When Dollingo launched in 2012, a language learning app became a poster of gamefound. The app is shameless and brilliant so that users can overcome the lessons with lines, leader boards and timely challenges.
Many users, including me and my 1300 days, fell to the cartoon masks and strange social media posts in Dollingo. The company is never afraid to panic. The pair’s owl is cute, but Dollingo has adopted a successful strategy for not doing its customers. The app regularly sends me a notification from my boyfriend, requesting me not to “break” us (our friend’s line). Look, it doesn’t happen Not Task
But this spring, Dollingo had made a huge misuse of messaging on adoption of AI, and brought many users closer to the ends of the end.
In April, Louis Wan Ahn, CEO of Dolingo Announced to employees He is going through Dollingo AI. The company will look for AI’s skills in hiring future services, and the use of AI will be assessed in employees’ performance studies. It will also move to change the contract workers wherever possible. This was the statement that many users were caught in the curve, and when I read it, they honestly surprised me. I am seriously convinced that many companies would like to replace expensive human workers with machines. To recognize it There is another thing.
The memo was a shot: Dollingo’s lead AI views as a pyramid shift, which is like adopting a mobile phone in the 2010s. At this time, the general wisdom may have ordered that the language learning program should give priority to a widely adopted platform like PC. Instead, the company went “mobile first”. This condition certainly paid. Dolingo saw 103 million users in a month in 2024. Now, she wants to go to “AI First”.
It is too soon to tell what the long -term effects of the decision will be. But in a short period, The result has been loud and angry In social media. The long -standing user is deleting the app, destroying 1000+ days of lines. Has been announced Painted as a failure In several posts. Dollingo subdidate melted so well that the modes placed a maratrium on posts about AI.
Meanwhile, Dolingo stock prices have increased by more than $ 500 (by June 2, 2025), which shows that consumers can feel about AI, the older boys who can bring the money nearby think it’s here to stay here.
Wan Ahn later made the second statement, “AI First” by not pushing the shift, But is stirring it in a soft tongue.
He wrote, “I am not seeing AI as a place for my employees.” “I see it as a tool to accelerate my work, at the same or better quality standards. And as soon as we learn to use it, and use it responsibly, the better we will be in the long run.”
Dollingo is not the only company. In the tech industryWorkers are being evaluated on the use of their AI, encouraging AI tools in the service of future productivity, and asked to train their own changes.
It is not being prepared in this way. Rather, executives are talking about their AI measures like Louis Wan Ahn: as a tools tools to help people, instead of replacing them.
Dolingo’s AI shift has been running for years
Of course, Dollingo has been moving towards “AI First” for years. In 2023, it used Openi’s GPT4 to make AI features Who are only available in Dolingo MaxMost subscription tires on the app.
One of these “describe my answer”, which tells users why their response to an exercise is wrong. Generally, if something goes wrong you, Dolingo will give you the right answer, but it will not be explained Why? You were wrong When it comes to types or wrong spelling, the error may be clear. But if they are basically misunderstood, this is a gramical concept.
Earlier, Dollingo hosted a forum where users could directly see the explanation of other users and local speakers in the app. It was removed in 2022.
Now, Dolingo Subsidate is surprised to find answers to their questions with users. And many of them are accusing them of their confusion. This is the goat of the sacrifice Unconscious conversationTranslating errors, and simply simple Strange exercises.
Without verification from Doingo, it is impossible to say which of these problems is in fact due to the implementation of Dollingo’s AI. In some cases, consumers are Really to face software worms Instead AI-created lessons.
But elsewhere, AI has realized the Dollingo lessons for the worst. As a result, Riley McLead writes The Irish course he was pursuing has been ruined by AI Vices Those who do not pronounce the Irish words correctly – a serious situation for a language It is literally at risk.
I talked to the former doolingo user Kali R, who is learning Japanese. They saw how the words were pronounced through the robotic voiceover in Word banks, there was a similarity between how the same words were described in the exercises.
Kali said, “This is the only aspect of how the Japanese is written, it is usually not possible to tell how Kanji is considered when you see it in isolation.” “It is understood that an automated process of producing content will make such a mistake, but a human team who is actively developing this course, keeping in view the results of learning, will not do so.”
They also pointed to other users’ observations that Dolingo’s robotic sound is not correct To speak a Japanese pitch activeAn important aspect of the language, and a local English speaker cannot easily lift.
