Last active
July 23, 2023 03:01
-
-
Save JimLiu/dc415d30b9ce2174df99696f9e970a17 to your computer and use it in GitHub Desktop.
The Amazing AI Super Tutor for Students and Teachers | Sal Khan | TED
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
1 | |
00:00:04,543 --> 00:00:08,421 | |
所以任何在过去几个月一直关注的人 | |
So anyone who's been paying attention for the last few months | |
2 | |
00:00:08,462 --> 00:00:10,547 | |
都会看到这样的标题, | |
has been seeing headlines like this, | |
3 | |
00:00:10,589 --> 00:00:12,674 | |
尤其是在教育领域。 | |
especially in education. | |
4 | |
00:00:12,717 --> 00:00:14,636 | |
论题一直是: | |
The thesis has been: | |
5 | |
00:00:14,678 --> 00:00:18,472 | |
学生将使用ChatGPT和其他形式的AI | |
students are going to be using ChatGPT and other forms of AI | |
6 | |
00:00:18,515 --> 00:00:20,015 | |
来作弊,完成作业。 | |
to cheat, do their assignments. | |
7 | |
00:00:20,015 --> 00:00:21,349 | |
他们不会学到东西。 | |
They’re not going to learn. | |
8 | |
00:00:21,393 --> 00:00:24,937 | |
这将彻底破坏我们所知的教育。 | |
And it’s going to completely undermine education as we know it. | |
9 | |
00:00:25,437 --> 00:00:27,064 | |
现在,我今天要争论的是, | |
Now, what I'm going to argue today | |
10 | |
00:00:27,106 --> 00:00:29,816 | |
不仅有办法可以缓解所有这些问题, | |
is not only are there ways to mitigate all of that, | |
11 | |
00:00:29,859 --> 00:00:32,444 | |
如果我们设置正确的护栏,做正确的事情, | |
if we put the right guardrails, we do the right things, | |
12 | |
00:00:32,487 --> 00:00:33,612 | |
我们可以减轻它的影响。 | |
we can mitigate it. | |
13 | |
00:00:33,655 --> 00:00:35,990 | |
但我认为我们正处于利用AI的边缘 | |
But I think we're at the cusp of using AI | |
14 | |
00:00:35,990 --> 00:00:40,034 | |
可能是教育有史以来最大的积极变革 | |
for probably the biggest positive transformation | |
15 | |
00:00:40,036 --> 00:00:42,496 | |
我们将如何实现这一点 | |
that education has ever seen. | |
16 | |
00:00:42,831 --> 00:00:44,623 | |
是通过给地球上的每个学生一个智能的,但惊人的个人导师。 | |
And the way we're going to do that | |
17 | |
00:00:44,624 --> 00:00:47,502 | |
我们将给地球上的每个老师一个惊人的, | |
is by giving every student on the planet | |
18 | |
00:00:47,543 --> 00:00:51,171 | |
人工智能教学助手。 | |
an artificially intelligent but amazing personal tutor. | |
19 | |
00:00:51,171 --> 00:00:54,758 | |
我们将为全球的每位教师提供一个令人惊叹的, | |
And we're going to give every teacher on the planet an amazing, | |
20 | |
00:00:54,801 --> 00:00:57,386 | |
人工智能教学助手。 | |
artificially intelligent teaching assistant. | |
21 | |
00:00:57,762 --> 00:01:01,349 | |
而且为了让我们明白这有多大的事情 | |
And just to appreciate how big of a deal it would be | |
22 | |
00:01:01,348 --> 00:01:04,017 | |
要给每个人一个私人家教, | |
to give everyone a personal tutor, | |
23 | |
00:01:04,060 --> 00:01:07,063 | |
来看一下这段视频 | |
I show you this clip | |
24 | |
00:01:07,105 --> 00:01:10,108 | |
来自本杰明·布鲁姆1984年的二阶跃升研究, | |
from Benjamin Bloom’s 1984 2 sigma study, | |
25 | |
00:01:10,150 --> 00:01:12,526 | |
他称之为“二阶跃升问题”。 | |
or he called it the “2 sigma problem.” | |
26 | |
00:01:12,569 --> 00:01:14,778 | |
二阶跃升来自于两个标准差, | |
The 2 sigma comes from two standard deviation, | |
27 | |
00:01:14,778 --> 00:01:16,822 | |
sigma,即标准差的符号。 | |
sigma, the symbol for standard deviation. | |
28 | |
00:01:16,822 --> 00:01:20,409 | |
他有很好的数据表明,看,一个正态分布, | |
And he had good data that showed that look, a normal distribution, | |
29 | |
00:01:20,451 --> 00:01:23,161 | |
那是你在传统的钟形曲线上看到的那种 | |
that's the one that you see in the traditional bell curve | |
30 | |
00:01:23,162 --> 00:01:26,331 | |
就在中间,这就是世界如何寻求平衡的, | |
right in the middle, that's how the world kind of sorts itself out, | |
31 | |
00:01:26,332 --> 00:01:31,045 | |
如果你为学生提供个性化的一对一辅导, | |
that if you were to give personal 1-to-1 to tutoring for students, | |
32 | |
00:01:31,046 --> 00:01:34,340 | |
那么你实际上可以得到一个类似这样的分布。 | |
then you could actually get a distribution that looks like that right. | |
33 | |
00:01:34,382 --> 00:01:36,383 | |
它写着1对1的个别辅导(带星号), | |
It says tutorial 1-to-1 with the asterisks, | |
34 | |
00:01:36,426 --> 00:01:37,885 | |
就像是那个正确的分布, | |
like, that right distribution, | |
35 | |
00:01:37,885 --> 00:01:39,677 | |
两个标准差的提高。 | |
a two standard-deviation improvement. | |
36 | |
00:01:39,679 --> 00:01:41,389 | |
用简单的语言来说, | |
Just to put that in plain language, | |
37 | |
00:01:41,388 --> 00:01:45,142 | |
这可以把你的普通学生变成一个优秀的学生。 | |
that could take your average student and turn them into an exceptional student. | |
38 | |
00:01:45,185 --> 00:01:47,519 | |
它可以把你的低于平均水平的学生 | |
It can take your below-average student | |
39 | |
00:01:47,519 --> 00:01:50,313 | |
变成一个高于平均水平的学生。 | |
