一口气搞清楚ChatGPT

Summary

视频脚本详细回顾了从图灵测试到ChatGPT的发展历程，探讨了ChatGPT的技术突破、社会影响以及对未来工作和教育的潜在影响。同时，讨论了ChatGPT与搜索引擎结合的可能性，以及微软和谷歌在AI领域的竞争态势。

Takeaways

• 📝 ChatGPT的起源可以追溯到1950年，图灵提出了著名的图灵测试，用于评估机器是否具有智能。
• 🤖 早期的聊天机器人如ELIZA和ALICE基于模式匹配技术，通过关键词触发预设的回答。
• 🧠 机器学习的兴起带来了新的语言学习模型，不再依赖于人工设置的规则和回答。
• 🌟 2001年的Smarter Child聊天机器人因其自然对话能力而广受欢迎，吸引了超过3000万用户。
• 💡 人工神经网络的突破使得机器能够模拟人脑神经元的连接，处理复杂信息。
• 🚀 Transformer模型的出现极大地提高了机器学习的速度和效率，为自然语言处理带来了革命性的变化。
• 🏢 OpenAI作为非盈利组织成立，旨在推动AI技术的发展，并将研究成果公开。
• 💸 由于AI研究需要巨额投资，OpenAI转型为有限利润公司，并获得了微软等公司的投资。
• 📈 ChatGPT通过不断迭代，从GPT-1到GPT-3，参数数量增长了数亿倍，使得其对话能力接近人类。
• 🔍 ChatGPT的训练数据只到2021年，因此它不知道近期的事件，微软将其与Bing搜索引擎结合，弥补了这一不足。
• 🌐 ChatGPT的普及可能会对搜索引擎市场产生重大影响，特别是对谷歌的搜索业务构成挑战。

Q & A

ChatGPT是如何帮助用户提高视频脚本写作效率的？

ChatGPT可以根据用户提供的大纲自动生成详细的脚本，用户无需从头开始撰写，节省了大量的时间和精力。

Alan Turing提出的“图灵测试”是如何定义机器智能的？

图灵测试认为，如果在非面对面的文本对话中，人们无法准确判断对方是人类还是机器人，那么在一定程度上，这台机器就具有智能。

早期的聊天机器人Eliza是如何工作的？

Eliza通过预设的模式匹配规则和关键词触发来回答用户，它模拟了一个心理治疗师的角色，通过尽量少说话来减少错误，给人一种在与人交流的错觉。

ChatGPT与以往的聊天机器人有何本质区别？

ChatGPT基于深度学习和神经网络，能够处理更复杂的对话和问题，生成更自然、更准确的回答，而不是简单地基于模式匹配。

OpenAI是如何推动ChatGPT技术发展的？

OpenAI作为非盈利组织，致力于推动AI技术的发展，通过公开研究和专利，吸引了大量投资和人才，最终推出了先进的ChatGPT模型。

ChatGPT在训练过程中是如何优化对话质量的？

ChatGPT在训练中加入了人类反馈机制，通过人工与AI的互动，不断调整和优化模型，使其对话更加流畅和贴近人类的交流方式。

ChatGPT在技术发展史上的突破性成就是什么？

ChatGPT通过深度学习和神经网络，极大地提高了机器理解和生成自然语言的能力，使得机器能够更准确、更自然地与人类进行交流。

ChatGPT的普及对社会和经济有哪些潜在影响？

ChatGPT可能会改变许多行业的工作方式，提高效率，但也可能导致某些重复性和常规性工作的失业。同时，它也可能引发对教育、版权等社会制度的重新思考。

Google在AI领域的应对策略是什么？

Google虽然在AI领域有强大的技术积累，如Transformer和LaMDA，但因为担心影响其搜索引擎业务，所以对推出类似ChatGPT的产品持谨慎态度。

ChatGPT在教育领域的应用现状如何？

ChatGPT在教育领域应用迅速普及，但同时也带来了挑战，如学生可能过度依赖它完成作业，而现有的教育系统还未准备好如何应对这一变化。

如何理解ChatGPT在未来发展中的不确定性？

尽管ChatGPT已经展现出强大的能力，但它的发展仍然充满不确定性，可能会在某个领域突然出现突破，也可能带来预料之外的社会变革。

Outlines

00:00

🤖 聊天机器人的起源与发展

本段介绍了聊天机器人的历史和发展。从1950年图灵测试的提出，到1966年MIT实验室发明的Eliza，再到1995年的ALICE，聊天机器人经历了从简单命令到模式匹配的演变。随着机器学习和神经网络的发展，尤其是2017年Google提出的Transformer框架，聊天机器人的能力得到了显著提升。OpenAI的成立和ChatGPT的诞生标志着聊天机器人技术的一个新时代，ChatGPT通过大量数据训练和人类反馈机制，实现了更自然和流畅的对话能力。

