Building Human Intelligence at Scale, to Save the Next Generation from ChatGPT

Introduction

The Simons Institute is introduced as the leading venue for collaborative research in theoretical computer science. The lecture series, "Theoretically Speaking," is also introduced.

• The Simons Institute was established in 2012 with a generous gift from the Simons Foundation.

• The primary aim of the programs at the Simons Institute is to explore algorithmic thinking and science and engineering.

• The lecture series, "Theoretically Speaking," highlights exciting advances in theoretical computer science for a general audience.

Speaker Introduction

Po-Shen Loh, a professor of mathematics at CMU, is introduced as the speaker for the lecture.

• Po-Shen Loh has an active lecture schedule and reaches over 10,000 people through his lectures around the globe.

• Po-Shen Loh is a professor of mathematics at CMU and leads their math team to victories.

• Po-Shen Loh is also an innovator in education and healthcare.

Lecture Begins

Po-Shen Loh begins his talk on building human intelligence at scale to save the next generation from chat GPT.

• Po-Shen Loh expresses surprise at seeing some familiar faces in the audience.

• The objective of this talk is not about computer science behind GPT but what should be done next with education and the next generation.

Education and GPT-3

In this section, the speaker talks about how GPT-3 is going to shake up the US higher education industry for college admissions. The speaker also discusses how private equity companies can use GPT-3.

GPT-3 and College Admissions

• The speaker discovered that GPT-3 could write college admission essays.

• During a talk at Yale, the speaker asked an admission officer how they plan to deal with college essays next year with the availability of GPT-3. The officer responded by saying that chat GPT cannot write good essays.

• One of the goals of the essay was to reduce the number of colleges people apply to, but when you can generate an essay in five seconds, it becomes a different question.

• The speaker cautions against using chat GPT for cheating on college applications.

Private Equity and GPT-3

• Private equity companies can buy high and sell low by using GPT-3.

Fixing a Distressed Company

In this section, the speaker talks about a conversation he had with someone who runs a private equity firm and is tasked with fixing distressed companies. The speaker discusses how the person suggested laying off people as a way to fix the company.

Fixing a Distressed Company

• The person running the private equity firm suggested finding more efficient ways to write blog posts for search engine optimization by using chat GPT.

• The marketing team started to worry about their jobs because they were being told to use chat GPT to do their job.

• This made the speaker realize that layoffs and automation could affect his life in unexpected ways.

Using Chat GPT for Tutor.com

In this section, the speaker talks about how a Carnegie Mellon student used chat GPT to make money on tutor.com during winter break.

Making Money on Tutor.com

• The student created a program that was connected to tutor.com and answered questions as if he was a tutor.

• He manually answered questions if his program failed, and made $600 before getting shut down for violating terms of service.

• The speaker was surprised that chat GPT could answer math questions accurately and show all of its work.

Introduction

In this section, the speaker talks about their job and how they came to be interested in GPT-4.

Interest in GPT-4

• The speaker's interest in GPT-4 was piqued when it was released on March 14th.

• They were particularly interested in its performance on various exams, as shown by a visual from the GPT-4 team.

• The height of the bar represents how high of a percentile the GPT scores or ranks on standard exams.

• The blue is the old GBT 3.5, while the green is GPT-4, which is more sophisticated.

• It was able to take people from or take this thing on uh the tool but they try to do the AP Chemistry exam went from like about 20th percentile to about 70th percentile of humans.

American Math Competitions

In this section, the speaker discusses American math competitions and their relevance to GPT-4.

AMC Exams

• The AMC (American Math Competitions) are used to select people for the United States international math Olympia team.

• These exams are designed to be different from textbook problems and are created by committees specifically trying to make non-textbook problems.

• While most exams are supposed to be something you've learned how to do, AMC problems are supposed to be something you haven't seen before.

Harder Math Questions

In this section, the speaker explores harder math questions that can test GPT-4's capabilities.

Arithmetic Series Question

• The speaker poses an arithmetic series question that is not super sophisticated yet and could be learned in high school algebra.

• It shows every step of working out using a formula.

• The speaker is interested in testing GPT-4's capabilities with harder math questions.

