- Các startup edtech như Prep (Việt Nam), CoLearn (Indonesia) và ELSA không lo ngại về sự xuất hiện của ChatGPT, mà coi đó là cơ hội để thị trường chấp nhận các sản phẩm học tập cá nhân hóa của họ.
- Prep sử dụng AI để giúp học sinh luyện tập kỹ năng nói IELTS, cung cấp phòng nói ảo với trợ lý AI và phản hồi tức thì về kỹ năng tiếng Anh. Họ thuê giáo viên tiếng Anh chất lượng cao (điểm IELTS 8-8,5) để thu thập dữ liệu tốt, giúp cung cấp phản hồi chi tiết và có giá trị hơn ChatGPT.
- CoLearn đã thu hẹp công cụ hỗ trợ bài tập AI miễn phí và chuyển sang cung cấp các lớp học trực tuyến về toán và khoa học cho học sinh lớp 4-12, tập trung vào tương tác thường xuyên giữa giáo viên và học sinh.
- ELSA, với 7 triệu người dùng hàng ngày, cung cấp trải nghiệm cá nhân hóa dựa trên trình độ tiếng Anh của từng người dùng. AI tạo sinh giúp ELSA tạo nội dung nhanh hơn và phù hợp hơn với thực tế. Công ty đang phát triển mô hình ngôn ngữ lớn riêng, tích hợp công nghệ sở hữu độc quyền và công nghệ hiện có để tạo sự khác biệt.
- Geniebook giải quyết vấn đề "ảo giác" của các mô hình ngôn ngữ lớn bằng cách phát hành các tính năng mới theo từng giai đoạn và có giáo viên xác minh đề xuất của AI, đảm bảo tính chính xác của nội dung cung cấp cho học sinh.
- Theo Maria Spies, CEO của Holon IQ, việc sử dụng AI không phải là yếu tố quyết định để các công ty edtech chiến thắng thị trường, mà là khả năng mang lại kết quả giáo dục tốt hơn so với đối thủ cạnh tranh.
📌 Mặc dù ChatGPT đe dọa các startup edtech, nhưng các nhà sáng lập vẫn tự tin cạnh tranh bằng cách tập trung vào học tập cá nhân hóa, xây dựng mô hình AI nội bộ, tích hợp công nghệ hiện có và thuê giáo viên chất lượng cao. Prep cung cấp phản hồi chi tiết hơn ChatGPT nhờ dữ liệu từ giáo viên giỏi, trong khi ELSA phát triển mô hình ngôn ngữ riêng để tạo sự khác biệt. Cuối cùng, khả năng mang lại kết quả giáo dục tốt hơn mới là yếu tố quyết định thành công.
https://www.techinasia.com/chatgpt-threatens-edtech-startups-but-founders-unfazed
#Tech in Asia
ChatGPT threatens edtech startups, but founders are unfazed
For Prep CEO and co-founder Tu Pham, ChatGPT is a “friend” of his business rather than an “enemy.”
That might seem like a bold statement, given that Prep – a Vietnamese edtech startup that has raised US$9 million in disclosed funding – is using AI to help students practice their IELTS speaking skills.
Prep offers virtual speaking rooms, where students interact with an AI assistant who provides immediate feedback on their English skills across a diverse range of social settings.
However, many students could turn to ChatGPT’s latest version, which features voice-to-text capabilities. This advancement has sent shockwaves through the industry, as it enables users to engage in real-time conversations with the AI chatbot in 12 languages, including English and Mandarin.
Pham is not fazed. When the startup introduced virtual speaking rooms in early 2022, it initially appealed to a small group of students interested in high-tech products.
“More students got interested [in Prep’s AI-enabled speaking rooms],” Pham says. “So for me, it’s not really a competition for the company. It only opened more opportunities for the market to adopt us.”
Contrary to popular belief, generative AI will not have a detrimental impact on edtech businesses, according to edtech founders and industry experts Tech in Asia interviewed for this story.
But they agree that competition will become fierce, and companies that will succeed are those willing to make big bets on personalized learning products and high-quality human teachers.
Time to get personal
Pham, a former school teacher, co-founded Prep with fellow educator Nam Tran in 2020.
They initially built the company to aid teachers in developing lesson materials but then turned it into a language learning assistant for Vietnamese students prepping for IELTS, or the International English Language Testing System. It’s a standardized English language proficiency test for non-native speakers, which has become a requirement at many schools in Vietnam in recent years.
Data from Tech in Asia shows that most edtech startups in Southeast Asia cater to K-12 students or offer language learning services. Funding to the sector peaked in 2021 at US$948 million but dropped more than 8x to US$147 million.
Amid the tech winter, even major edtech companies such as LingoAce and Ruangguru had to lay off hundreds of employees as their expansion plans were affected by the return to in-person schooling and hit hard by broader macroeconomic realities.
Indonesia-based tutoring platform CoLearn, which was founded in 2020 and has secured nearly US$30 million in funding, was one of the early edtech companies tapping into AI. It offered a free AI-enabled assignment assistance tool called Ask to grade school students in Indonesia.
At its peak, the AI tool had 4 million users, says CoLearn co-founder Marc Irawan in a recent interview with Tech in Asia. CoLearn had expected that the free service would incentivize students to sign-up for its paid online tutoring services. But the plan didn’t materialize.
In recent years, the startup has scaled down the marketing of its AI-enabled tool and shifted resources to offering online extra classes in math, and science for students in Grades 4 to 12 in Indonesia. The system mimics a school setting: Students are enrolled on a semester basis and have the same set of classmates all year round.