05:01

🧠 人工智能的神经网络革命

这一部分讲述了人工神经网络的起源和重要性。人工神经网络模拟人脑神经元的连接方式，通过隐藏的神经节点进行信息判断和输出。虽然神经网络的概念早在1960年代就已存在，但直到2010年代互联网时代，数据和计算能力的充足才使得神经网络得以实际应用。特别是Transformer框架的提出，极大地提高了机器学习的速度和效率，为自然语言处理模型的发展奠定了基础。

10:01

💡 ChatGPT的诞生与技术突破

本段描述了ChatGPT的诞生过程和技术突破。OpenAI作为一个非盈利组织，旨在推动技术发展，其研究成果和专利均公开。ChatGPT通过大量数据训练，无需人工监督，自主学习。随着参数数量的增加，ChatGPT的能力得到了显著提升，尤其是GPT-3模型，参数量达到了1750亿，使得ChatGPT能够处理各种领域的问题。然而，ChatGPT在训练过程中缺乏有效的反馈机制，导致其有时回答可能出现偏差。为此，OpenAI引入了人类反馈的强化学习机制，进一步提升了ChatGPT的训练效果和对话质量。

15:02

🌐 ChatGPT对搜索引擎的影响

这一部分探讨了ChatGPT对搜索引擎可能带来的影响。ChatGPT的出现可能会改变人们获取信息的方式，从而影响到搜索引擎的市场地位。特别是当ChatGPT与搜索引擎结合时，用户可以直接通过对话获得所需信息，减少了对传统搜索引擎的依赖。微软和谷歌在这方面的竞争尤为激烈，微软通过投资OpenAI并将其技术整合到Bing搜索引擎中，而谷歌则急忙推出了基于LaMDA的对话AI Bard。这一变革可能会对搜索引擎市场格局产生重大影响。

20:03

🚀 生成式AI的快速发展与挑战

本段讨论了生成式AI的快速发展及其对社会的挑战。生成式AI技术，包括聊天、编程和绘画等，正在迅速发展，吸引了大量资本投入。这种技术的快速发展可能会导致某些工作岗位的消失，尤其是那些重复性和常规性的工作。同时，ChatGPT等AI工具的出现也对教育系统产生了影响，许多学生开始使用这些工具来帮助完成作业。社会需要找到方法来整合这些AI工具，同时解决由此产生的版权等问题。

Keywords

💡ChatGPT

ChatGPT是一个由OpenAI开发的人工智能聊天机器人，它能够理解和生成自然语言文本，用于模拟人类的对话。在视频中，ChatGPT被描述为能够写作、编程、搜索信息，并在多个领域提供帮助，它的出现颠覆了人们对聊天机器人的传统认知。

💡Turing test

图灵测试是由艾伦·图灵提出的一个哲学测试，用于判断机器是否具有智能。测试的核心是，如果一个人在与机器进行文本对话时无法区分对方是人还是机器，那么机器在一定程度上被认为是智能的。

💡机器学习

机器学习是人工智能的一个分支，它使计算机系统能够通过经验自我改进，而无需进行明确的编程。它依赖于算法和统计模型，通过分析和识别数据中的模式来进行预测或决策。

💡神经网络

神经网络是一种模仿人脑神经元结构的计算模型，用于识别复杂的模式和关系。它由大量的节点（类似于神经元）组成，这些节点通过加权连接相互传递信息，并通过学习调整权重，以改善其预测能力。

💡Transformer

Transformer是一种深度学习模型架构，它在自然语言处理（NLP）领域取得了重大突破。它通过自注意力机制（Self-Attention）允许模型同时处理整个输入序列，而不是像传统的循环神经网络（RNN）那样逐个单词处理，大大提高了处理长文本的能力。

💡OpenAI

OpenAI是一个人工智能研究实验室，它致力于确保人工智能（AI）的发展能够对全人类产生积极影响。OpenAI是ChatGPT的开发者，它的目标是推动AI技术的进步，并将其研究成果公开，以促进整个领域的发展。

💡生成式AI

生成式AI是指能够自主生成新内容的人工智能技术，如文本、图像、音乐等。这种AI通过学习大量数据，掌握创造内容的模式和规则，从而能够创造出原始的、逼真的作品。

💡失业

失业是指个人在劳动力市场中没有工作的状态，可能是因为技术进步导致某些职位被机器取代，或者是经济环境变化导致的就业机会减少。

💡教育

教育是指系统地传授知识、技能和价值观的过程。它包括学校教育、家庭教育、职业培训等多种形式。在视频中，提到ChatGPT在教育领域的应用，如帮助学生完成作业，但同时也引发了对现有教育体系是否适应AI技术的讨论。