GPT-3's Ability to Solve Math Problems

In this section, the speaker discusses how GPT-3 can solve math problems by ingesting teacher manuals and recognizing patterns.

GPT-3's Method of Learning

• GPT-3 can mechanically go through a lot of techniques because it has ingested teachers' manuals.

• The way to train these large language models is by feeding them language, and all standard arithmetic series problems have teachers' manuals explaining how to do that.

Recognizing Patterns in Math Problems

• GPT-3 can use the sum of the series 1 to 50 to solve one plus two plus 50.

• People often think someone is good at math if they know all the tricks, but that's not what we really say. It's like saying someone is smart if they know all of Wikipedia.

Examples of GPT-3 Solving Math Problems

In this section, the speaker provides examples of math problems that GPT-3 can solve.

Factorizing Numbers

• The speaker was surprised when he saw that GPT-3 could factorize numbers using tricks found in books.

• A math contest problem was given where participants had to factorize 99.

Solving Complex Equations

• The speaker shares a story about a math contest where participants had to find whole numbers X and Y so that 30X + 31Y + 28(1) = 365.

Conclusion

In this section, the speaker concludes his discussion on GPT-3's ability to solve math problems.

Impressive Capabilities

• The speaker was impressed by GPT-3's capabilities in solving complex math problems.

• The speaker describes the Monmouth Math Contest and how participants used different methods to solve a problem that GPT-3 could also solve.

Introduction

In this section, the speaker introduces the topic of GPT and its capabilities.

GPT's Capabilities

• The speaker mentions that GPT can do a lot of things and emphasizes the need to be careful with its usage.

• The speaker presents a math contest problem as an example to showcase GPT's abilities.

Math Contest Problem

In this section, the speaker discusses a math contest problem and how it relates to GPT's capabilities.

Solving the Problem

• The speaker attempted to solve the math contest problem using GPT but failed.

• The solution to the problem is presented, which involves multiplying 28 by 5.

Concerns about GPT

In this section, the speaker expresses concerns about GPT's impact on future generations and job loss.

Impact on Future Generations

• A parent in attendance expressed fear about GPT after seeing its capabilities.

• The speaker explains that many people have not explored GPT's full potential and are underestimating its impact on future generations.

Job Losses

• Outsourcing companies in some parts of the world are laying off thousands of employees due to advancements in language models like GPT.

• The speaker highlights how text-based communication jobs are at risk due to language models being able to learn from patterns in customer service work.

Rethinking Education

In this section, the speaker discusses how education should be rethought given advancements in language models like GPT.

Learning Methods

• Many educational systems teach students methods for solving problems rather than encouraging them to come up with new ways of thinking.

• Language models like GPT excel at finding patterns and following them, which is similar to how many students are taught in school.

Researchers

• The speaker acknowledges that researchers are working on language models like GPT and it's not a bad thing.

Teaching Students to Come Up with New Ideas

In this section, the speaker discusses how to teach students to come up with new ideas and why it is important.

Importance of Teaching Students to Come Up with New Ideas

• Practicing a lot of things does not necessarily teach students how to come up with new ideas.

• Many students are not used to coming up with their own ideas and have been taught that mathematics is all about having someone already tell them how to do it.

• By teaching students how to come up with their own way of solving a problem, they find out that they are capable of solving any question even if they've never seen it before.

How To Teach Students To Come Up With New Ideas

• The speaker teaches by asking the class for ideas on how to solve a problem or prove a theorem.

• The instructor's job is then to use the student's ideas to solve the problem.

• This teaching style works well in gifted classrooms or math clubs where people have agreed they want to think but may not work in general classrooms where only some students are interested in thinking.

Challenges Faced When Teaching Students To Come Up With New Ideas

• Some students may resist this teaching style because they expect teachers to tell them how to do things rather than figuring it out themselves.

• This teaching style requires more effort from instructors as they need to be awake and ready for unexpected solutions.

Math Team and the Importance of Human Interaction

In this section, the speaker talks about how being part of a math team in his public school helped him discover his ability to solve new questions. He emphasizes the importance of human interaction in facilitating brainstorming sessions and helping students feel loved and supported.

