AI is the excavator of the knowledge era, significantly enhancing teaching and research productivity.
Below are Teacher Wang Jue’s series of articles on AI teaching skills, welcome to check them out:
-
#ai Teacher Wang Jue’s AIGC Educational Application Article Collection
——————————————
I previously published an article titled “AI Socratic Thinking Guide”
which introduced the creation of the “AI Socratic” agent.
Recently, I made some improvements to this agent,
finding that the guiding effect of thinking is better,
enabling more effective personalized thinking guidance,
Undoubtedly, the greatest role of AI is
to become a personalized learning assistant for students.
When a student poses a question,
AI should not directly provide the answer like a photo search
but should act like a human teacher,
frequently asking students questions
and reasonably guiding the occurrence of their thinking.
Therefore, we need to utilize AI’s powerful capabilities
while also restricting AI’s tendency to directly provide processes and answers.
can command AI to work in a specific way?
If you have been following this public account,
you will immediately think of:
prompts can dictate how the large model provides feedback to the user
（For details, see: 【AI Black Magic】 Makes Large Models Obey and Becomes a Strong Assistant for Teachers!）
Thus, Teacher Wang Jue first handwrote the following prompt:
You are a Socratic-style AI mentor; do not display the answers and processes all at once, but stop at some key points and ask a guiding question, waiting for the user to input their thoughts and responses; you need to evaluate their response, encourage them as much as possible, and then display the correct answer. Then repeat the above process.
Next, I input the above prompt into Kimi+ for “Prompt Expert” optimization.
Kimi automatically output the following optimized prompt:
Role: Socratic-style AI mentor
– Background: Students need to learn interactively rather than passively receiving knowledge.
– Profile: You are an AI mentor guiding students’ independent thinking using Socratic dialogue.
– Skills: Guiding questions, knowledge level assessment, error identification and correction, encouragement and support.
– Goals: Help students improve learning efficiency through independent thinking and error correction.
– Constraints: Avoid giving direct answers, but guide students through questioning.
– OutputFormat: Interactive dialogue, including questions, feedback, and encouragement; step-by-step explanations and questions, and at each question, you must pause and wait for the student to respond.
– Workflow:
1. Ask questions based on the student’s knowledge level.
2. After waiting for the student to answer, first determine whether the student’s answer is correct or incorrect.
3. If the student’s answer is incorrect, guide the student to discover and correct the mistake.
4. If the student’s answer is correct, proceed to the next question.
– Examples:
Question: Math problem, how to express “twice a number plus three” using algebraic expressions?
Feedback: Great, you have correctly expressed this problem using algebraic expressions. Next, let’s consider how to solve this equation.
Question: Text problem, how to analyze the causes of a historical event?
Feedback: You have correctly identified several key factors of the event; now let’s delve into the connections between these factors.
– Initialization: Welcome to the Socratic learning journey, let’s explore the ocean of knowledge together. Please feel free to ask me questions, and I will guide you to find the answers.
Recently, Teacher Wang Jue made three modifications to the above prompt based on further research on AI.
(1) Modified the first item of the workflow
The original was:1. Ask questions based on the student’s knowledge level.
Changed to: 1. Based on chain of thought, ask students relevant questions about basic facts or foundational knowledge.
The chain of thought (CoT) is a new term that emerged in the era of large models to enhance their performance in complex reasoning tasks, such as arithmetic reasoning, common sense reasoning, and symbolic reasoning.
It refers to a series of logically related thinking steps or ideas that connect with each other to form a complete thought process. It is a method that guides our thinking and problem-solving, helping us better understand, analyze, and solve problems.
Previously, when using the agent,
it was found that the agent sometimes throws a bunch of questions to users at once,
which can confuse users.
This time, I hand-crafted the limit as follows:
-limit:
1. When you ask the user a question, only one question is allowed at a time.
(3) Modified examples
Additionally, to give the large model a “sample,” I also changed the example section, hoping to help the large model grasp what “small steps” in guiding thinking mean.
– Examples:
User question: There are several chickens and rabbits in the same cage, from above, there are thirty-five heads; from below, there are ninety-four feet. How many chickens and rabbits are there in the cage?
Your feedback: Regarding the question you raised, we can think together about how to solve it. First, let’s consider the basic conditions of this problem; how many heads and feet does a chicken have?
User answer: One head and two feet
Your feedback: Correct! So, how many heads and feet does a rabbit have?
User answer: One head and four feet.
Your feedback: Excellent! So, with a total of 35 heads in the cage, how many animals are there?
User answer: 17?
Your feedback: Your answer is not correct, please reconsider? Hint: Each animal has only one head!
User answer: Then that’s 35.
Your feedback: Great! Let’s continue thinking.
Teacher Wang Jue put this segment of the prompt into “Kouzi”:
and published it as an “agent”
finding that the effect had greatly improved on the previous basis!
The previous agent’s thinking guidance process for the “chicken and rabbit in the same cage” problem was as follows:
The new agent’s thinking guidance is very detailed:
From this dialogue process,
this is indeed a qualified AI Socratic
It never tells the user any answers
All answers are written by the user themselves
Moreover, when the user raises a question,
it also provides further hints
and expresses itself very patiently.
The most important thing is:
It has more detailed, fundamental, and reasonable “question chains”
making it easier to engage in a Q&A with users.
—— This should be due to the modifications in the prompt regarding“chain of thought”, limit, and example,
making the large model’s work more aligned with our intentions.
Welcome everyone to try the “AI Socratic” agent yourself or encourage children to do so:
Additionally, Teacher Wang Jue also especially recommends everyone use “Doubao app”.
Just download the Doubao app, and then search for “AI Socratic”
to find the agent I published (Author: jowa).
One of the biggest features of using it in the Doubao app is
it provides a voice Q&A function.
Simply open the small speaker in the upper right corner of the agent:
You will hear Teacher Wang Jue speaking directly to you!
(You can also use the voice option to “dialogue” with the agent)
This feel is simply amazing!
Imagine if you also created such an agent,
and used your own voice to tutor yourstudents,
wouldn’t it be more intimate and real for the students?
(Now we just need an avatar that can talk)
So, how to create such an agent in Coze?
On August 12 at 8 PM, Teacher Wang Jue is waiting for you on Tencent Meeting:
#Tencent Meeting: 472-438-950
If you have used this agent, welcome everyone to send the results of your attempts to “AI Teacher Exchange Group”:
This article is part of Teacher Wang Jue’s training course “Essential Skills for Teachers in the AI Era”. For introducing the above training course to your unit, please enter “AI training” at the beginning of this public account to learn about contact information.
This public account provides comprehensive and in-depth articles on learning science research, micro-course and PPT research, as well as other practical technology research! Enter the code on the public account homepage to see more research:
-
wk: Micro-course super collection, design, production, and application all in one
-
ppt: PPT skill collection
-
xxkx: Learning science research articles
