LearnLM, Phi-3, ChatGPT Edu: Building AI Models for Education
C
Contenjoo
Last modified
Date
Empty
With the advancement of AI technology, the use of AI in education is also rapidly expanding. Large tech companies such as Google, Microsoft, and OpenAI are trying various approaches to optimize their AI models for education. Among them, Google's LearnLM, Microsoft and Khan Academy's Phi-3, and OpenAI's ChatGPT Edu are attracting attention recently.
Google's LearnLM is characterized by developing a conversational tutoring model specialized for education based on the Gemini model of the verb. The research team disclosed the development process and results of LearnLM in detail through a massive 86-page research paper. They recognized that defining and measuring effective teaching methods is not easy, and identified core teaching strategies such as encouraging active learning, managing cognitive load, deepening metacognition, stimulating motivation and curiosity, and adapting to learners.
LearnLM evolved into a model optimized for education through fine-tuning, not just simple prompting. The research team fine-tuned LearnLM step by step using various datasets such as human tutoring data, generative AI role-play data, math problem conversation data, high-quality conversation data written by teachers, and safety data. Through this, LearnLM was able to learn natural conversation, structured and consistent teaching methods, step-by-step problem-solving explanations, and high-quality teaching practices.
The performance of LearnLM was evaluated in various ways, including quantitative, qualitative, automated, and human evaluation. First, it passed the existing benchmark test in terms of safety and bias, and received high scores from human evaluators in terms of improving learner confidence, promoting student participation, and encouraging active learning compared to the base model. In automated evaluations, LearnLM also showed superior performance to the base model in most educational tasks.
However, LearnLM also has limitations. It is still lacking in fully implementing the advantages of human teachers, such as contextual understanding, personalized guidance, nonverbal communication, and multimodal interaction in actual educational settings. Above all, additional empirical research seems necessary to determine whether the improvement in educational capacity directly leads to actual learning outcomes.
Microsoft and Khan Academy are exploring the development of cost-effective and accessible AI tutoring tools through the Small Language Model (SLM) Phi-3, which is differentiated from existing large language models. Phi-3 provides many of the features of large models while being easy to use even for resource-constrained institutions. The two companies plan to combine Khan Academy’s extensive educational content with the Phi-3 model to enhance tutoring capabilities, especially in math.
OpenAI has released ChatGPT Edu, a specialized version for universities based on its flagship model, GPT-4. It has advanced features such as data analysis, web search, and document summarization, as well as strong security and privacy protections to ensure that AI can be safely utilized within universities. University members can build and share customized GPT models through ChatGPT Edu.
As competition in AI in the education sector accelerates, large tech companies are each seeking to differentiate themselves with their own strategies. Google is leading the development of education-specific models such as LearnLM based on its strengths in both extensive research capabilities and product and distribution networks. Microsoft is working on developing an economical and accessible solution based on SLM through a partnership with Khan Academy. OpenAI is accelerating its advance into the higher education market with ChatGPT Edu.
Of course, there are still many issues to be resolved in the development of AI for education. Above all, it is urgent to empirically prove whether AI's improved educational capabilities lead to actual learning outcomes. It is also necessary to consider ethical issues and ways to resolve the digital divide. Nevertheless, the efforts of the education, industry, and academia to cooperate and use AI technology as a driving force for educational innovation are highly commendable. A series of innovative cases, such as LearnLM, Phi-3, and ChatGPT Edu, allow us to gauge how AI can change the future of education.
Subscribe to '오늘배움'
Get the latest on education trends, AI-based learning know-how, and corporate talent development insights at Learn Today. Receive useful news as we explore the future of digital learning together.