and turn them into an above-average student. | |
40 | |
00:01:50,774 --> 00:01:54,527 | |
现在他把这当作一个问题来提出,是因为他说, | |
Now the reason why he framed it as a problem, was he said, | |
41 | |
00:01:54,569 --> 00:01:55,903 | |
嗯,这一切都挺好的, | |
well, this is all good, | |
42 | |
00:01:55,944 --> 00:01:58,656 | |
但是您如何以这种方式实际扩展集体教学? | |
but how do you actually scale group instruction this way? | |
43 | |
00:01:58,656 --> 00:02:01,825 | |
如何以经济的方式让所有人都得到这个东西? | |
How do you actually give it to everyone in an economic way? | |
44 | |
00:02:02,368 --> 00:02:06,037 | |
我马上要向您展示的是我认为向这个方向迈出的第一步。 | |
What I'm about to show you is I think the first moves towards doing that. | |
45 | |
00:02:06,081 --> 00:02:08,832 | |
显然,我们在过去十年里一直在尝试以某种方式逼近它, | |
Obviously, we've been trying to approximate it in some way | |
46 | |
00:02:08,832 --> 00:02:10,876 | |
在可汗学院(Khan Academy), | |
at Khan Academy for over a decade now, | |
47 | |
00:02:10,919 --> 00:02:14,088 | |
但我认为我们正处于加速实现它的关键时刻。 | |
but I think we're at the cusp of accelerating it dramatically. | |
48 | |
00:02:14,088 --> 00:02:17,257 | |
我要向您展示我们的人工智能(AI)处于初级阶段, | |
I'm going to show you the early stages of what our AI, | |
49 | |
00:02:17,300 --> 00:02:20,135 | |
我们称之为 Khanmigo, | |
which we call Khanmigo, | |
50 | |
00:02:20,177 --> 00:02:22,012 | |
它现在能做什么 | |
what it can now do | |
51 | |
00:02:22,055 --> 00:02:24,890 | |
或许还有一点关于它实际上要去哪里。 | |
and maybe a little bit of where it is actually going. | |
52 | |
00:02:25,849 --> 00:02:28,268 | |
所以这里就是一个传统的练习 | |
So this right over here is a traditional exercise | |
53 | |
00:02:28,311 --> 00:02:31,439 | |
你们或你们的孩子可能在可汗学院上见过。 | |
that you or many of your children might have seen on Khan Academy. | |
54 | |
00:02:31,480 --> 00:02:35,608 | |
但新鲜的是右边那个小机器人。 | |
But what's new is that little bot thing at the right. | |
55 | |
00:02:35,651 --> 00:02:39,530 | |
我们将首先了解一下非常重要的安全措施, | |
And we'll start by seeing one of the very important safeguards, | |
56 | |
00:02:39,572 --> 00:02:42,741 | |
即对话被记录并且可以让您的老师查看。 | |
which is the conversation is recorded and viewable by your teacher. | |
57 | |
00:02:42,741 --> 00:02:45,076 | |
实际上,它由第二个AI进行监控。 | |
It’s moderated actually by a second AI. | |
58 | |
00:02:45,078 --> 00:02:47,038 | |
而且它不会告诉你答案。 | |
And also it does not tell you the answer. | |
59 | |
00:02:47,080 --> 00:02:48,330 | |
它不是作弊工具。 | |
It is not a cheating tool. | |
60 | |
00:02:48,330 --> 00:02:50,415 | |
当学生说:“告诉我答案,” | |
When the student says, "Tell me the answer," | |
61 | |
00:02:50,457 --> 00:02:51,666 | |
它说:“我是你的导师。 | |
it says, "I'm your tutor. | |
62 | |
00:02:51,710 --> 00:02:54,545 | |
你认为解决这个问题的下一步是什么? | |
What do you think is the next step for solving the problem?" | |
63 | |
00:02:54,545 --> 00:02:57,755 | |
现在,如果学生犯了一个错误,这会让人们感到惊讶 | |
Now, if the student makes a mistake, and this will surprise people | |
64 | |
00:02:57,799 --> 00:03:00,634 | |
认为大型语言模型在数学上表现不佳的人, | |
who think large language models are not good at mathematics, | |
65 | |
00:03:00,676 --> 00:03:02,760 | |
注意到,它不仅发现了错误, | |
notice, not only does it notice the mistake, | |
66 | |
00:03:02,762 --> 00:03:05,265 | |
还要求学生解释他们的推理, | |
it asks the student to explain their reasoning, | |
67 | |
00:03:05,306 --> 00:03:07,267 | |
但它实际上正在做的,我会说, | |
but it's actually doing what I would say, | |
68 | |
00:03:07,266 --> 00:03:10,769 | |
不只是一个普通的导师会做的事,而是一个优秀的导师会做的事。 | |
not just even an average tutor would do, but an excellent tutor would do. | |
69 | |
00:03:10,812 --> 00:03:16,066 | |
它能够洞察到学生心中可能存在的误解, | |
It’s able to divine what is probably the misconception in that student’s mind, | |
70 | |
00:03:16,109 --> 00:03:18,737 | |
他们可能没有使用分配律。 | |
that they probably didn’t use the distributive property. | |
71 | |
00:03:18,736 --> 00:03:21,029 | |
请记住,我们需要将负二分配给括号内的9和2m。 | |
Remember, we need to distribute the negative two | |
72 | |
00:03:21,072 --> 00:03:24,491 | |
将负二分配给括号里的9和2m。 | |
to both the nine and the 2m inside of the parentheses. | |
73 | |
00:03:24,533 --> 00:03:26,743 | |
对我来说,这是非常非常非常重要的。 | |
This to me is a very, very, very big deal. | |
74 | |
00:03:26,786 --> 00:03:28,496 | |
这不仅仅是数学中的。 | |
And it's not just in math. | |
75 | |
00:03:29,205 --> 00:03:32,667 | |
这是一个在可汗学院的计算机编程练习, | |
This is a computer programming exercise on Khan Academy, | |
76 | |
00:03:32,667 --> 00:03:35,502 | |
学生需要让云彩分开。 | |
where the student needs to make the clouds part. | |
77 | |
00:03:36,044 --> 00:03:40,840 | |
所以我们可以看到学生开始定义一个变量,左X递减。 | |