Importance of Math Teams

• Being part of a math team helped the speaker discover his ability to solve new questions.

• Math teams facilitate brainstorming sessions that allow students to come up with new ideas.

Importance of Human Interaction

• The speaker emphasizes the importance of human interaction in facilitating brainstorming sessions.

• Helping students feel loved and supported is an important aspect that cannot be automated.

Reaching Underserved Students

In this section, the speaker discusses the challenges involved in reaching underserved students who may not have parents encouraging them to excel at mathematics.

Challenges Faced by Underserved Students

• Underserved students may not have parents who encourage them to excel at mathematics.

• There is a gap that needs to be bridged when it comes to reaching these students.

Automating Education

In this section, the speaker discusses whether education can be fully automated or if there will always be a need for human involvement.

Automation vs Human Involvement

• There are two ways to attempt to solve the problem of automating education: by putting all teachers out of work or by having people do it. The speaker is on the side of employing people rather than putting them out of work.

• While it may be possible for a Silicon Valley company to come up with an automated solution, the speaker is personally not interested in pursuing that route.

The Importance of New Ideas and Being Human

In this section, the speaker discusses what he believes the next generation needs to do: come up with new ideas and learn how to be as human as possible.

Next Generation's Role

• The next generation needs to come up with new ideas and learn how to be as human as possible.

• There is a psychological aspect involved in education that cannot be fully automated.

Fuzzy Problems and Game Theory

In this section, the speaker discusses how problems in the physical world are often fuzzy and difficult to optimize. He explains that learning to solve these types of problems is important because they are harder to automate. The speaker also introduces the concept of game theory as a way to solve problems by aligning incentives.

Learning to Solve Fuzzy Problems

• Physical world problems are often fuzzy and difficult to optimize.

• Learning to solve fuzzy problems is important because they are harder to automate.

• Game theory can be used as a way to solve problems by aligning incentives.

Teaching People How to Come Up with Their Own Ideas

• The speaker wants to teach people how to come up with their own ideas in a way that can be done today.

• This approach cannot replace school but can be an addition.

• The ideal age group for teaching this is sixth grade or younger.

Using Game Theory

• The speaker approaches problem-solving through the lens of game theory.

• He started a program for high school students where he teaches them how to communicate better using professional comedians and actors/actresses.

Communication Skills for Math Students

In this section, the speaker talks about his program for high school math students who want to get into top universities. He explains that communication skills and interpersonal skills are crucial factors in getting accepted into these universities.

The Problem for High School Math Students

• High school math students who want to get into top universities face a challenge with their communication and interpersonal skills.

• The speaker started a program for high school students where he teaches them how to communicate better using professional comedians and actors/actresses.

Conclusion

In this section, the speaker concludes his talk by summarizing his approach to problem-solving and teaching people how to come up with their own ideas.

Summary

• The speaker's approach to problem-solving involves using game theory to align incentives.

• He wants to teach people how to come up with their own ideas in a way that can be done today.

• Communication skills and interpersonal skills are crucial factors in getting accepted into top universities.

Time-Saving Lecture Preparation

In this section, the speaker talks about how he saved time by preparing for lectures using domain knowledge. He also recommends this approach to anyone in academia.

Using Domain Knowledge to Prepare Lectures

• Spent a year doing classes for three hours a week.

• Never had to prepare for a single lecture again.

• Saves more time than you spend.

• Recommended for anyone in academia.

Importance of Combining Math and Performing Arts Skills

In this section, the speaker discusses the correlation between math ability and performing arts training. He emphasizes that people with both skills are not only good at interviews but will be extremely successful in the world.

Correlation Between Math Ability and Performing Arts Training

• High school students taught to combine math and performing arts skills.

• People with both skills are good at interviews and will be successful in the world.

• Strong sense of what to do with powerful skills is important.

Admissions Criteria for High School Program

In this section, the speaker talks about the admissions criteria for their high school program. They emphasize that being nice is an important factor in their selection process.

Admissions Criteria for High School Program

• Stringent admissions process based on several criteria.

• Being nice is an important factor.

• Students who meet criteria get paid instead of paying tuition.