💡版权

版权是指对文学、艺术和科学作品的原创性表达形式的法律保护。版权赋予作者或创作者对其作品的复制、分发、展示和表演等方面的独占权。

Highlights

ChatGPT的出现在资本市场上引起了轰动。

ChatGPT能够写作、编码并查找信息，显示出强大的写作能力。

从历史上看，聊天机器人的发展可以追溯到1950年图灵测试的提出。

早期的聊天机器人如Eliza和ALICE基于模式匹配进行工作。

机器学习的兴起标志着一种新的学习方式，机器通过大量示例学习而非依赖预设规则。

人工神经网络的灵感来自人脑，通过模拟神经元连接处理信息。

Transformer学习框架的提出是自然语言处理领域的重大突破。

ChatGPT的开发公司OpenAI是由Elon Musk和Peter Thiel等人成立的非盈利组织。

ChatGPT通过增加人类反馈机制来优化训练效果。

ChatGPT能够流畅地进行跨领域对话，但其并不完全理解所说的话。

ChatGPT的训练数据只到2021年，它不知道近期的事件。

ChatGPT的运营成本高昂，每天需要花费约1000万美元。

ChatGPT的快速崛起可能会颠覆人们对聊天机器人的认知。

ChatGPT的出现可能会使某些职业的常规工作变得不再必要。

ChatGPT在教育领域的应用引发了关于其对现有教育系统影响的讨论。

ChatGPT的版权问题，即AI创作的作品的版权归属，是一个待解决的问题。

ChatGPT的发展可能会对社会秩序产生重大影响，需要新的系统来适应。

ChatGPT的未来发展充满不确定性，可能会在某一领域突然实现突破。

Transcripts

is what you all want me to talk about

I’m getting all kind of private messages

so let’s talk about it today

To save myself the trouble

I asked ChatGPT

if it could write me an outline for a video

Here you go

1234567, all listed out

Can you list in detail

Can you list in detail

Can you list in detail

Can you write a script for me?

I don’t even have to write a draft

You certainly can’t delve into the manuscript

If I follow the script

If I follow the script

However

If I follow the script

just look at its ability to write

just look at its ability to write

just look at its ability to write

and decently

I was shocked

American medical license

bar exam

with the ability to write novel, code and look up information

It’s like

anything that can be conveyed in words

It can do it

This thing

how did it suddenly appear out of nowhere?

Before this, there were also chatbots

but why is this one turning the world upside down

and excites

the capital market

And what are its problems

and how are those tech giants counter it

Who will be put out of work because of it

Although I’m not

Although I’m not

but today, let’s

put all the pieces together

and talk about

ChatGPT

All that you need to know about

Talking about chat bot

we’ll have to go back in time to 1950

Known as the Father of Computer Science

Father of Artificial Intelligence, Alan Turing

published an

epoch-making paper

He came up with a very philosophical

imitation game

The famous

Turing test

It means that if you are not in a face-to-face

text conversation

can you accurately determine whether the person you are talking to

is an actual person

or a robot

If it’s hard to tell

then to a certain extent

the machine is intelligent

Turing exam

is easy and simple to understand

and quite interesting

Hence it attracted

Hence it attracted

to attack it

In the beginning

are very simple commands

It uses some language technique

and trick

to make you feel that

you are talking to a person

For example in 1966

in MIT lab

they invented a chat bot

called Eliza

The developer is very clever

He set Eliza up as a psychotherapist

You see for these kind of therapist

normally they listen more and talk less

So it can ask

Do you have any thought?

And people can reply

and it ask again

How was your rest yesterday

and people reply

The less it talks, the less mistake it makes

It really makes people believed that

it’s listening and communicating with you

In fact, behind it is some

very simple code of if….

then….

For example if it sees the word

“mother”

It’ll tell you

Tell me about your family

Keywords like this

are about 200 words

30 years later

in 1995

Eliza came out with a junior named ALICE

It has evolved to be very powerful

Although still incomparable to ChatGPT

but it can handle

everyday conversation

But in essential

Regardless it’s Alice or Eliza

The principle

is based on Pattern Matching

Pattern Matching

When it sees a keyword

it’ll pick up one

pre-planned answer

For example if it hears Hi How are you

Have you eaten

If it hears Mother

it’ll reply tell me about your family

Something like this

In fact, even now

on some e-commerce site

or banking site’s chat bot

they are still

based on this model

If you are chatting with it

and mention refund

it’ll send you the procedure for refund

or if you say ATM

it’ll send you the map of nearest ATM

This pattern matching

although not very intelligent

did reduce a lot of that

mechanical repetitive answer from human

From the perspective of intelligence

These rule-based robot

no matter how complicated the rules are

or more preplanned answers

there won’t be infinite answer

nor can it create new answer

So

if you are trying to use the Turing test

to become real intelligent

it’s impossible to realise

with pattern matching

And so a new school of

language learning emerged.