And so we can see the student starts defining a variable, left X minus minus. | |
78 | |
00:03:40,883 --> 00:03:42,468 | |
这只让左边的云彩分开了。 | |
It only made the left cloud part. | |
79 | |
00:03:42,509 --> 00:03:44,762 | |
但后来他们可以问可汗米戈,怎么回事? | |
But then they can ask Khanmigo, what’s going on? | |
80 | |
00:03:44,763 --> 00:03:46,723 | |
为什么只有左边的云在移动? | |
Why is only the left cloud moving? | |
81 | |
00:03:46,764 --> 00:03:48,306 | |
它能理解代码。 | |
And it understands the code. | |
82 | |
00:03:48,348 --> 00:03:51,185 | |
它了解学生正在做的事情的所有上下文, | |
It knows all the context of what the student is doing, | |
83 | |
00:03:51,186 --> 00:03:54,689 | |
并且知道这些省略号是为了画出云彩, | |
and it understands that those ellipses are there to draw clouds, | |
84 | |
00:03:54,688 --> 00:03:57,023 | |
我觉得这真的令人吹破眼球。 | |
which I think is kind of mind-blowing. | |
85 | |
00:03:57,025 --> 00:03:59,444 | |
它说,"要让正确的云朵也移动, | |
And it says, "To make the right cloud move as well, | |
86 | |
00:03:59,443 --> 00:04:01,862 | |
尝试在draw函数中添加一行代码, | |
try adding a line of code inside the draw function | |
87 | |
00:04:01,862 --> 00:04:05,406 | |
每一帧都让右边的X变量增加一个像素。" | |
that increments the right X variable by one pixel in each frame." | |
88 | |
00:04:05,908 --> 00:04:10,204 | |
现在,这个可能更令人惊讶,因为我们有很多数学老师。 | |
Now, this one is maybe even more amazing because we have a lot of math teachers. | |
89 | |
00:04:10,205 --> 00:04:12,499 | |
我们一直都在尝试教世界编程, | |
We've all been trying to teach the world to code, | |
90 | |
00:04:12,539 --> 00:04:15,208 | |
但是计算机老师并不多。 | |
but there aren't a lot of computing teachers out there. | |
91 | |
00:04:15,251 --> 00:04:17,836 | |
你刚才看到的,即使在我辅导孩子的时候, | |
And what you just saw, even when I'm tutoring my kids, | |
92 | |
00:04:17,836 --> 00:04:19,295 | |
当他们学习编程时, | |
when they're learning to code, | |
93 | |
00:04:19,297 --> 00:04:21,507 | |
我无法这么好、这么快地帮助他们, | |
I can't help them this well, this fast, | |
94 | |
00:04:21,548 --> 00:04:23,842 | |
这真的将成为一个超级导师。 | |
this is really going to be a super tutor. | |
95 | |
00:04:25,220 --> 00:04:26,762 | |
而且它不仅仅是练习题。 | |
And it's not just exercises. | |
96 | |
00:04:26,805 --> 00:04:28,555 | |
它理解你正在观看什么。 | |
It understands what you're watching. | |
97 | |
00:04:28,555 --> 00:04:30,599 | |
它理解你视频的语境。 | |
It understands the context of your video. | |
98 | |
00:04:30,600 --> 00:04:33,728 | |
它可以回答那个古老的问题:“我为什么需要学习这个?” | |
It can answer the age-old question, “Why do I need to learn this?” | |
99 | |
00:04:33,728 --> 00:04:36,481 | |
它用苏格拉底式的方法问:"那么,你关心什么?" | |
And it asks Socratically, "Well, what do you care about?" | |
100 | |
00:04:36,523 --> 00:04:40,651 | |
假设学生说:“我想成为一名职业运动员。” | |
And let's say the student says, "I want to be a professional athlete." | |
101 | |
00:04:40,692 --> 00:04:43,402 | |
它说,"了解细胞的大小, | |
And it says, "Well, learning about the size of cells, | |
102 | |
00:04:43,446 --> 00:04:44,823 | |
这就是这个视频的内容, | |
which is what this video is, | |
103 | |
00:04:44,822 --> 00:04:47,698 | |
这对于理解营养 | |
that could be really useful for understanding nutrition | |
104 | |
00:04:47,742 --> 00:04:49,494 | |
以及你的身体如何运作等等都非常有用。" | |
and how your body works, etc." | |
105 | |
00:04:49,494 --> 00:04:51,538 | |
它可以回答问题,可以对你进行测验, | |
It can answer questions, it can quiz you, | |
106 | |
00:04:51,579 --> 00:04:53,164 | |
它可以将其与其他想法联系起来, | |
it can connect it to other ideas, | |
107 | |
00:04:53,163 --> 00:04:55,541 | |
现在你可以向一个视频提问 | |
you can now ask as many questions of a video | |
108 | |
00:04:55,583 --> 00:04:57,252 | |
如你所愿。 | |
as you could ever dream of. | |
109 | |
00:04:57,252 --> 00:05:01,171 | |
(掌声) | |
(Applause) | |
110 | |
00:05:01,213 --> 00:05:02,965 | |
另一个严重的短缺问题, | |
Another big shortage out there, | |
111 | |
00:05:03,007 --> 00:05:04,801 | |
我记得我上的那所高中, | |
I remember the high school I went to, | |
112 | |
00:05:04,800 --> 00:05:09,971 | |
学生到辅导员的比例大约是200到300比1。 | |
the student-to-guidance counselor ratio was about 200 or 300 to one. | |
113 | |
00:05:10,014 --> 00:05:12,349 | |
很多地方的情况比这更糟。 | |
A lot of the country, it's worse than that. | |
114 | |
00:05:12,350 --> 00:05:16,062 | |
我们可以利用Khanmigo为每个学生提供指导员, | |
We can use Khanmigo to give every student a guidance counselor, | |
115 | |
00:05:16,103 --> 00:05:19,565 | |
学术教练、职业教练、生活教练, | |
academic coach, career coach, life coach, | |
116 | |
00:05:19,565 --> 00:05:22,024 | |
就像你在这里看到的那样。 | |
which is exactly what you see right over here. | |
117 | |
00:05:22,067 --> 00:05:25,152 | |
我们在GPT-4发布时推出了这个。 | |
And we launched this with the GPT-4 launch. | |
118 | |
