ChatGPT has posed a major challenge to edtech firms, compelling stakeholders to either leverage its capabilities or risk being left behind. Last month, US-based edtech company Chegg reported in a trading update that its business has been negatively impacted by the growing use of generative AI. Although the first-quarter results were relatively stable, the effects have become more pronounced over the past two months.
While some edtech platforms dismissed the potential of ChatGPT and faced near failure, others, like Khan Academy, recognized its value early on and adopted it. For example, in March, Khan Academy introduced Khanmigo, a virtual tutor and learning guide powered by GPT-4.
In India, edtech firms are beginning to follow this trend, but the industry is currently in a fragile state. After experiencing rapid growth during the pandemic, these firms are now struggling with instability as conditions return to normal. Many are dealing with challenges in maintaining stability, resulting in downsizing and layoffs. As a result, generative AI could prove to be a crucial asset, offering the potential for highly personalized learning experiences for students.
“Edtech companies are poised to serve as a UX layer, offering context, capacity, and creativity that complement generative AI. When this UX layer is integrated with Natural Language Processing (NLP) and supported by extensive machine learning data, it can guide generative AI to craft meaningful learning experiences,” explained Rishabh Ranjan, Chief Data Scientist at SwiftChat by ConveGenius, in a conversation with AIM.
Indian edtech firms are leveraging generative AI.
In India, several edtech firms are harnessing large language models (LLMs) to deliver hyper-personalized learning experiences to students. According to Mayank Kumar, co-founder and managing director of upGrad, the company is considering developing its own proprietary LLM. “These advanced models have significantly enhanced our educational services and provided valuable insights to our learners. We are also looking ahead to potentially creating our own proprietary LLM to further advance our offerings.”
upGrad has developed a GPT-powered chatbot that enables learners to conduct mock interviews at their convenience. “Our chatbot provides real-time performance scores and feedback, along with guidance and corrective measures to help learners enhance their skills,” said Kumar. They are also working on creating a coaching bot.
Akshay V, founder of EdZola, shared with AIM that they have been utilizing GPT model APIs long before the launch of ChatGPT to explore how generative content can benefit nonprofits and educational institutions. Recently, Byju’s, the leading startup in India, unveiled Wiz, a collection of AI models designed for hyper-personalized learning. This suite includes BADRI, MathGPT, and TeacherGPT, which will be integrated into Byju’s entire product range, from Pre-K to Gray.
Cuemath aims to enhance math education by enabling teachers to tailor their teaching methods to individual students. “We’re developing various use cases and experimenting with technology that provides teachers with copilots to generate prompts for student guidance and problem-solving,” said Manan Khurma, CEO and Founder of Cuemath.
Cuemath is also investigating the potential of large language models (LLMs) to create content for math learning, practice, and exam preparation. “If our experiments prove successful, we will be able to design personalized test papers based on each student’s learning journey and problem-solving abilities, rather than applying the same level of complexity to all students,” Khurma added.
SkillUp Online, which specializes in IT certification, is leveraging generative AI to enhance learner engagement by offering personalized and immersive teaching support. “We are also exploring how this technology can empower teachers and instructors to deliver content more effectively and efficiently,” Ratan Deep Singh, CEO and Chief Evangelist of SkillUp Online, told AIM.
Additionally, other companies like ConveGenius, Unacademy, and Doubtnut are also utilizing generative AI. AIM reached out to several other edtech firms, including Vedantu and TimesPro, but they chose not to disclose any information on their use of this technology.
Generative AI is adding to regulatory challenges.
Regulating edtech firms in India has been an ongoing challenge. While significant measures have been implemented, the advent of generative AI introduces new regulatory concerns. “AI technologies in education can revolutionize learning, but they also bring up issues related to privacy, security, bias, and ethical considerations,” stated Advocate Satya Muley, founder of Satya Muley & Co., in a previous conversation with AIM.
In response, the Internet and Mobile Association of India (IAMAI) has taken a proactive approach by establishing the India Edtech Consortium (IEC). This independent body aims to encourage self-regulation within the edtech sector. The IEC’s main goals are to create a comprehensive code of conduct for edtech companies and to set up an effective grievance redressal mechanism for consumer issues. However, the consortium’s focus on AI systems and ethical considerations within edtech startups remains somewhat limited.
Large Language Models (LLMs) like GPT-4 exhibit fewer hallucinations compared to GPT-3.5, but they are not entirely free from errors. For instance, a professor at Texas A&M-Commerce failed over half of his students when ChatGPT mistakenly claimed authorship of their papers, leading the university to withhold their diplomas. Such incidents underscore the challenges associated with LLMs. Tools like Byju’s MathGPT could potentially provide incorrect answers to students or generate factually inaccurate content. Imagine a student failing an exam due to learning from erroneous information generated by an LLM. Byju’s has stated that their models have a 90% accuracy rate, which still leaves some room for error.
Kumar from UpGrad explained that they address these challenges through multiple approaches. They employ a robust filtering mechanism, which includes data filtering, comprehensive training, and human review. “Our team of human reviewers is essential in assessing the text generated by LLMs. Their expertise ensures the accuracy, safety, and quality of the content produced,” he concluded.