Creating an Ecosystem for Drama People

In this section, the speaker talks about creating an ecosystem where drama people can find jobs while auditioning. This ecosystem connects with another part of their program that teaches live video math classes.

Creating an Ecosystem for Drama People

• Drama people given jobs in the ecosystem.

• Live video math classes taught to students.

• Connection between the two parts of the program is important.

Importance of Communication Skills

In this section, the speaker emphasizes the importance of communication skills and how they are developed through talking and getting feedback.

Importance of Communication Skills

• Learning to communicate effectively is important.

• Talking and getting feedback helps develop communication skills.

• Live video math classes provide opportunities for students to talk about math.

High School Students Teaching Math on Zoom

In this section, the speaker introduces a Zoom class taught by two high school students who are both very good at math. The speaker explains that for privacy reasons, they cannot show images of the middle school students being taught, but viewers can see the interplay between the instructors.

Introduction to the Zoom Class

• The Zoom class is taught by two high school students who are very good at math.

• Viewers can see the instructors on screen and their interactions with each other.

• For privacy reasons, images of the middle school students being taught cannot be shown.

Solving a Math Problem in a Zoom Class

In this section, the high school student instructors solve a math problem in front of their class using symmetry. They explain their thought process and how they arrived at their solution.

Solving a Math Problem Using Symmetry

• The high school student instructors solve a math problem using symmetry.

• They explain how they arrived at their solution step-by-step.

• One instructor points out that symmetry makes problems more elegant and fun to solve.

Running Large-Scale Zoom Classes Taught by High Schoolers

In this section, the speaker discusses how his organization is able to produce large-scale Zoom classes across the world taught by people who know math well. He explains that his team brought in high schoolers with strong domain knowledge and Performing Arts experts to make sure classes are entertaining.

Producing Large-Scale Zoom Classes Taught by High Schoolers

• The speaker's organization produces large-scale Zoom classes across the world taught by people who know math well.

• High schoolers with strong domain knowledge were brought in to teach the classes.

• Performing Arts experts were also brought in to make sure classes are entertaining and engaging for students.

I understand the task. However, I need the transcript to be able to create a summary in markdown format with timestamps. Please provide me with the transcript.

The Incentive Aligned Ecosystem

In this section, the speaker discusses how the ecosystem is incentive-aligned and how it benefits high school students and performing artists.

Facilitating Discussion

• High school students facilitate discussions in middle school classes.

• They ensure that ideas generated by middle school students are implemented.

• They play videos to guide the discussion and introduce new questions.

• Their job is done after an hour.

Incentive Alignment

• The ecosystem is completely incentive-aligned.

• High school students do not need to prepare for a pass or do anything other than participate in the program.

• Performing artists are paid for their work.

• There are enough high school students and performing artists to make this program scalable.

Scalability

• Approximately 100,000 high school students could benefit from this program each year.

• This number is on the same order of magnitude as the number of people who participate in math competitions like Mathcounts AMC.

• The ratio of high schools to middle schools is 10:1, which means that this program could potentially scale to millions of participants.

Success Story

• The speaker has been working in education for eight years before coming up with this idea in October 2021.

• He realized that he should focus on things that people want rather than what he thinks they need.

• As soon as he saw that his idea was very tight, he started building the technology.

Enrichment Classes for Middle School Students

The speaker discusses the market for enrichment classes and how their program fits into it. They also explain the benefits of their program beyond just academic enrichment.

Known Market for Enrichment Classes

• There is a known market for enrichment classes for middle school students in the South Bay Area.

• The program was thrown into this market, undercutting prices while increasing quality.

Benefits Beyond Academic Enrichment

• The program encourages high school students to show love, care, and encouragement towards middle school students.

• Co-teaching with another student builds confidence and helps them develop social skills.

• Two high school students are always included in each class to encourage friendship building and networking opportunities.

Scholarship Opportunities

• The program also has a scholarship arm that specifically targets underserved communities by recruiting high motivation, low access students from Title One schools.

• Philanthropists have donated classes to help these students at no cost to them.

The Importance of Storytelling in Teaching Entrepreneurship

In this section, the speaker discusses the importance of storytelling in teaching entrepreneurship and shares some personal stories to illustrate his point.