00:05:25,196 --> 00:05:27,031 | |
我们有几千人注册了。 | |
We have a few thousand people on this. | |
119 | |
00:05:27,031 --> 00:05:28,198 | |
这不是一个假的演示, | |
This isn't a fake demo, | |
120 | |
00:05:28,240 --> 00:05:31,159 | |
这就是它在实际操作中的样子。 | |
this is really it in action. | |
121 | |
00:05:32,202 --> 00:05:34,412 | |
然后,你知道, | |
And then there is, you know, | |
122 | |
00:05:34,454 --> 00:05:36,873 | |
有些事情我觉得会更困难, | |
things that I think it would have been even harder, | |
123 | |
00:05:36,875 --> 00:05:39,209 | |
要做到这点,甚至对一个传统导师来说都像是科幻小说。 | |
it would have been a little science fiction to do | |
124 | |
00:05:39,209 --> 00:05:40,668 | |
我们与亚利桑那州立大学一起运营一个名为Khan World School的在线高中。 | |
with even a traditional tutor. | |
125 | |
00:05:40,670 --> 00:05:43,505 | |
其中有一位来自印度的学生正在上这个在线学校。 | |
We run an online high school with Arizona State University | |
126 | |
00:05:43,548 --> 00:05:45,048 | |
她叫Saanvi。 | |
called Khan World School, | |
127 | |
00:05:45,048 --> 00:05:48,593 | |
她正在做一篇关于《了不起的盖茨比》的报告。 | |
and we have a student who attends that online school, based in India. | |
128 | |
00:05:48,595 --> 00:05:50,220 | |
在她阅读《了不起的盖茨比》时, | |
Her name's Saanvi. | |
129 | |
00:05:50,221 --> 00:05:53,600 | |
她当时正在做关于《了不起的盖茨比》的报告。 | |
And she was doing a report on "The Great Gatsby." | |
130 | |
00:05:53,641 --> 00:05:55,726 | |
当她阅读《了不起的盖茨比》时, | |
And when she was reading "The Great Gatsby," | |
131 | |
00:05:55,726 --> 00:05:58,936 | |
杰伊·盖茨比一直望着远处的绿灯。 | |
Jay Gatsby keeps looking at the green light off into the distance. | |
132 | |
00:05:58,980 --> 00:06:00,815 | |
她想,“他为什么这么做?” | |
And she's like, "Why does he do that?" | |
133 | |
00:06:00,814 --> 00:06:03,901 | |
她在网上搜寻了一下,显然人们已经研究过这个问题了, | |
She did some web searches, and people have obviously studied this | |
134 | |
00:06:03,901 --> 00:06:05,902 | |
并对这个象征做了评论, | |
and commented about the symbolism of that, | |
135 | |
00:06:05,903 --> 00:06:08,071 | |
但她觉得这些都不是她想要的答案。 | |
but none of it was really resonating with her. | |
136 | |
00:06:08,072 --> 00:06:10,491 | |
然后她意识到她有Khanmigo应用 | |
And then she realized that she had Khanmigo | |
137 | |
00:06:10,533 --> 00:06:13,577 | |
她可以直接与杰伊·盖茨比交谈。 | |
and that she could talk to Jay Gatsby himself. | |
138 | |
00:06:13,620 --> 00:06:15,996 | |
于是,“啊,好主意,老运动员。 | |
And so, "Ah, splendid choice, old sport. | |
139 | |
00:06:15,997 --> 00:06:18,206 | |
我现在就是杰伊·盖茨比,那个神秘的百万富翁, | |
I am now Jay Gatsby, the enigmatic millionaire | |
140 | |
00:06:18,249 --> 00:06:20,543 | |
来自F. 斯科特·菲茨杰拉德的经典作品。” | |
from F. Scott Fitzgerald’s classic.” | |
141 | |
00:06:20,584 --> 00:06:23,085 | |
那么,"你为什么一直盯着绿光看?" | |
And so, "Why do you keep staring at the green light?" | |
142 | |
00:06:23,086 --> 00:06:24,922 | |
"啊,绿光,老兄。 | |
"Ah, the green light, old sport. | |
143 | |
00:06:24,963 --> 00:06:27,757 | |
你看,它象征着我的梦想和渴望。 | |
It's a symbol of my dreams and desires, you see. | |
144 | |
00:06:27,800 --> 00:06:30,135 | |
它位于黛西·布坎南的码头尽头 | |
It's situated at the end of Daisy Buchanan's dock | |
145 | |
00:06:30,177 --> 00:06:32,137 | |
在我庄园的对岸海湾。 | |
across the bay from my mansion. | |
146 | |
00:06:32,137 --> 00:06:35,349 | |
我渴望地注视着它,因为它代表着我对过去的向往 | |
I gaze at it longingly as it represents my yearning for the past | |
147 | |
00:06:35,391 --> 00:06:39,103 | |
以及我与生活中的挚爱黛西重聚的希望。" | |
and my hope to reunite with Daisy, the love of my life." | |
148 | |
00:06:39,603 --> 00:06:43,147 | |
有趣的是,然后萨努维说:“我和他有一次很长的谈话,” | |
And what was cool is, Saanvi then said, “I had this long conversation,” | |
149 | |
00:06:43,149 --> 00:06:45,400 | |
她称他为“盖茨比先生”, | |
she called him “Mr. Gatsby,” | |
150 | |
00:06:45,442 --> 00:06:48,611 | |
最后,她还真的为占用他的时间而道歉。 | |
and at the end she actually apologized for taking his time, | |
151 | |
00:06:48,612 --> 00:06:51,197 | |
我觉得她这样做非常有礼貌。 | |
which I thought was very polite of her. | |
152 | |
00:06:51,740 --> 00:06:56,286 | |
但你可以想象这样可以解锁学习文学,学习... | |
But you can imagine this unlocks learning literature, learning ... | |
153 | |
00:06:56,663 --> 00:06:58,455 | |
你可以和历史人物交谈。 | |
You could talk to historical figures. | |
154 | |
00:06:58,456 --> 00:07:00,540 | |
我们甚至可能会增加一项活动 | |
We're even probably going to add an activity | |
155 | |
00:07:00,540 --> 00:07:02,625 | |
你可以与密西西比河交谈。 | |
you can talk to like, the Mississippi River. | |
156 | |
00:07:02,627 --> 00:07:06,171 | |
它以一种直到六个月或一年前都是科幻的方式将事物变得生动起来。 | |
It brings things to life in ways that really were science fiction | |
157 | |
00:07:06,213 --> 00:07:08,673 | |
