch back to the relevant task.
“Are many AI models working and on track?” Bicknell says.
The aim is to provide an immersive conversational experience while keeping users safe and focused on their learning activities.
The Future of AI in Duolingo
The success of Duolingo’s AI projects to date, including its latest achievements at GPT-4, is a testament to how AI will continue to grow in growing its business, creating new user experiences, and achieving its mission of becoming the world’s personal language tutor. means to take precedence.
Future new implementations will include leveraging AI and generating course content. “People always want more courses and more advanced topics covered,” Bicknell says.
“Historically, this has been a constraint…we want our students to achieve the highest standards of learning, but it is a very time-consuming and expensive process.”
Generating new course content has traditionally been a bottleneck, and this is one task where especially the latest generation of language models have proven to be very efficient.
“They are creative and help generate a lot of exercises, so we’re looking at incorporating that into our workflow,” Bicknell says.
Further progress has been made by integrating GPT-4 into the company’s English testing system. Specifically, research is underway on how to determine a test taker’s language proficiency from a small sample of the test taker’s writing.
And of course, digging deeper into personalization possibilities remains a top priority. The ultimate goal is what can be learned from a user’s response and how it can be leveraged to keep a particular user engaged and satisfied that they are receiving a valuable learning experience. becoming more and more understandable.
Overall, we believe this strategy will provide invaluable insight into how AI will become a critical aspect of education systems around the world, both in online and classroom-based learning. Understanding how personalization, engagement, and experience affect how we teach and learn in all contexts has the potential to be truly transformative, making learning more effective. valuable and accessible for future generations.
you can click here Watch a webinar with Klinton Bicknell, Head of AI at Duolingo. There, we delve deeper into the potential impact of AI on the future of education.