This is also the most important part

in artificial intelligent

Machine learning

As the name implies, the basic principle

is to let the machine learn

Meaning, I wont be setting some

rules and answers

I’ll just dump lots of ready-made example

for you to learn and find the pattern

Sounds more impressive now right

it also complies to

our understanding on the logic of learning

Based on this principle

In 2001

There is Smarter Child

Smarter Child

That's when the robot went viral

Why did it go viral?

First, it used some of the more advanced models

of machine learning at the time

to make conversation more natural

Moreover in 2000

A large number of chat apps had sprung up

like AOL Windows Yahoo

So Smart Child

swept up all these chatting platforms

and let billions of people all over the world

to have a conversation with it

No matter what you ask

doesn’t matter the quality of answer

it’ll chat with you at least for a sentence or two

It can be said as the predecessor of ChatGPT

Something fun like this

immediately became popular all over the world

Attracted more than 30 million users

to have conversation with it

It receives more than 1 billion pieces of

information every day

information every day

it was bought by a giant company

Guess who

Microsoft

Microsoft has been coveting

this sector since that early

Although this Smarter Child

is already good in chatting

but it is still far from

passing the Turing test

In two sentences you’ll know

that’s a machine

Let's keep making progress

In 2010

There’s one area of machine learning

is starting to shine

Artificial neural network

Artificial neural network

Our brain

depends on

more than 10 billion neurons

through network connection

to judge and convey information

Although each of these neurons is very simple

but when combine

can judge very complex information

So this artificial neural network

is to simulate the model

of human brain

After entering information

It will go through the judgment of

several hidden neural nodes

like neuron

and give you an output

In fact, the idea of this neural network

has long been existed

We can trace it

back to 1960s

But it needs two things for support

A large amount of data and powerful computing power

which were not available before

So this neural network thing

is only a talk on paper

Later in the 2010s

The era of internet

Data is certainly available

the computing power

continues to improve exponentially

This is what makes neural networks

finally working

People realised that

This mode is really good for solving

those that people know by just looking

intuitive thing

For example when you look at a face

you can immediately know who the face belongs to

Except for Liu Qiangdong

I’m face blind

I don’t know if she is pretty or not

Before this, it’s very difficult

for computer to figure out

who the face belongs to

But with neural link

Machine learning can slowly figure out the pattern

It is now widely used

not just face recognition

voice recognition, automated driving

including few years ago

AlphaGo that beats professional Go player Ke Jie

are learning in this way

So this neural network

can make great achievements in those fields

we mentioned just now.

But back in the realm of writing

it didn’t go so well

it didn’t go so well

Because usually machine learning

Because usually machine learning

Recurrent Neural Network

RNN to process text

The way it works is

to look at word by word in order

then process it word by word

The problem with that is

It can't do a lot of learning at the same time

and the sentence can’t be too long either

Otherwise, when it learns the latter

it forgets the former

Until 2017

Google published a paper

and proposed a new learning framework

called Transformer

The exact mechanism is more complicated

It’s not something I can figure out

but the result is that it can let the machine

but the result is that it can let the machine

Before this you have to learn word by word

like a series circuit

Now you can learn at the same time

like parallel circuit

In this way, the speed and efficiency of training

has greatly improved

With the Transformer

the machine can now learn words

very easily

Many of today's natural language processing models

are actually built on

its infrastructure

The T in Google’s BERT

including the T in ChatGPT

including the T in ChatGPT

Alright now

there have been very strong breakthroughs

in technology

Everything is ready

all that’s needed now is people and money

It's time for ChatGPT to debut

In 2015

few tech giants like

Elon Musk, Peter Thiel

Elon Musk, Peter Thiel

Founded a non-profit organization called OpenAI

the parent company of ChatGPT

to conduct research on AI

Its an non profit organisation

not for earning money

Simply for the sake of

pushing the technology forward

Because of this, the research including patents

are made public

Look at the investor list

we can hear

the familiar name, Elon Musk

he gradually discovered that

his Tesla also needed to invest a lot of

research in AI

for automated driving

In order to avoid

conflict of interest between Tesla and OpenAI

In 2018

3 years after OpenAI was founded

he stepped down from the board

So now OpenAI

actually has no relation to Musk

Bye~

The OpenAI guys

are indeed incredible

In 2017

Google introduced Transformer

and they quickly conduct research and learning

based on this foundation

and published a paper in 2018

to introduce a new language learning model

Previous models of language learning

basically required human supervision

artificially set some labels for it.