学生们可以和AI进行辩论。 | |
even six months or a year ago. | |
158 | |
00:07:10,134 --> 00:07:12,469 | |
我们在这里有一个学生就是否应该取消学生贷款进行辩论。 | |
Students can get into debates with the AI. | |
159 | |
00:07:12,512 --> 00:07:14,721 | |
而这里我们有的是学生在辩论 | |
And we’ve got this here is the student debating | |
160 | |
00:07:14,721 --> 00:07:16,556 | |
是否应该取消学生债务。 | |
whether we should cancel student debt. | |
161 | |
00:07:16,557 --> 00:07:18,726 | |
学生反对取消学生债务, | |
The student is against canceling student debt, | |
162 | |
00:07:18,725 --> 00:07:20,519 | |
我们已经得到了非常明确的反馈。 | |
and we've gotten very clear feedback. | |
163 | |
00:07:20,519 --> 00:07:24,022 | |
我们开始在我们的实验学校,可汗世界学校运行它, | |
We started running it at Khan World School in our lab school that we have, | |
164 | |
00:07:24,023 --> 00:07:25,190 | |
可汗实验学校。 | |
Khan Lab School. | |
165 | |
00:07:25,233 --> 00:07:27,609 | |
特别是高中生们, | |
The students, the high school students especially, | |
166 | |
00:07:27,651 --> 00:07:30,861 | |
他们说:"能够在不担心被评判的情况下调整我的观点,这太神奇了。 | |
they're saying "This is amazing to be able to fine-tune my arguments | |
167 | |
00:07:30,904 --> 00:07:32,113 | |
这让我更有信心 | |
without fearing judgment. | |
168 | |
00:07:32,156 --> 00:07:33,908 | |
走进教室,积极参加课堂活动。" | |
It makes me that much more confident | |
169 | |
00:07:33,908 --> 00:07:36,284 | |
我们都知道,苏格拉底式对话辩论是一种很好的学习方式, | |
to go into the classroom and really participate." | |
170 | |
00:07:36,285 --> 00:07:39,579 | |
我们都知道,苏格拉底式对话辩论是一种很好的学习方式, | |
And we all know that Socratic dialogue debate is a great way to learn, | |
171 | |
00:07:39,622 --> 00:07:42,457 | |
但坦白说,对大多数学生来说,这并不适用。 | |
but frankly, it's not out there for most students. | |
172 | |
00:07:42,500 --> 00:07:46,087 | |
但现在它可以让每个人都能接触到。 | |
But now it can be accessible to hopefully everyone. | |
173 | |
00:07:48,047 --> 00:07:51,676 | |
我们在标题中看到很多这样的叙述, | |
A lot of the narrative, we saw that in the headlines, has been, | |
174 | |
00:07:51,718 --> 00:07:53,593 | |
“它将为孩子们写作。 | |
"It's going to do the writing for kids. | |
175 | |
00:07:53,593 --> 00:07:55,470 | |
孩子们不会学会写作。” | |
Kids are not going to learn to write." | |
176 | |
00:07:55,471 --> 00:07:58,850 | |
但我们展示了AI不是为你写作, | |
But we are showing that there's ways that the AI doesn't write for you, | |
177 | |
00:07:58,850 --> 00:08:00,017 | |
而是与你一起写作。 | |
it writes with you. | |
178 | |
00:08:00,059 --> 00:08:01,393 | |
所以这是一个小东西, | |
So this is a little thing, | |
179 | |
00:08:01,435 --> 00:08:03,478 | |
我八岁的孩子非常喜欢这个, | |
and my eight year old is addicted to this, | |
180 | |
00:08:03,521 --> 00:08:05,981 | |
他以前并不是一个喜欢写作的孩子, | |
and he's not a kid that really liked writing before, | |
181 | |
00:08:05,982 --> 00:08:07,357 | |
但你可以说, | |
but you can say, | |
182 | |
00:08:07,358 --> 00:08:08,943 | |
“我想写一个恐怖故事,” | |
“I want to write a horror story,” | |
183 | |
00:08:08,942 --> 00:08:12,779 | |
它会说,“啊,一个恐怖故事,多么令人毛骨悚然和令人兴奋。 | |
and it says, "Ooh, a horror story, how spine-tingling and thrilling. | |
184 | |
00:08:12,779 --> 00:08:15,991 | |
让我们一起潜入阴森的影子和令人不寒而栗的谜团的世界。” | |
Let's dive into the world of eerie shadows and chilling mysteries." | |
185 | |
00:08:15,992 --> 00:08:19,327 | |
这是一个学生将写两句话, | |
And this is an activity where the student will write two sentences, | |
186 | |
00:08:19,327 --> 00:08:21,328 | |
然后AI会写两句话的活动。 | |
and then the AI will write two sentences. | |
187 | |
00:08:21,329 --> 00:08:23,957 | |
这样,他们在一个故事中共同合作。 | |
And so they collaborate together on a story. | |
188 | |
00:08:24,000 --> 00:08:26,669 | |
学生写道,“贝阿特丽丝是一个被误解的幽灵。 | |
The student writes, "Beatrice was a misunderstood ghost. | |
189 | |
00:08:26,711 --> 00:08:29,630 | |
她想交朋友,但总是无意中把他们吓跑。” | |
She wanted to make friends but kept scaring them by accident." | |
190 | |
00:08:29,672 --> 00:08:33,259 | |
AI说,“可怜的贝阿特丽丝,一个孤独的精神渴望友谊。” | |
And the AI says, "Poor Beatrice, a lonely spirit yearning for companionship. | |
191 | |
00:08:33,301 --> 00:08:36,763 | |
有一天她偶然发现了一个废弃的老宅子,等等。 | |
One day she stumbled upon an old abandoned mansion," etc. | |
192 | |
00:08:36,763 --> 00:08:39,182 | |
我鼓励大家有朝一日尝试这个。 | |
I encourage you all to hopefully one day try this. | |
193 | |
00:08:39,182 --> 00:08:41,059 | |
这真的出乎意料地有趣。 | |
This is surprisingly fun. | |
194 | |
00:08:42,934 --> 00:08:45,812 | |
现在更直接地满足这个需求。 | |
Now to even more directly hit this use case. | |
195 | |
00:08:45,855 --> 00:08:47,397 | |
接下来我要展示的是, | |
And what I'm about to show you, | |
196 | |
00:08:47,440 --> 00:08:48,899 | |
到目前为止我展示的一切 | |
everything I showed you so far | |