The Speaker's Background

• The speaker grew up in Silicon Valley and knew Steve Jobs.

• He looks for certain factors that increase the likelihood of success, such as whether a person tried to read the encyclopedia cover to cover as a child.

Using Storytelling to Engage Students

• Story elements are crucial because they help students remember techniques.

• High school students are taught various tactics, including storytelling, which they can deploy when needed.

• Fashion designer Jake Barnes uses storytelling to inspire others.

Tips for Effective Communication and Performance

In this section, the speaker shares tips for effective communication and performance that he has learned from his community of high school students.

Sharing Tips with High School Students

• Every week during office hours, high school students share tips about communication or performance.

• The community intentionally shares tricks so that each person can learn from others.

Finding Success Through Idealism and Being a Good Person

In this section, the speaker discusses how being an idealistic person who is also good can lead to success in new technologies.

Idealism and Being a Good Person

• When new technologies are invented, idealistic people who are also good tend to be successful.

Protecting Against Speakers and Purpose of Companies

In this section, the speaker talks about how they protect against their speakers and the purpose of companies.

Protecting Against Speakers

• The speaker explains that they don't take any investors to protect against their speakers.

• They are an entrepreneur who has built a cash flow positive bootstrappable ecosystem where you get paid before delivery of product.

• This is called exponential growth.

• They don't want somebody who will be on the board of directors saying they have a way of saving one percent of your costs by doing X.

Purpose of Companies

• Whenever a piece of technology is made, whether it's a company, the speaker is curious about what is the purpose of the companies, who controls them, who's on the board of directors, and what is their objective.

• Governance plays an important role in powerful companies in the world.

• Entities in the world are called companies and they have rules that follow their charter or whatever it is.

• When you start looking at this then you see that incentives don't always align with idealistic goals.

Learning Outcomes and Feedback from Students

In this section, the speaker talks about learning outcomes and feedback from students.

Feedback from Students

• The speaker traveled around the whole world and met some students who were taking his class.

• One student said that this was the only class he would wake up at 8am to do because times were different.

• There are lots of people in China and Hong Kong who want to do this so there are open dedicated times for them.

Learning Outcomes

• The goal was for students to be able to think about new ideas which seems to be happening anecdotally based on feedback received so far.

• Research on actual outcomes is still being conducted.

• There is a market for people looking for these kinds of classes and they can already see that there's an improvement.

• A third-party research agency wants to join to analyze what they're doing as the scholarship side grows.

Philanthropy and Equity Gap

In this section, the speaker talks about philanthropy and equity gap.

Philanthropy

• The speaker has two arms: one arm is purely growing this thing and helping people learn while the other arm is trying to make sure that this new thing doesn't exponentially expand the equity gap because he also works in education.

• They have spent a lot of time trying to make sure that this new thing doesn't exponentially expand the equity gap.

Equity Gap

• Public school districts are spending money to try to help students learn but there are definitely going to be some outliers in those public school systems.

• As their scholarship side grows, they have a third-party research agency which will analyze what they're doing so that it doesn't run forever until it's okay.

The Importance of Combining Domain Expertise and Performing Arts

In this section, the speaker discusses how combining domain expertise and performing arts can help students learn more effectively.

Improvisational Company Training

• High school students with improvisational company training and the support of Performing Arts experts could approximate teaching staff very quickly.

• One student was told he sounded like Professor Lowe after attending a summer camp.

Passion vs Pragmatism in Education

• Many people try to get into Ivy League schools simply because they are good schools with good rankings, not because they have a passion for the subject matter.

• People who learn skills out of pragmatism rather than passion may not have outstanding skills that are in demand in an AI era.

• Education should inspire people to have passion and go above and beyond what society expects of them.

Success Through Networking

• Success is not about getting into a famous university but about having lots of capable friends who can help you find job opportunities.

• The value of an Ivy League education is largely in its brand.

This transcript has limited content, so there are only three sections.

Building a Network of Connections

In this section, the speaker explains how he views the world as a mathematician and how he builds networks of connections to help people get jobs. He also discusses the criteria for getting into his high school program.