but for GPT

it doesn’t need all that

You just need to put in data

and it’ll learn till it gets

That's about it

In June 2018, OpenAI

introduced 1st Gen of GPT

In November 2019

they increased the amount of training data

and introduced GPT-2

Actually for machine learning

they require two things

One is model, another is parameter

The model determine how the machine learns

With the same data

I can learn faster and better than anyone

then you’re great

As for parameter

It needs large volume of computation

To put it bluntly, you need to dump in lots of money

No matter how good the model is

you still need to put in lots of money to train and verify

You cannot have one without the other

OpenAI team

is very confidence with the model

but the next step requires money

For every single step forward

it needs

another level of magnitude of data to support

and all these

need money to support

For example DeepMind by Google

the company that came out with AlphaGo

their annual expenditure goes up to 400 or 500 million dollars

In the beginning at OpenAI

they received $1 billion investment

but it’s not enough

but it’s not enough

at this time it is still a non-profit organisation

When Musk stepped down

and the $1 billion sentiment is no longer sufficient

where am I going to find more people with the same sentiment

due to capital pressure

In 2019 OpenAI transformed from

non-profit organisation

but it didn’t change completely to profit-making organisation

they still needed the sentiment

and transformed the organisation to

Capped-profit company

What does it mean?

What does it mean?

cannot exceed 100 times

Once exceed 100, the amount beyond that

will not be retrievable by investors

and will belong to OpenAI

I’m curious

If my

investment return is close to 100

I'd get the money out and invest again

then won’t I get the 100?

Regardless,

OpenAI became capped-profit company

it means if you invest in it

you can get a return

Here comes Microsoft

with investment of $1 billion

Then the investment must be

a win-win for both sides

At the OpenAI side, first they got the money

second, Microsoft built them

the world’s fifth super computer

which greatly improved its training efficiency

Meanwhile Microsoft

also obtained OpenAI’s team and technology

And of course

the research from OpenAI

will no longer be public

Microsoft is definitely not investing on sentiment

Once OpenAI got the support of super computation

they were starting to prepare on a miracle

The first generation

has only 120 million parameter

GPT-2 has 1.5 billion parameter

6 months later

they came up with GPT-3

and the parameter rose by 100 times

became 175 billion

The effect was really good

so close

to the current ChatGPT

Just ask whatever

and it’ll give you answer

At that time, there was already

a wave of sensation in the industry.

However this pure machine trained GPT-3

has a problem

and that is sometimes it gives really good answer

but sometimes it’s a little bit off

Another problem is that

no matter how much you increase the parameter

the improvement made is very limited

This is because during training

it didn’t have a very good respond mechanism

Meaning, there’s no one to tell it

which answer is correct

or which kind of answer is not good

For example if I'm playing chest

I want to win right

Winning is good

so I train myself to win

But for chatting

it’s hard to make the judgement

How do I know if the answer is good

or not

Can only learn

So in order to solve this problem

During training, OpenAI

added human feedback mechanism

You can chat with me and I’ll tell you

if you are doing good or bad

The professional term is

Reinforcement Learning from Human Feedback

That’s why when using ChatGPT

you can feel that

it can be very lean and talkative

This is because the people training it

likes it that way

If the person training it

is very humorous

then the ChatGPT

would probably be telling you jokes all the time

After adding

Reinforcement Learning from Human Feedback

It has greatly improved both the

efficiency and the effect of training.