“Our teachers on the platform never speak for more than five minutes alone. The interactions happen regularly during class to ensure students are still paying attention,” Irawan says.
Ever since the company scaled down its assignment assistance service, CoLearn has not tapped or created any AI-enabled service. “It’s not cheap to run an AI model,” the co-founder adds. “For now, we are seeing more long-term engagement in our live classes and that’s what we’re focusing on.”
For edtech startups that have invested significantly in AI, providing a personalized experience for users remains a top priority.
English language learning platform ELSA, which claims to have 7 million daily users on its mobile app, says it has configured its offerings based on each user’s level of English proficiency.
On the app, users can create their own study sets based on an English word they wish to master. For example, someone who wants to use the word “exaggerate” more often can build a module that allows them to listen to native English speakers using the word in different contexts.
ELSA also features role plays, which allow users to practice speaking in English based on sample social interactions. Unlike Prep, however, ELSA enables users to customize their own role-play scenarios.
Founded in 2015, ELSA initially catered to anyone wanting to learn English, whether an adult exploring it for fun or a young child using it for school. The startup has since expanded, offering specialized versions of the app for schools and companies.
Michael Ngo, ELSA’s general manager for Southeast Asia and Taiwan, noted that genAI has actually enabled ELSA to produce personalized content “much faster” and one that’s more applicable in real-world settings.
Founded in 2015, ELSA initially catered to anyone wanting to learn English, whether an adult exploring it for fun or a young child using it for school. The startup has since expanded, offering specialized versions of the app for schools and companies.
Michael Ngo, ELSA’s general manager for Southeast Asia and Taiwan, noted that genAI has actually enabled ELSA to produce personalized content “much faster” and one that’s more applicable in real-world settings.
“Hallucinations” – the tendency of large language models (LLMs) to generate incorrect answers and present them as facts – are a common concern for edtech startups using genAI.
It’s a feedback that regional edtech startup Geniebook commonly receives from parents, its CEO Neo Zhizhong says in an interview with Tech in Asia. After all, the platform is mainly known for its AI-generated worksheets, which are personalized based on students’ level of expertise on a subject.
To address this issue, Neo says the company has started releasing new features in phases to ensure their quality and accuracy. For some features, a human teacher must verify what the AI model recommends.
For example, worksheets submitted by students on the platform are initially assessed by the AI model, but a human teacher decides on the final grading to ensure accuracy.
Similarly, the platform’s Genie Ask, which allows students to ask questions about their worksheet problems, is AI-enabled but still overseen by human checkers.
Questions that have never been previously asked on the platform will be answered by a human teacher, according to Neo. Eventually, when a similar question is posed again, the AI takes over and mimics the previous response.
“One critical role that edtech companies will have to play moving forward is as gatekeepers of accurate information,” the CEO says. “That includes ensuring that the content we provide is verified before distributing it to students.”
ELSA also employs an added layer of verification on its platform to ensure that users are not taught “harmful language,” general manager Ngo says.
These edtech startups are betting that, with human teachers involved, their models can offer more value to students and educators than relying solely on a chatbot. However, it’s not an easy or cheap endeavor.
Prep CEO Pham says the company had to hire “high-quality” English teachers – those who typically score between 8 and 8.5 on the IELTS band (where 9 is the highest) – to gather “good data” for its platform. This approach is “crazy expensive,” he notes.
These teachers, who write detailed, and well-informed feedback to students with their essays, are the basis of Prep’s AI-enabled correction function for users’ essays. So for Pham, finding such teachers is critical.
While ChatGPT can provide similar feedback, Pham says Prep’s feedback is more extensive. For instance, when ChatGPT is given an essay with 20 grammatical errors, it might only detect 10. In contrast, Prep’s model can identify 15 or more.
“ChatGPT won’t provide feedback that is as detailed, and informative as what our platform can do, which is backed by teachers,” he adds.
Pham says the startup can afford to employ such teachers because the feature that allows students to check their grammar in written essays is offered as a premium service.
For example, IELTS courses on the platform cost at least US$135. Meanwhile, a TOEIC course that does not have this feature only costs US$40.
AI-first or edtech-first?
ELSA’s services are currently built on existing LLMs. However, Ngo says the company is developing its own LLM with a Portugal-based AI team that’s working to leverage ELSA’s existing data.
Ngo says edtech startups today are grappling with a dilemma similar to what many companies faced when cloud computing first emerged as an outsourcing option. Although using cloud services may be more cost-effective initially, scaling the company can become more expensive in the long run.
If you only offer what general generative AI models already do, then how would you differentiate?
For Ngo, it’s all about striking a balance between developing in-house AI models and using existing solutions in the market.
To stay competitive in the long run, he believes that edtech companies need to “build your own LLM, layer it with your in-house proprietary technology, as well as integrate existing technology on top of that.”
Does this mean edtech companies should prioritize becoming AI-first? “Because if you only offer what general generative AI models already do, then how would you differentiate?” Ngo adds.
At least ELSA has the resources to do so. Despite the tech winter, it raised US$22.53 million last year, bringing its total funding to date to US$60 million.
However, Maria Spies, CEO of education market intelligence platform Holon IQ, believes that the use of AI alone can’t be the defining factor for edtech companies to win the market.
“If your edtech startup can show that you deliver better educational outcomes while offering the same service as your competition, then you’re a winner,” she says. “Everyone – customers and investors – will look for outcomes.”