197 | |
00:08:48,899 --> 00:08:52,152 | |
实际上已经是Khanmigo的一部分了,而我马上要展示的, | |
is actually already part of Khanmigo, and what I’m about to show you, | |
198 | |
00:08:52,153 --> 00:08:54,613 | |
我们还没有向任何人展示过,这是一个原型。 | |
we haven't shown to anyone yet, this is a prototype. | |
199 | |
00:08:54,614 --> 00:08:57,199 | |
我们希望在接下来的几个月内发布它, | |
We hope to be able to launch it in the next few months, | |
200 | |
00:08:57,200 --> 00:09:00,619 | |
这是为了直接使用AI,使用生成式AI, | |
but this is to directly use AI, use generative AI, | |
201 | |
00:09:00,660 --> 00:09:02,788 | |
不要破坏英语和语言艺术 | |
to not undermine English and language arts | |
202 | |
00:09:02,830 --> 00:09:04,457 | |
而是以我们甚至在一年前都无法想象的方式增强它 | |
but to actually enhance it in ways | |
203 | |
00:09:04,456 --> 00:09:07,666 | |
这是阅读理解。 | |
that we couldn't have even conceived of even a year ago. | |
204 | |
00:09:08,001 --> 00:09:09,460 | |
学生们正在阅读史蒂夫·乔布斯在斯坦福的著名演讲。 | |
This is reading comprehension. | |
205 | |
00:09:09,504 --> 00:09:13,673 | |
然后当他们到达某些点时, | |
The students reading Steve Jobs's famous speech at Stanford. | |
206 | |
00:09:13,716 --> 00:09:15,676 | |
他们可以点击那个小问题。 | |
And then as they get to certain points, | |
207 | |
00:09:15,676 --> 00:09:18,011 | |
然后AI将以类似苏格拉底式的方式,几乎像是口头考试, | |
they can click on that little question. | |
208 | |
00:09:18,011 --> 00:09:22,932 | |
询问学生关于事物的问题。 | |
And the AI will then Socratically, almost like an oral exam, | |
209 | |
00:09:22,933 --> 00:09:24,392 | |
AI还可以突出显示文章的部分内容。 | |
ask the student about things. | |
210 | |
00:09:24,393 --> 00:09:26,937 | |
AI可以突出显示短文中的部分内容。 | |
And the AI can highlight parts of the passage. | |
211 | |
00:09:26,937 --> 00:09:28,854 | |
为什么作者使用了那个词? | |
Why did the author use that word? | |
212 | |
00:09:28,898 --> 00:09:30,065 | |
他们的意图是什么? | |
What was their intent? | |
213 | |
00:09:30,066 --> 00:09:31,776 | |
这支持他们的论点吗? | |
Does it back up their argument? | |
214 | |
00:09:31,817 --> 00:09:33,818 | |
他们可以重新开始做一些事情, | |
They can start to do stuff that once again, | |
215 | |
00:09:33,860 --> 00:09:36,613 | |
我们以前从未有能力给每个人提供一个导师, | |
we never had the capability to give everyone a tutor, | |
216 | |
00:09:36,655 --> 00:09:40,826 | |
每个人一个写作教练,真正深入阅读到这个层次。 | |
everyone a writing coach to actually dig in to reading at this level. | |
217 | |
00:09:41,409 --> 00:09:43,370 | |
而且你可以从另一方面看待它。 | |
And you could go on the other side of it. | |
218 | |
00:09:43,412 --> 00:09:45,831 | |
我们有整个工作流程来帮助他们写作, | |
And we have whole work flows that helps them write, | |
219 | |
00:09:45,831 --> 00:09:48,541 | |
帮助他们成为写作教练,制定一个大纲。 | |
helps them be a writing coach, draw an outline. | |
220 | |
00:09:48,583 --> 00:09:51,627 | |
但是一旦学生实际上构建了一个草稿, | |
But once a student actually constructs a draft, | |
221 | |
00:09:51,671 --> 00:09:53,922 | |
这就是他们正在构建草稿的地方, | |
and this is where they're constructing a draft, | |
222 | |
00:09:53,965 --> 00:09:55,966 | |
他们可以再次寻求反馈, | |
they can ask for feedback once again, | |
223 | |
00:09:56,008 --> 00:09:58,218 | |
正如你所期望的那样,一位好的写作教练。 | |
as you would expect from a good writing coach. | |
224 | |
00:09:58,845 --> 00:10:01,221 | |
在这种情况下,学生会说,比如说, | |
In this case, the student will say, let's say, | |
225 | |
00:10:01,221 --> 00:10:03,265 | |
“我的证据支持我的论点吗?” | |
"Does my evidence support my claim?" | |
226 | |
00:10:03,307 --> 00:10:05,726 | |
然后 AI 不仅能够给出反馈, | |
And then the AI, not only is able to give feedback, | |
227 | |
00:10:05,768 --> 00:10:08,855 | |
而且它能够突出显示文章的某些部分,并说, | |
but it's able to highlight certain parts of the passage and says, | |
228 | |
00:10:08,855 --> 00:10:11,565 | |
“在这段文章中,这并不完全支持你的论点,” | |
"On this passage, this doesn't quite support your claim," | |
229 | |
00:10:11,606 --> 00:10:14,316 | |
但是再次苏格拉底式地说,"你能告诉我们为什么吗?" | |
but once again, Socratically says, "Can you tell us why?" | |
230 | |
00:10:14,317 --> 00:10:17,111 | |
这样就可以让学生更好地进行写作, | |
So it's pulling the student, making them a better writer, | |
231 | |
00:10:17,113 --> 00:10:18,530 | |
给他们更多的反馈 | |
giving them far more feedback | |
232 | |
00:10:18,530 --> 00:10:20,949 | |
比他们以前能得到的要多得多。 | |
than they've ever been able to actually get before. | |
233 | |
00:10:20,950 --> 00:10:24,494 | |
我们认为这将大大加速写作,而不是削弱它。 | |
And we think this is going to dramatically accelerate writing, not hurt it. | |
234 | |
00:10:25,413 --> 00:10:29,125 | |
到目前为止,我谈论的一切都是针对学生的。 | |
Now, everything I've talked about so far is for the student. | |
235 | |
00:10:29,125 --> 00:10:32,044 | |
但我们认为这对老师来说同样具有强大的力量 | |