In March 2022

the introduced GPT-3.5

where the conversation was optimised

In November 2022, they introduced

It's actually a very, very simple

chat interface

No matter what you ask

it could give you

all the answer

that sound reasonable

Of course there’ll be some problems

we’ll talk about it later

But if you look at it roughly

It really could talk about anything

And the language expression

is really like talking

After half a century

This time ChatGPT

This time ChatGPT

Turing test easily

Impressive right

This reminds me of

Fu Tu Niu Niu Overseas version

More than 70% of Futu's employees

are engaged in product and R&D

relying on technological innovation

to make investing easier

You can invest around the world

with just one account

Moomoo prepared exclusive benefit

for Lin’s subscribers

You’ll get one Under Armour stock upon opening an account

If you deposit an equivalent of HKD 1Ok

you’ll get

Google stock worth $100

Lately ChatGPT is going viral

If you’d like to see

which concept stocks is getting viral because of ChatGPT

you just need to search ChatGPT

then you can see

US, Hong Kong stocks

Not only famous tech giants

like Microsoft and Google

you’ll also find

some unheard

potential stocks

For example

if I want to know which stock has potential

like TSMC

you can see the rating from Wall Street analyst

Target Price Forecast

And which stocks give

positive or negative signaks

Moomoo has it all

All those people are usually concern about

like Financial aspect, Technical aspect, Fundamental aspect

are all available

Not only they sort out

these information for free

the graphic visualisation

is quite intuitive too

including real-time updates

on global AI news

are even translated for you

Apart from ChatGPT

they have a Concept Segment

where you can see other concept stocks

like robotic science,IOT

Even if you’re not into buying stocks

it’s good to get an understanding

They also have millisecond quotes

and support 0.0037 seconds

and support 0.0037 seconds

So you can really see

they are making miracle in technology

Recently in Japan, Moomoo open up

functional experience of the platform

The users

If you are interested

click on the link down below

and experience it for yourself

Alright, let’s get back to ChatGPT

Anyway it

has subverted

most people's perception of chatbots

including me

So in just two short months

ChatGPT's monthly active users exceeded 100 million

The rate of expansion must be the fastest in history

The rate of expansion must be the fastest in history

But honestly

with ChatGPT being so subversive

The shock that the product itself

brings to people

has far surpassed those data

Until now

when I look at its answer

It didn’t all come out

in one go

it came out bit by bit

just like how a person is talking to you

Sometimes it really does give me goosebumps

But I guess a year from now

It shouldn't be surprising

for everyone to see this again.

Alright now let’s see

how does ChatGPT able to

chat on questions

of any fields

To put it simply

A large language model like GPT

it essentially calculates the

next word, next sentence

what should appear next

it’s a matter of probability

For example when it says I’m very

and the continuation to that

so many words in the database

it could be I’m very happy, I’m very healthy

I’m very anxious, I’m very hungry, etc

but you need to have a context

For example if the text above says the weather is nice today

then it could compute

that it’s I’m very happy

Actually every answers, every words

are just simple

It's calculated based on the correlation of previous text

when it learns enough

Hundreds of billions of parameters and words

After finding patterns through these complex models

It forms a

very large neural network

You don't need to tell it

What is programming and what is video scripting

It’ll know from learning more and more

It’ll know from learning more and more

This is what a video script should look like

So I asked it to write one

ChatGPT video script for me

From the conclusion of the correlation

it gives out answer word by word

It is still a language model

imitating how human talks

But does it know

the meaning of what it says?

At least the current ChatGPT version

doesn’t completely understand

It's like a kid who has a really good memory

but doesn't really know anything

and imitating adult

But make us think

it knows everything

This is why

This is why

it’s almost close to perfection

very much like human

But there are often

some logical mistakes

for us it’s like stupid mistake

of addition, subtraction, multiplication and division

This is because

it is actually a language model

For now

actually

GPT also often

has a lot of fabricated answers

Meaning to say

it doesn’t know what it’s talking about

but it was just not making sense

Including a lot of moral and ethical problems

For example if you ask what does it think about human

It’ll say

Human beings are inferior and selfish

It's the worst kind of creature

and should be wiped out

Then it must not know

what it is talking about

don't know where it learned this from

but all these nonsense problems

are all problems with the current version of ChatGPT

Although now

it’s just simply imitating

But as it get better and better

at imitating

and that in 99.99% of the cases

it can answer correctly

So whether it really understands

or just imitating

it doesn't really matter much

This was a question which

Alan Turing had already discussed in his paper

on the Turing Test

Rather than us asking

Can machines think like humans

might as well ask

Can machines do what humans do

It’s deep

Actually, I think

One of ChatGPT's major breakthroughs was

to greatly improve the efficiency of communication

between humans and machines

Human beings communicate information

primarily with words

The computer uses code

Humans have always accommodated computers

You’ll have to learn programming first

And figure out

program it to a

language that computer could understand

and let it execute

including search

We also change our questions

into few keywords

and then search

It changes

Computers can slowly understand people now

I can talk to it directly

then it’ll translate itself

and execute

Everyone thought ChatGPT was amazing

It knows everything that you ask

But the amazing thing about it

isnt’t that it

can do these tasks

It is mainly

it could accurately understand your question

And then contextualize

from its vast database

and come out with the most appropriate information

and tell you in human words

This communication link

is actually the most amazing part

It has such a powerful interface

Then we can more easily

hand over a lot of things

to the machine

Wouldn't that make things

much more efficient

Can you imagine

if we connect it

to a speech recognition system

like Siri

and let it talk to you freely

and if you could connect it to

professional analysis interface

like analysing AI stocks

and connect it to

programming and computing machine

as well as visual generation

then everyone of us

could be like in the movie

Iron Man and his assistant

For example if you ask it to compute

Mobius ring

and it’ll start computing

and then you say

Superb

You see how ChatGPT

opens up so many possibilities at once

and it is the hottest thing in the market now

the major shareholder behind it, Microsoft

must be really happy

So they started to invest more money

and in January they announced

to invest $10 billion

It was valued at $29 billion

This time the deal

between Microsoft and OpenAI

is quite interesting

For Microsoft

after they invested $10 billion

the revenue that OpenAI received

they have to give Microsoft 75%

until they get the $10 billion back

Meaning to say Microsoft is making sure

that the money they invested will get a return

Also, Microsoft hold

49% of OpenAI stake

And there’s a

100 fold upper limit of return on investment

A rather peculiar deal

When this deal is done

on February 7th

Microsoft held a press conference

They announced to incorporate ChatGPT into

their own search engine Bing

Microsoft called it “Copilot for the Web”