Building a Random Network

• The speaker builds a random network where every node has six to ten connections with other people who are good at math and have strong interpersonal skills.

• This network is not limited in size and helps connect people with others who are on their way to various positions of power.

Criteria for High School Program

• To get into the speaker's high school program, one must meet three criteria: being good at math, having strong interpersonal skills, and having some reason for doing something (passion).

• The program rejects people who only meet one or two of these criteria.

Creating Value

• The speaker encourages people to move away from an attitude of "how do I get stuff for myself" to an attitude of "how do I create value with my situation."

• He believes that adding value without worrying about what comes back will ultimately lead to more success and happiness.

The Traveling Salesperson Problem

In this section, the speaker talks about how he started giving talks after the pandemic and decided to travel across the United States of America to declare his message. He also discusses how GPT can be used by high school students to write essays and how it affects the curriculum.

Changing Curriculum

• The speaker believes that changing the curriculum is difficult at a national scale.

• Instead, he focuses on creating an out-of-band solution that is better than what is currently available.

• Researchers document these solutions, which may lead to changes in curriculum over time.

Using GPT for Essays

• High school students are using GPT to write their essays.

• Detecting such usage is difficult, and it's better to do evaluations in class or have students write essays in class.

• The speaker has talked to actual teachers who face this problem.

Teaching Math in Schools

In this section, the speaker discusses whether teaching math in schools is still relevant given advancements like GPT.

Changing Curriculum

• Changing all of the curriculum is challenging at a national scale.

• The speaker believes that as a single person, he cannot convince powerful people who control curriculum to change it.

Prompt Engineering

• Teach kids prompt engineering instead of changing the entire curriculum.

Assessing High School Students

In this section, the speaker talks about assessing high school students and how they take advantage of their lack of sophistication when lying.

Assessing Honesty

• Most high school students haven't learned to be sophisticated liars yet.

• Therefore, assessing honesty can be done by taking advantage of this fact.

Interviewing for Empathy

In this section, the speaker discusses how they interview potential teachers for their program and how they assess whether or not someone cares about people.

Assessing Empathy in Interviews

• The speaker interviews potential teachers with a couple of people to get a sense of whether or not they care about people.

• They ask questions to determine if the candidate cares about others, but not always the same question.

• The speaker emphasizes that they take quiet people too and that it's a skill to interview for empathy.

• People who are high school age may have trouble getting past this stage because they haven't thought much about caring for others.

Practicing Before Teaching

• Candidates must practice before teaching a paid class, meaning someone is paying for the class or the scholarship donor is paying.

• They must be checked out on being able to deliver what is expected before teaching a paid class.

Building a New Normal

• The program aims to build a new normal among high school math students by emphasizing empathy and kindness as important qualities in addition to academic success.

• The goal is to create an environment where being kind and empathetic is valued in addition to being good at math.

Artificial General Intelligence

In this section, the speaker briefly discusses artificial general intelligence (AGI), its potential impact on research jobs, and their thoughts on the matter.

AGI and Job Perspectives

• A member of the audience asks about AGI potentially taking over research jobs and how it might affect job perspectives.

• The speaker admits that this is not their area of expertise and that there are others who would know more about it.

• They mention that the fraction of jobs in the world that are at risk of being taken over by AGI is very small, so it's not something they usually worry about.

The Importance of Teaching Critical Thinking

In this section, the speaker discusses the importance of teaching critical thinking and how it can benefit individuals and society as a whole.

Teaching General Subjects with Critical Thinking

• The speaker believes that teaching critical thinking is more important than just teaching math or other general subjects.

• His goal is to help people learn how to think because he thinks it's good for them.

• He wants to teach sixth graders how to come up with new ideas and how they think with only 120 hours of class or 200 hours of class.

Impact on Society

• The speaker believes that it's valuable for AI researchers to consider the impacts of what they're making on the rest of the world.

• He thinks that clever people can come up with ways to mitigate or assist with these impacts.

• He also wants to unite the world by connecting people through his program.

Personal Goals

• One of the speaker's goals is for half of the population in his country to be able to solve a specific math question within 24 hours, which requires thinking outside the box.

• He believes that if half the people could solve that very first question, then we have people who can think right.