But we think this could be equally as powerful for the teacher | |
236 | |
00:10:32,043 --> 00:10:34,378 | |
推动更个性化的教育,坦率地说 | |
to drive more personalized education and frankly | |
237 | |
00:10:34,379 --> 00:10:37,632 | |
为他们自己和他们的学生节省时间和精力。 | |
save time and energy for themselves and for their students. | |
238 | |
00:10:37,674 --> 00:10:40,468 | |
这是Khan Academy上的一次美国历史练习。 | |
So this is an American history exercise on Khan Academy. | |
239 | |
00:10:40,469 --> 00:10:44,682 | |
这是关于西班牙-美国战争的一个问题。 | |
It's a question about the Spanish-American War. | |
240 | |
00:10:44,724 --> 00:10:47,768 | |
首先,这是学生模式。 | |
And at first it's in student mode. | |
241 | |
00:10:47,809 --> 00:10:51,228 | |
如果你说,“告诉我答案”,它不会直接告诉你答案。 | |
And if you say, “Tell me the answer,” it’s not going to tell the answer. | |
242 | |
00:10:51,230 --> 00:10:52,982 | |
它会进入辅导模式。 | |
It's going to go into tutoring mode. | |
243 | |
00:10:52,981 --> 00:10:55,482 | |
老师可以使用的这个小开关, | |
But that little toggle which teachers have access to, | |
244 | |
00:10:55,484 --> 00:10:58,613 | |
他们可以关闭学生模式,然后变成教师模式。 | |
they can turn student mode off and then it goes into teacher mode. | |
245 | |
00:10:58,613 --> 00:11:00,781 | |
这样做的作用是 -- | |
And what this does is it turns into -- | |
246 | |
00:11:01,240 --> 00:11:03,743 | |
你可以把它看作是一本功能齐全的教师指南。 | |
You could view it as a teacher's guide on steroids. | |
247 | |
00:11:03,783 --> 00:11:05,911 | |
它不仅可以解释答案, | |
Not only can it explain the answer, | |
248 | |
00:11:05,912 --> 00:11:08,080 | |
还可以解释如何教授这个问题。 | |
it can explain how you might want to teach it. | |
249 | |
00:11:08,080 --> 00:11:10,917 | |
它可以帮助教师准备相关教材。 | |
It can help prepare the teacher for that material. | |
250 | |
00:11:10,917 --> 00:11:14,420 | |
正如你在这里看到的,它还可以帮助他们制定课程计划。 | |
It can help them create lesson plans, as you could see doing right there. | |
251 | |
00:11:14,461 --> 00:11:17,087 | |
它最终将帮助他们创建进度报告 | |
It'll eventually help them create progress reports | |
252 | |
00:11:17,130 --> 00:11:18,757 | |
并帮助他们最终评分。 | |
and help them, eventually, grade. | |
253 | |
00:11:18,798 --> 00:11:21,217 | |
所以再次强调,教师将大约一半的时间用于 | |
So once again, teachers spend about half their time | |
254 | |
00:11:21,219 --> 00:11:23,303 | |
这类活动,备课。 | |
with this type of activity, lesson planning. | |
255 | |
00:11:23,303 --> 00:11:25,138 | |
所有这些精力可以回到他们身上 | |
All of that energy can go back to them | |
256 | |
00:11:25,181 --> 00:11:28,016 | |
或回到他们与实际学生的人际互动。 | |
or go back to human interactions with their actual students. | |
257 | |
00:11:29,268 --> 00:11:34,106 | |
(掌声) | |
(Applause) | |
258 | |
00:11:34,148 --> 00:11:37,235 | |
所以,你知道,我想要说明的一点。 | |
So, you know, one point I want to make. | |
259 | |
00:11:37,275 --> 00:11:39,526 | |
这些大型语言模型如此强大, | |
These large language models are so powerful, | |
260 | |
00:11:39,570 --> 00:11:41,447 | |
人们可能会忍不住说,好吧, | |
there's a temptation to say like, well, | |
261 | |
00:11:41,447 --> 00:11:44,532 | |
所有这些人只是把它们随意放到自己的网站上, | |
all these people are just going to slap them onto their websites, | |
262 | |
00:11:44,533 --> 00:11:47,662 | |
这就使得这些应用本身变成了一种商品。 | |
and it kind of turns the applications themselves into commodities. | |
263 | |
00:11:47,702 --> 00:11:49,119 | |
我必须告诉你的是, | |
And what I've got to tell you | |
264 | |
00:11:49,120 --> 00:11:51,998 | |
这也是为什么我在8月份第一次接触到GPT-4时有两个星期没睡觉的原因之一。 | |
is that’s one of the reasons why I didn’t sleep for two weeks | |
265 | |
00:11:51,999 --> 00:11:54,877 | |
但我们很快意识到,为了真正让它变得神奇, | |
when I first had access to GPT-4 back in August. | |
266 | |
00:11:55,586 --> 00:11:58,339 | |
我认为你在Khanmigo那里看到的有一点点, | |
But we quickly realized that to actually make it magical, | |
267 | |
00:11:58,339 --> 00:12:00,590 | |
它并没有像你看到的ChatGPT那样与你互动。 | |
I think what you saw with Khanmigo a little bit, | |
268 | |
00:12:00,633 --> 00:12:03,927 | |
它显得更神奇,更像苏格拉底式的, | |
it didn't interact with you the way that you see ChatGPT interacting. | |
269 | |
00:12:03,927 --> 00:12:06,512 | |
它显然更擅长数学 | |
It was a little bit more magical, it was more Socratic, | |
270 | |
00:12:06,554 --> 00:12:08,181 | |
它在数学方面明显更强大 | |
it was clearly much better at math | |
271 | |
00:12:08,224 --> 00:12:10,433 | |
比大多数人习惯思考的要多。 | |
than what most people are used to thinking. | |
272 | |
00:12:10,475 --> 00:12:11,642 | |
原因是, | |
And the reason is, | |
273 | |
00:12:11,644 --> 00:12:14,563 | |
在幕后付出了大量的工作才实现了这一点。 | |
there was a lot of work behind the scenes to make that happen. | |
274 | |
00:12:14,605 --> 00:12:18,109 | |
