ChatGPT has presented a significant challenge to the Edtech Firms forcing stakeholders to either embrace its capabilities or risk falling behind. Last month, US-based edtech firm Chegg, in a trading update, stated that the company’s business has been adversely affected by the increasing use of generative AI. While the first-quarter results were relatively unaffected by this trend, the impact has become more noticeable in the past two months.

At a time when some edtech platforms undermined the power of ChatGPT and ended up at the cusp of failure, there were others like Khan Academy that assessed the potential of generative AI early and embraced it. For instance, in March, Khan Academy launched Khanmigo, a GPT-4-powered virtual tutor and learning guide. 

In India, edtech firms are following suit, however, the industry is currently facing a precarious situation. While these firms experienced rapid growth during the pandemic, they are now grappling with instability as normalcy gradually returns. Many of them are facing difficulties in maintaining stability, leading to downsizing and layoffs within the sector. Hence, generative AI could emerge as a saviour for the sector as it could help provide hyper-personalised learning experiences to students. 

“Edtech firms will continue to be a UX layer (that provides context, capacity, and creativity) above generative AI. And if this UX layer is enabled with Natural Language Processing (NLP) and has rich Machine Learning data, edtech firms can prompt generative AI to create meaningful learning experiences,” Rishabh Ranjan, chief data scientist at SwiftChat by ConveGenius, told AIM.

Indian edtech firms leveraging generative AI

Today, in India, we have a host of edtech firms using Large Language Models (LLMs) to provide hyper-personalised learning experiences to students. Mayank Kumar, co-founder & managing director at upGrad told AIM that the edtech firm is exploring the idea of building their own proprietary LLM. “These cutting-edge models have proven to be invaluable assets in enhancing our educational offerings and providing valuable insights to our learners. However, we also have an eye towards the future, as we are actively exploring the possibility of building our own proprietary LLM.” 

upGrad has also built a GPT-powered chatbot to allow learners to conduct mock interviews at their convenience. “The chatbot is designed to offer real-time performance scores and feedback to our learners while also providing them with guidance/corrective measures to improve their skill sets,” Kumar said. Additionally, they are also in the process of developing a coaching bot.

Akshay V, founder at EdZola also told AIM that they have been using GPT models API’s long before ChatGPT was announced to explore how generative content can help nonprofits and educational institutions. Most recently, Byju’s, the most valuable startup in India, announced Wiz, a series of AI models for hyper-personalised learning. The suite includes BADRI, MathGPT, and TeacherGPT and these models will be integrated across the entire product portfolio of Byju’s from Pre-K to Gray.

Similarly, Cuemath wants to solidify the maths learning approach by helping teachers customise their teaching method for each of their students. “We are currently working on multiple use cases and experimenting with this technology with copilots for our teachers to create prompts that they can refer to, cue students and help them solve the problem,” Manan Khurma, chief executive & founder at Cuemath, told AIM.

Additionally, Cuemath is exploring the use of LLM to generate content for maths learning, practice and exam preparation. “If the experiments turn out to be successful, we will be able to curate unique test papers for students that will be specifically based on their learning journeys and their abilities to solve maths rather than putting them all to solve the same level of complex mathematical problems,” Khurma added.

SkillUp Online, which offers online IT certification, is also using generative AI to drive deeper engagement with learners, providing personalised and immersive teaching assistance. “It is also exploring how the technology can be used to empower teachers and instructors to deliver content more effectively and efficiently,” Ratan Deep Singh, chief executive & chief evangelist, SkillUp Online, told AIM.

Furthermore, there are a few other firms which are leveraging generative AI such as ConveGenius, Unacademy and Doubtnut. AIM also reached out to a host of other edtech players in the industry like Vedantu and TimesPro among others, however, they refused to share any details in this regard.

Generative AI adding to Regulatory challenges

Regulating the edtech firms in India has remained a concern for a long time. Even though significant steps have been taken in this regard, the use of generative AI gives regulatory concerns a whole new dimension. “AI-based technologies in education have the potential to transform the way students learn, but they also raise concerns about privacy, security, bias, and ethical considerations,” Advocate Satya Muley, founder at Satya Muley & Co, had earlier told AIM.

In this regard, the Internet and Mobile Association of India (IAMAI) has taken a proactive step by establishing the India Edtech Consortium (IEC), an autonomous body aimed at fostering self-regulation within the edtech industry. The IEC’s primary objective is to develop a comprehensive Code of Conduct for edtech companies, along with implementing an effective grievance redressal mechanism for addressing consumer concerns. However, the consortium’s focus on AI systems and ethical considerations within edtech startups remains relatively limited.

LLMs like GPT-4 hallucinates comparatively lesser compared to GPT3.5, its predecessor. However, it does hallucinate. A professor at Texas A&M-Commerce failed more than half of his students when ChatGPT incorrectly asserted that it had written their papers. Consequently, the university chose to withhold the affected students’ diplomas. Incidents like this highlights the challenges that come with LLMs. Tools like Byju’s MathGPT, for example, could provide wrong answers to students, or LLMs could generate content which are factually wrong. Now, imagine a situation where a student fails an exam by learning inaccurate content generated by a LLM? Byju’s have mentioned that their models have an accuracy of 90%, which means there is room for error, even though minute.

Kumar said upGrad deals with these challenges in multiple facets. They have a robust filtering mechanism, which includes data filtering, comprehensive training and also human review. “Our team of human reviewers plays a crucial role in evaluating the text generated by LLMs. Their expertise and discernment contribute to maintaining the accuracy, safety, and overall quality of the generated content,” he concluded. 

The post The Secret Weapon for the Survival of Indian Edtech Firms appeared first on Analytics India Magazine.