duolingo launches new gpt-4-powered ai tools

Duolingo, the mobile application that teaches users foreign languages, is launching a new subscription tier that incorporates the new AI model from OpenAI – GPT-4 – to improve students’ experience during their virtual learning process.

The new paid subscription tier will be called Duolingo Max and will be the most expensive and feature-rich alternative for premium users. Duolingo has used GPT-4 to build two tools for its students.

The first one is called ‘Explain My Answer’ and its purpose is to create a thorough explanation of why an answer given by the student is right or wrong and what the correct answer is and why.

Duolingo says that it is often frustrating for people to be wrong many times about something they have studied hard while not knowing the exact reason.

Explain My Answer will act as a tutor in the sense that it can provide details and insights that can help students clarify their doubts.

Roleplay Lets Users Engage in Real-World Conversations with the AI

In addition, Duolingo is launching a new tool for its users to practice the language they are learning with the AI model – this new feature will be called ‘Roleplay’ and it will function similarly to a chatbot.

The GPT-4 model has been trained by the Duolingo team with the help of OpenAI to simulate real-life conversations with users. Those who engage in the chats will gain experience points and there will be various world characters available for them to talk to.

Also read: Snapchat Reveals New ChatGPT-Powered Chatbot Called My AI

For now, Duolingo Max will only be available to Spanish and French English Speakers that use the app on iOS-powered devices. The company plans to make the tool available to more courses and operating systems in the future.

Duolingo Max costs $29.99 per month or $167.99 per year – this is more than double its previous top subscription tier – Super Duolingo – which costs $12.99 per month.

Human tutors have been feeding the AI model to create these tools. The process has been broken down by Duolingo into four stages starting with curriculum design and finishing with lesson personalization.

In the case of the Roleplay feature, all of the scenarios were created by humans and are designed to be fun and effective lessons.

In addition, the company says that the answers generated by the AI tool are being “constantly” reviewed by tutors to make sure that the tone is appropriate and the conversation is factually accurate.

However, the learning app acknowledged that the technology is not perfect – they have promised to keep working with OpenAI on further refining the AI model until mistakes are flushed out of the system.

During this testing phase, users will be able to report any inaccurate responses or oddities that occur during the interactions they have with the two new features. These will be used by Duolingo to keep training the AI model.

GPT-4 is the Most Powerful (And Expensive) Model to Date

Duolingo’s announcement is taking place on the same day that OpenAI presented GPT-4 to the world.

The new AI model is the most powerful created by the Microsoft-backed AI company to date and its capabilities exceed those of its highly-popular predecessor ChatGPT.

According to OpenAI, GPT-4 can process both text and image prompts and it can come up with much more insightful answers and analyses.

Users can input as many as 25,000 words of context to the model, which increases its use cases significantly.

“GPT-4 is 82% less likely to respond to requests for disallowed content and 40% more likely to produce factual responses than GPT-3.5 on our internal evaluations,” the company said about the new model.

Additional human feedback, including the one submitted by users of previous models, was used to train GPT-4 and over 50 experts joined forces to provide feedback on the new version before it was released.

There is a waitlist for the GPT-4 application programming interface (API), with the price for using the API significantly higher than ChatGPT.

GPT-4 will cost $0.03 for every 1,000 prompt tokens and $0.06 for completion tokens for the shortest context format – for the longest format of 32,000 words the price doubles.