Anyone with an interest in leveraging generative AI (more specifically, large language models) to enhance STEM related education (physics, chemistry, biology, engineering, mathematics, computer science education, etc...) for students can utilize this website!  Please just cite this website if you are using it for academic publishing

The guiding principle for the work below, is that, the way academics, researchers and industry people discover methods on how to improve the performance of various task of large language models (or multi modal large language models), is PARALLEL to how we (as teachers) try to teach our STEM students how to improve their task performance on course assessments (such as assignments, quizzes, midterm tests and final exams).   Simple examples for which professional industrial LLM modellers are implementing to improve "quantitative reasoning" in math and science question answering are  "zero shot" or "chain of thought "(CoT) to answer a question.  

Let me explain further. As teachers, we have actually used "zero shot" and "CoT" methods in our teaching strategy when lecturing to our students on how to solve science and math questions, for centuries!  For example, the general strategy to teaching our students how learn about a topic/idea/concept (e.g., limits in Calculus) and how to solve a math and science problem, is to first introduce the idea, give the definition, and give a "simple example" of how to apply the idea to solve a given problem.  To build their understanding of the concept and idea,  we show them how to solve more complicated problems, by showing them the step by step solutions.  Students, then, will most likely mimic and understand how to solve similar problems, by following our step by step solutions.

Therefore, the aim of this work is to (1) find out and list all the researchers work (in academia or industry) on improving task performance on LLMs on quantitative reasoning, and (2) explore whether it is feasible to integrate or transform the prompt engineering technique (for improving task performance of LLMs on solving science and math problems) into an active teaching pedagogy in science and math education to help improve STEM students' learning outcomes.

Finally, I believe the grand goal here is to help students eventually learn how to use AI to solve real world problems (e.g., in science, engineering, etc....).   Why: there is already evidence that AI helps us solve some of the most difficult scientific problems: Google Deepmind has shown us, using AI and deep learning, we maybe able to solve many scientific problems in the world, starting with AlphaGo to beat professional Go players; AlphaFold to fold proteins (now up to AlphaFold3),  AlphaTensor to speed up a calculation at the heart of many different kinds of code; AlphaDev to make key algorithms used trillions of times a day run faster; and finally, FunSearch (function search) to solve an old mathematics problem.  In fact, as a classically trained mathematician, I also believe that AI is set to revolutionize mathematics.  Albeit, Google Deepmind's approach is more fundamental and foundational, students these days should be aware and have access to AI (in this case, generative AI), which will help them get an intuitive feel of what AI is, how they work, what they are able to do (and not do), and hopefully, begin to use them to solve some hard problems.  

Few shot Prompting is where you would string together two or more CoTs, before asking the original question.  In fact, when companies  (OpenAI, Google, Meta, Anthropic, etc...) evaluate their (open sourced or proprietary) models, they always use few-shot prompting to test their models against on bench marked test against competing models.  

Language Models are Few-Shot Learners https://arxiv.org/abs/2005.14165 ​

Click here for examples on relating, e.g., the concept of "limits" in Calculus, voltage and quantum superposition in Physics, and cellular respiration in Biolology to Blackpink!, and more!.

Click here for on how to personalize your examples.

Please click in this Section to find out research on generative AI in science and mathematics education; including mathematics, computer science, physics, chemistry and biology education!

This is joint work with Prof. Stephen CHOW and his Ed.D. student Mr. Lim.  In this work, we explore how using a teachable machine for plastics classification education influences students' knowledge, attittude and behavior towards green sustainability measures.