it’s like a web assistant

Actually there’s another problem with ChatGPT

the problem is that the training data

is only up till 2021

Meaning to say

it doesn’t know the recent events

When Microsoft combines it with Bing

the logical side they use ChatGPT

while news and information

can be searched with Bing

Isn’t it a strong alliance

ChatGPT For example if I ask ChatGPT

do you know Lin's channel?

it would say no

If I ask Bing

it’ll say Lin's channel

is a fun and useful

content creator

It's a good example for many people

who want to pursue their dreams

I'm a little embarrassed by that

So it’s viral for a reason

And Microsoft is sneaky

the chat function

can only be used on their own

Edge web browser

I have to say

their marketing

I give full mark

So in the face of all this publicity

the most anxious is Google

Why?

Because ChatGPT is likely to shake up

their biggest piece of the pie

Search engine

Imagine if I ask ChatGPT

and it can organize the language to tell me

So when I want to search for something

I don't have to go through

them myself

I can just ask ChatGPT

then no one will use search engine anymore

Can Google not panic

It now occupies 93% of the

93% global search engine market

That's a solid monopoly

Although Microsoft’s Bing is in second place

but it’s only 3%

Advertising revenue brought by the search business

can account for 60% of Google's total revenue

everyone was doing fine

and suddenly there’s GPT

Actually

Google has been leading

in AI sector

The Transformer

is created by Google

They have actually been testing

a robot named BERT

It’s similar to ChatGPT

but they didn’t spend a lot of energy

to train it

They also have another robot

which is more impressive, called LaMDA

It's based entirely on normal human conversation

So it can even make jokes

Or express emotions

It’s not at all like you ask

and it answers

Because it does speak so naturally

It even fooled

a test developer who was

working inside Google

I believe that LaMDA already has consciousness

Almost like a seven or eight year old

So

Google has actually

been very strong on chatbots

But its position is

quite different from Microsoft's

Google

is already the top in search engine sector

then they had to build a robot

and cut down their cash cow

Surely not unless it's a last resort

So that’s why

I think

the LaMDA

is more focus on conversation and chat

and not like ChatGPT

can answer any question

And they haven’t released these AI robots

is also because

they are worried about their reputation

After all their focus is on search engine

which needs to be strict and accurate

If they introduced

an untrained

speak nonsense robot

then that’s outrageous

On the other hand, training on such a large scale

requires a lot of computing power and burns money

Each question consumes roughly

10 to 100 times as much energy

as a Google search today

For example ChatGPT

now has to spent

$10 million per day to operate

So you can see

Microsoft's first-mover advantage

is indeed very reasonable

Not only they invested in the right company

but they also

have the ruthless hand to spend all these money

In the face of strong public opinion pressure from Microsoft

Coupled with overwhelming media coverage

Google can’t hold on longer

Just after ChatGPT was launched

Google initiated

Code Red

This is the moment

of our life or death

We need to focus the entire company

on the AI circuit

Because the key to this thing is that you have to be fast

How fast?