He said, “It was not able to learn the language as a means of wrong.
Two years later, the Kali R deleted the app and started its 700+ day.
AI should be good in it
The thing is, learning language is a field where AI’s large language models are Can Be actually useful. These LLMs are not reliable truth -speaking, but they can be a practical partner of the conversation.
Duolingo has long been criticized for how he teaches users to speak effectively – the app naturally focuses more on reading and listening, and the lessons of “speaking” are more than that they are about to actively remember memory words. The latter is important for real flow in another language.
Dolingo is trying to solve this flaw with two and more special AI features that allow users Talk to Dolingo’s cartoon masks. The most influential of these is the video call, where users can make a short “phone call” with Lily, Dollingo resident’s sad Goth Girl.
Earlier this year, I had some conversations with Lily during the Dollingo Max -free trial. In each of them, she would ask me a question, repeat me what she understood by my answer, and then asked a simple follow. We talked about things like animals or fruits I like, or how my holiday is going. It forced me to remember Italian words on the fly, without a word bank to help me.
Photo: Dolingo through Polygon
This is the area where the LLMS Excel is: to create a human language based on speech patterns.
Unfortunately, the LLMS fails in areas in which the doolingo is trying to disrupt. In his blood in the machine newsletter, journalist Brian Merchant Talked with the former doolingo employee Whose job was from writing lessons to how to write lessons, from training to non -existent.
The employee told the merchant, “We had been working with his AI tool for a while, and he was unable to write lessons without humans,” the employee told the merchant.
The lead of the Dollingo led, the flaws in the system are the costs they see as the modern edge. Dollingo’s lessons are not considered good.
“We can’t wait until the technology is 100 % perfect.” “We will rather move forward immediately and take a lot of successful films more occasionally than to move slowly and lose the moment.”
Users who are taped in these conversations are not confused. A recent research has shown it Confessing in the use of AI can cause people less confidence. This is the situation that seems to be running on Dolingo Subsiddate, where consumers are in a permanent battle To know what AI is and what is not.
Some are deleting the app as Kali R. But there is a sad fear that the silent majority cannot care about any of these problems or even know them. Dollingo Subardet has more than 508,000 members – which is less than 5 % of Dollingo, which is reported to be 116 million monthly users. And Subreddit itself is not fully anti -II. Many users accept it, or upright Don’t think that there is no use to fight the maize.
Dolingo’s AI policy apps in question
The use of my own doolingo has always been predicted on an assumption: it will not be Injury Learning your language. There has been a lot of ink spreading on the fact that doolingo may not likely make you fluent in another language. SureI have always argued, I know this. But do 5 minutes of Italian work every day when I am very slow or cheap or unable to find a tutor Nothing is better than nothing. I am still learning, even if I am not tied to the flow and fluently.
But the arrival of AI content puts this justification at risk. However, language students do not know what they do not know.
Kali R said, “I really don’t care that unless it is monitored and if no one is producing real Japanese, someone is ready to pull.” Instead of pulling a plug, people in Dollingo are actively enabled the Japanese to learn the Japanese language, hoping that someday AI will teach Japanese instead of Japanese.
There is no clear road map to go there. LLM can be taught to speak language. It is not clear that they can be taught Teach.
Doolingo
Dolingo faces a scale problem: he wants to present many language courses, and requires time and money to make these courses. It turned to the AI to fill the frustrating gap where a person can be fine, but never can be so fast.
The thing that makes the dolingo crawling even more unpleasant is that it is most likely to have an effect Tongues with small user base – Like Irish or Navajo, both endangered languages. The majority of the app users are studying English, French, or Spanish. These are the courses that see the real investment of resources.
Doolingo Gets a great press To create lessons that consumers intend to get acquainted with Nawaju. But if the use of AI is used to “scale” the Navajo program, will seem to be used with very few human guards to ensure that exercises are correct?
“Without AI, we will take decades for decades for more and more learners,” Wan Ahn wrote in his first statement. “We owe our learners to get this content to ASAP.”
My question is… why? Why do we need more content for users immediately, when this content can be incorrect or low quality? It is here that the Dollingo mission to make the language accessible plays its role as a public trading company. Lessons need a scale so that the user is on the app, so that the app can make money.
Actually learning a language – or even one mere running water – has no part to play.
