HomeAI News
Learn 4 new uses of ChatGPT from the TED speech of the founder of Khan Academy
10

Learn 4 new uses of ChatGPT from the TED speech of the founder of Khan Academy

McMuffin
McMuffin
May 5th, 2023
View OriginalTranslated by Google

The original text is shared by @starzq.eth⛩️ on Twitter. If you have benefited from the article, please go to the original tweet and like it!

Yesterday, the founder of Khan Academy shared on TED the case of how to use GPT-4 to achieve 1-to-1 tutoring to improve student performance. It was very exciting. I also learned 4 new usages of ChatGPT and shared them with you

  • Reading Comprehension: Talk to characters directly, no more guessing
  • 1 on 1 tutoring: teach them how to fish
  • AI Debater: Exercise Thinking Ability
  • AI Tutor: Sharing Subject Significance

The following details are expanded, and the full video is at the end

Reading Comprehension: Talk to characters directly, no more guessing

"What is the purpose of XX doing this?" [Reading comprehension] is definitely the nightmare of most students. The author himself may have forgotten why he wrote it like this, but the answerer must say (make up) the reason...

Why not let ChatGPT bring in the character and tell us directly?

Khan gives an example: "Why does Gatsby always stare at the green light?"

This student turned Khanmigo (Khan Academy's GPT-4 based ChatBot) into Gatsby. This Khanmigo really entered the scene in a second. The answer is as follows

"Ah, green light, man! You know what, it's a symbol of my dreams and longings. It's at the end of Daisy Buchanan's pier, across the bay from my mansion. I stare at it long because it Represents my longing for the past..."

"...and the hope of being reunited with Daisy, the love of my life. The green light is a constant reminder of the elusiveness of the American dream, and the enormous distance between reality and vision. So tell me, have you ever had a seemingly distant What about unattainable dreams or longings?"

Talking to the character himself is such a great use! This way makes studying literature and history interesting

Then I asked "Lu Xun" an eternal problem

1 on 1 Mentoring: Teach him to fish, not teach him to fish

Many people's concerns about GPT are that students can get answers directly without developing the corresponding ability

Khan’s response: Students can copy homework with or without GPT. And on the contrary, its customized Khanmigo will greatly reduce homework copying and improve students' abilities at the same time, because

  • It will be displayed on Khanmigo: All chat records will be seen by the teacher to prevent students from "taking shortcuts"
  • When the student said, "Tell me the answer," Khanmigo responded, "What do you think is the next step in solving this problem?" Take students through the problem step by step
  • When students make mistakes, it can find out and point out in time, strengthening the reasoning process
  • This is a process of "teaching him to fish", allowing students to know what they are and why they are 6/17

Khan said that they got the permission of GPT-4 in August last year, and then spent 6 months doing the prompt project, and a lot of fine-tune, so that Khanmigo learned how to "teach him to fish" step by step

When we encounter problems looking for ChatGPT, we can also try a similar prompt "Teach me how to do xxx step by step" instead of directly displaying the answer

AI Debater: Exercise Thinking Ability

The third example that impresses me is "debate".

As we all know, things become more and more clear when debated, but suitable "debaters" are hard to find. With the power of GPT-4, every student can learn about the essence of inquiry through Socratic dialogue and debate

AI Tutor: Sharing Subject Significance

This example also impressed me. Students ask "Why study cytometry? My goal is to be a professional athlete"

Khanmigo replied "Great! As a professional athlete, understanding the science of size can help you understand how your body works at the cellular level. For example, understanding the interactions between cells, proteins, and molecules can help you..."

Is there any reason not to study hard?

Meaning is the reason why we do most things, but as a newcomer, it is difficult to get why we need to learn xxx, and in the end, we did not master a lot of important knowledge

Wang Huiwen also gave an example, saying that he regretted not learning well at the beginning is linear algebra. If someone had told him that this subject was the foundation of artificial intelligence, he would have mastered it well

With ChatGPT, students will never have such regrets again

To sum up, the 4 important uses of ChatGPT tools in education are also very helpful for adult learning

Reading Comprehension: Talk to characters directly, no more guessing

1 on 1 tutoring: teach them how to fish AI Debater: Exercise Thinking Ability

AI Tutor: Sharing the Significance of Subjects In 1984, a study showed that 1-to-1 tutoring can greatly improve student performance, but the problem is that the cost is too high. AI allows everyone to have the opportunity

Talk about my views on GPT/LLM in education

Both human and AI capabilities can be abstracted as "data + algorithm + computing power"

The new era has arrived, and new industries will produce new data every few years. The requirement for our ability is to quickly call the algorithm (brain), and with the support of computing power (body), based on data output result

The New Normal: Spend 1 year becoming an expert, work for 3 years, then move to a new field

No knowledge and skills can last forever, our whole life will be spent constantly learning new knowledge and new skills

Traditional education emphasizes the importance of "data". In the new era, the shelf life of data is greatly shortened, and more importantly, algorithms and computing power

This reminds me of the battle between the two paradigms in the field of AI: BERT+Fine tuning vs LLM+prompting 13/17

  • BERT+Fine tuning (traditional AI): Let a child who has just learned to read go to the assembly line to train on screwing until he is 18 years old and becomes an excellent screwer—but he is not competent for other jobs.
  • LLM+prompting (ChatGPT): Use a large amount of computing resources to train children until they graduate from university—although they have not become skilled workers, they can learn by analogy in many fields, learn from one instance, and get started quickly.

The talent requirements in the new era are the same as building ChatGPT, the ability to learn by analogy >> a single skill in a certain field

Many people criticized ChatGPT for inaccurate answers. OpenAI co-founder @Greg Brockman responded that the most valuable thing is reasoning ability, which is the bottom layer of learning ability

So my attitude is to integrate with education as soon as possible and use AI to improve HI (Human Intelligence)

Reprinted from starzq.ethView Original

Comments

no dataCoffee time! Feel free to comment