How LLMs Will Revolutionize Education
What comes to mind when you think of AI in education?
Do you think of students using ChatGPT to cheat in class? That is a valid opinion, but narrow at the same time. There are many useful applications of AI that will change your perception of using it in our education system.
What are LLMs?
Large Language Models (LLMs) are a subset of artificial intelligence (AI) which uses artificial neural networks to learn from massive amounts of data. Artificial neural networks are just fancy words for saying an artificial brain. A human brain is made of around 86 billion neurons, with over 100 trillion connections!
ChatGPT, by OpenAI, is a large language model that was trained on hundreds of gigabytes of text from the internet. The latest version of ChatGPT is estimated to consist of 100 trillion parameters! If you were living under a rock, ChatGPT is essentially a very powerful autocomplete that has gained immense popularity in the last year. While traditional autocomplete on smartphone keyboards can complete a sentence, LLMs like ChatGPT can synthesize so much more information! It can even take small prompts and create its own stories, and poems, and even write code!
ChatGPT is not the only LLM out there, but it is the most accessible and has the largest training. There are other competitors like Google’s Bard, which is another LLM that accomplishes the same thing but with a different dataset.
The inner workings of neural networks
Artificial neural networks are made up of nodes, which can be compared to the neurons of a biological brain. These nodes can contain a numerical value between 0 and 1 to show the strength of activation.
Deep neural networks have more than 3 layers of nodes, interconnected with each other.
- The first layer is where the inputs go. In LLMs specifically, there is only 1 node for the input layer as it processes the entire input text, which is converted into a numerical format that can be understood mathematically.
- The middle layers consist of many nodes that are interconnected to each other, with varying strengths of connection.
- The last layer is the output of the neural network, with the number between 0 and 1 representing the confidence of its answer.
At first, the middle layers are set to random values and strengths. As data is inputted, the model’s answer is compared to the correct answer and adjusts all the knobs of the nodes and connections. The more data is inputted, the more accurate the model becomes over time.
LLMs like ChatGPT use deep neural networks to predict the next word when given a prompt. They keep predicting more and more words as they go, resulting in an output with lots more text. This method only works because of how large the dataset is and through lots of training and fine-tuning all the nodes and connections.
Problems with the education system
In our current education system, there is a high student-to-teacher ratio. When you were in school, have you ever wanted to ask the teacher something, but the teacher never got to you because others needed help? Maybe you were struggling with something but couldn’t talk to the teacher one-on-one or share it in front of the class?
The learning provided in school is not personalized to each individual, rather it is designed to meet the needs of many students. This has the side effect of the content being too difficult or too easy for individuals.
Individuals may prefer or understand different teaching methods that may not apply to a whole class. For example, some students prefer writing tests or exams while others may not. Some prefer working together in groups while others like to work by themselves.
The education system is also outdated and struggles to catch up with newer technologies that are crucial to leap forward in our society. Even in the time of the pandemic, the poor usage of technology and adaptation proves just how rigid our education system is.
How can LLMs help the education system?
- Personalized learning. LLMs have the potential to assist students when they need help, wherever and whenever. AI can integrate interests that an individual may have into lessons. This will keep the content engaging and fun to learn. If the learning is personalized to students’ interests, they can better grasp the information as it will be relevant to them.
- Tutoring. Students can learn at their own pace with guidance from AI. Talking to an LLM is more humanlike than searching the web, which can be helpful for students who need to ask questions or more clarification on a subject. When I was younger, we learned how to search effectively on the internet. Now, you could ask an AI like you were talking to a person and it will give you an answer! It’s awesome how much technology has improved in less than a decade.
- Help teachers with lesson planning. LLMs can assist teachers in lesson planning, formatting documents, building rubrics for assignments, creating tests, and so much more. Teachers are overworked and underpaid; with an AI model assisting teachers, they could get things done efficiently without sacrificing time.
- Providing feedback. AI could help provide feedback for students with drafts of their assignments. AI can act as a peer reviewer, providing constructive feedback. This could go the other way, too! Students could use their critical thinking to spot what’s wrong with AI-generated content, indirectly improving themselves.
Why is AI frowned upon in education?
There are many ethical problems with using AI in the education system. However, some solutions can make this possible in the near future.
- Students may use AI to cheat. While this is an issue, the education system should be the one to blame. Even before AI, students could have used other students to cheat on assignments, tests, or projects. All AI did was make general knowledge more accessible, much like the internet. The method of teaching and demonstrating knowledge must be reworked. AI is not going away, it is here to stay. By realigning the education system to demonstrate more critical and creative thinking that is personal, this issue can be mitigated.
- AI stole information without crediting it. Even though this is true, newer models such as Bing AI are working on including references to the websites that the information was from. While this is a tough issue to resolve, the fact that the information is not credited could be used to our advantage. Students could be forced to cite where they found their information, prompting them to research differently.
- AI can be biased or misinformed. Since the answers from the AI models depend on what data was fed into it, the AI could develop a political bias that would manipulate the answers. The data that is inputted into the models must be certified to include no bias, which is hard to do. Nevertheless, students could critique the AI and use critical thinking to make their own decisions on misinformation.
Are there any companies trying to make this happen?
Khanmigo by Khan Academy is an AI-powered guide that is in the works! Khanmigo will assist both students and educators by connecting ChatGPT-4 to the vast library of free resources that Khan Academy offers. Khanmigo will be able to assist students with the benefits of one-on-one tutoring for all students. It will also free up time for teachers, which is very precious. This is a fantastic integration of LLMs and education since Khanmigo can provide personalized learning to students and help them along their journeys.
MagicSchool is an AI assistant for teachers that is focused on saving teachers hours to help plan lessons and write assessments. Many of the teachers who have used MagicSchool have mentioned that it saves a lot of time and keeps learning personalized! Using AI tools like MagicSchool can prevent teachers from getting burned out.
Conclusion
Large language models will revolutionize the current education system by supporting students with convenient personalized learning. They will also cut the number of hours that teachers spend, helping them with lesson planning and creating assessments. However, to do so, the current education system needs to move away from just testing knowledge and move towards assessing the critical and creative thinking of the students. AI is not going anywhere, and it is time for the rigid education system to change. LLMs have great potential to ease the workload for teachers and personalize the learning for students.
References
AI for teachers — lesson planning and more!. MagicSchool.ai — AI for teachers — lesson planning and more! (n.d.). https://www.magicschool.ai/
Christensen, U. J. (2023, October 4). How AI will revolutionize human learning — but not the way you think. Forbes. https://www.forbes.com/sites/ulrikjuulchristensen/2023/05/25/how-ai-will-revolutionize-human-learning—but-not-the-way-you-think/?sh=41dd7e0a4c65
García, E., & Weiss, E. (2019, March 26). The teacher shortage is real, large and growing, and worse than we thought: The first report in “The perfect storm in the teacher labor market” series. Economic Policy Institute. https://www.epi.org/publication/the-teacher-shortage-is-real-large-and-growing-and-worse-than-we-thought-the-first-report-in-the-perfect-storm-in-the-teacher-labor-market-series/
Khan Academy. (n.d.). Khanmigo Education Ai Guide. Khan Academy. https://www.khanacademy.org/khan-labs
Layton, D. (2023, January 24). Chatgpt and Dall-E-2 — show me the data sources. LinkedIn. https://www.linkedin.com/pulse/chatgpt-dall-e-2-show-me-data-sources-dennis-layton/
Zhou, V. (2019, July 30). Neural Networks from scratch. Victor Zhou. https://victorzhou.com/series/neural-networks-from-scratch/