So fast that Google was about to twist its ankle

We mentioned earlier that Microsoft’s press conference

was on February 7th

where they announced to integrate ChatGPT

into their search engine

Google in a hurry

organised a press conference on February 8th

and introduced a conversational AI called Bard

This is developed based on their

Chatbot LaMDA

Just look at the stock prices of

Microsoft and Google after Google's announcement

You’d know how bad it is

for Google

You can't blame anyone in this business

You don't have to read any professional analysis

You just need to stay calm

watch their press conference

from beginning till the end

then you’ll know why

Everyone knows that

people are focusing on

AI chat

But during Google's 40 minutes press conference

they talked about their previous achievements

and later picture search

In the middle of this, the speaker

couldn't find the mobile phone for presentation

So have to skip this part

Later, when they finally got to the point

and begin to introduce Bard

they only talked for a few minutes

During the press conference

they also showed a video

introducing Bard

The terrible thing is that there are

factual mistakes in Bard's answer in the video

Actually to be honest

everyone can understand

if this type of chatbot

makes some factual error

However

the answer in the commercial video is incorrect

and even forgot to bring mobile phone

it was nothing but great cry and little wool

It’s obvious that

Google was doing it hastily

this is what the market is worrying about

Although ChatGPT is great

but everyone knows that

Google is the power house in AI sector

So even though you didn’t

make much noise

the outsiders would know you not to be messed with

and that you are holding back something big

They initiated the red alert

because it is to let

outsiders know that they take this matter seriously

Don’t sell out your stock first

Before the press conference

Google’s stock price is not worse than Microsoft

But they had to hastily

do something that

make a fool of themselves

So Google's market value evaporated

$100 billion

But in comparison, Microsoft is much more stable

Microsoft’s CEO

OpenAI’s CEO all came out

and personally explain

The nearly one-hour press conference

focused on the AI chat function

Plus various demos

Obviously well prepared

The AI war just started

Google was first caught off guard

by ChatGPT

And then out of panic

they did stupid mistake

The first battle was a disastrous defeat.

However this is only the first battle

Google is still Google after all

What’s gonna happen next

we’ll just wait and see

Of course this AI war

is not limited to these two companies

Meta, Baidu, Tencent, Ali

are all fighting to get in

Stocks that had anything to do with generative AI

started to soar

Nvidia, AMD

hardware manufacturer that provides computing power foundation

profited from this

Actually for AI chatting, AI painting

AI programming

all these generative AI

have been under development spurt

since two years ago

The amount raised over the past few years

From 2021, 2022

It's already taking off

It's over a billion dollars a year

As the year 2023 begins

Microsoft started with $10 billion

Capital has done all it can

to get into this track

This thing is developing so fast

will it cause many people to lose their jobs?

Who will lose their jobs?

Will it cost you your job?

technological innovation

It's always a double edged sword

it may create more jobs

Unemployment rate does not necessarily fall

The overall GDP will probably rise

But in the short term

It will certainly cause some people to lose their jobs

I was thinking how do you think we could

try to not lose our job

and even use this AI tool

to increase our productivity

My personal opinion

is that we have to avoid

repetitive routine work

When computers first came out

It might solve some

repetitive human tasks

Everyday doing the

same thing over and over again

You can fix it with a computer for loop

But now

it’s not just the repetitive works

Even routine work

as long as you have routine

even though every day you think you are creating content

but in actuality it doesn’t require much brain power

then this kind of task the computer

can quickly get it in the matter of minutes

What is routine work?

Let me give you an example

For example I let ChatGPT

write a fairy tales on Xiao Lin

It’ll come up with Xiao lin has a cat that can speak

it defeated dragon

saved the princess and became a hero

And I tell it, no it’s wrong

Xiao Lin is a woman, rewrite it

It’ll say Xiao lin is a woman

who has a cat that can speak

It defeated bad witch and became a hero

You see

this is the routine of a fairy tale

There’s an animal that can speak

defeated something and became a hero

Although the speaking cat

is basically useless in the story

but it is the standard of fairy tales

Similarly

There are some exceptionally skilled engineers

who can write codes

with their eyes shut

Writers who could

write 20 chapters of online novel in a day

Or some very basic

financial report of the company

basic design

basic legal advice, etc

For these tasks

once you get familiar you can do it with eyes shut

Because it has a routine

So now AI is learning these routine

so you don’t even have to do it anymore

AI will do it all

Attention, I’m not saying that

programmer, accountant or writer

or analyst will be replaced

It’s just that the routine part of

their work can be

easily learned by machine

So if you think there are some

routine tasks in your job

you need to be careful

Or at least

don’t put these routine work online

Else AI will learn it

Actually not just on unemployment

Because it is so

subversive

we can already see

the huge impact

on society

For example in education sector

it's only been online for a few months

Among the students

over the age of 18 in the United States

90% of them have used ChatGPT to help them with homework

Apart from sports

can it really help in any subject

How would I know

if you are doing the homework yourself

Of course it doesn't mean that

we can't use it to help

It's just that our current education system

is not ready for ChatGPT

It's as if we've spent hundreds of years

building a better

transportation system

but suddenly one day

all the cars could fly

From technical view point, flying car

is good in long-term

but right now when we don’t have a

complete new system

while everyone is flying their car

then it’ll be chaotic

The social order

would be greatly disrupted

So for companies and schools

they haven’t figure out how to integrate ChatGPT

into their existing system

For now they can only

ban it first

The content that AI created

the painting it created

who will own the copyright

These are really tough questions

So this generative AI

no one can say for sure

how it will develop in the future.

The ChatGPT team

did not have any special purpose

at the very beginning

Just put the data in

and let the machine learns

It was later that they discovered

how powerful it is

and could even connect to search engine

Everyone is basically tapping in the dark

You don’t know when one day

the AI is suddenly become enlightened in one field

Sometimes I feel that

It is actually quite exciting

to witness such a miraculous development of AI.

Pandora's box is

being opened bit by bit