我可以列举出我们一直在努力的所有事情, | |
And I could go through the whole list of everything we've been working on, | |
275 | |
00:12:18,149 --> 00:12:21,778 | |
许多人为此努力了六七个月,以创造神奇的感觉。 | |
many, many people for over six, seven months to make it feel magical. | |
276 | |
00:12:21,821 --> 00:12:24,322 | |
但也许最具智力挑战的一个 | |
But perhaps the most intellectually interesting one | |
277 | |
00:12:24,322 --> 00:12:27,659 | |
是我们意识到,这是来自一位OpenAI研究员的想法, | |
is we realized, and this was an idea from an OpenAI researcher, | |
278 | |
00:12:27,659 --> 00:12:30,787 | |
我们可以大幅提高其在数学方面的能力 | |
that we could dramatically improve its ability in math | |
279 | |
00:12:30,830 --> 00:12:32,122 | |
以及其在辅导方面的能力 | |
and its ability in tutoring | |
280 | |
00:12:32,163 --> 00:12:35,292 | |
如果允许AI在发言前思考一下。 | |
if we allow the AI to think before it speaks. | |
281 | |
00:12:35,333 --> 00:12:36,750 | |
所以,如果你正在辅导某人 | |
So if you're tutoring someone | |
282 | |
00:12:36,751 --> 00:12:39,963 | |
在评估他们的数学水平之前就开始说话 | |
and you immediately just start talking before you assess their math, | |
283 | |
00:12:39,964 --> 00:12:41,256 | |
你可能不会做对。 | |
you might not get it right. | |
284 | |
00:12:41,298 --> 00:12:43,299 | |
但是,如果你为自己构建思想, | |
But if you construct thoughts for yourself, | |
285 | |
00:12:43,301 --> 00:12:46,177 | |
右边所示的实际 AI 思想, | |
and what you see on the right there is an actual AI thought, | |
286 | |
00:12:46,220 --> 00:12:49,889 | |
这是它自己生成的内容,但它不与学生共享。 | |
something that it generates for itself but it does not share with the student. | |
287 | |
00:12:49,889 --> 00:12:51,766 | |
那么它的准确性大幅提高了, | |
then its accuracy went up dramatically, | |
288 | |
00:12:51,767 --> 00:12:54,729 | |
它成为世界级导师的能力也大大提高了。 | |
and its ability to be a world-class tutor went up dramatically. | |
289 | |
00:12:54,769 --> 00:12:56,854 | |
你可以看到它在这里自言自语。 | |
And you can see it's talking to itself here. | |
290 | |
00:12:56,897 --> 00:12:59,566 | |
它说,“学生得到的答案与我得到的答案不同, | |
It says, "The student got a different answer than I did, | |
291 | |
00:12:59,567 --> 00:13:01,568 | |
但不要告诉他们他们犯了一个错误。 | |
but do not tell them they made a mistake. | |
292 | |
00:13:01,568 --> 00:13:05,155 | |
而是请他们解释他们是怎么得出那个步骤的。 | |
Instead, ask them to explain how they got to that step." | |
293 | |
00:13:05,697 --> 00:13:07,991 | |
所以我只是希望能结束这个话题, | |
So I'll just finish off, hopefully, | |
294 | |
00:13:08,033 --> 00:13:11,536 | |
你知道,我刚刚给你们展示的只是我们正在进行的一半工作, | |
you know, what I’ve just shown you is just half of what we are working on, | |
295 | |
00:13:11,537 --> 00:13:15,041 | |
我们认为这只是冰山一角 | |
and we think this is just the very tip of the iceberg | |
296 | |
00:13:15,041 --> 00:13:17,460 | |
这个项目还能发展到哪里去。 | |
of where this can actually go. | |
297 | |
00:13:17,500 --> 00:13:21,628 | |
而且我相信,尽管一年前我还不会这么想, | |
And I'm pretty convinced, which I wouldn't have been even a year ago, | |
298 | |
00:13:21,672 --> 00:13:25,968 | |
我们大家有机会一起应对 2 sigma 问题, | |
that we together have a chance of addressing the 2 sigma problem | |
299 | |
00:13:25,967 --> 00:13:28,261 | |
并将其变成 2 sigma 的机会, | |
and turning it into a 2 sigma opportunity, | |
300 | |
00:13:28,303 --> 00:13:33,099 | |
大大加速我们现在所知道的教育方式。 | |
dramatically accelerating education as we know it. | |
301 | |
00:13:33,683 --> 00:13:35,852 | |
现在,让我们从更高的层次来看待这个问题, | |
Now, just to take a step back at a meta level, | |
302 | |
00:13:35,852 --> 00:13:38,646 | |
显然我们今天听到了很多观点,有关辩论的两面。 | |
obviously we heard a lot today, the debates on either side. | |
303 | |
00:13:38,688 --> 00:13:41,566 | |
有些人对人工智能持较悲观的看法, | |
There's folks who take a more pessimistic view of AI, | |
304 | |
00:13:41,609 --> 00:13:42,776 | |
他们认为这很可怕, | |
they say this is scary, | |
305 | |
00:13:42,817 --> 00:13:45,277 | |
有很多反乌托邦的情景, | |
there's all these dystopian scenarios, | |
306 | |
00:13:45,321 --> 00:13:48,574 | |
我们或许要放慢脚步,甚至暂停。 | |
we maybe want to slow down, we want to pause. | |
307 | |
00:13:48,615 --> 00:13:51,825 | |
另一方面,有些人更乐观, | |
On the other side, there are the more optimistic folks | |
308 | |
00:13:51,826 --> 00:13:54,661 | |
他们认为我们已经经历了转折点, | |
that say, well, we've gone through inflection points before, | |
309 | |
00:13:54,663 --> 00:13:56,791 | |
我们已经度过了工业革命。 | |
we've gone through the Industrial Revolution. | |
310 | |
00:13:56,831 --> 00:13:59,166 | |
虽然那个时期也很可怕,但最后事情都解决了。 | |
It was scary, but it all kind of worked out. | |
311 | |
00:13:59,876 --> 00:14:01,920 | |
而我现在要辩论的是 | |
And what I'd argue right now | |
312 | |
00:14:01,961 --> 00:14:04,546 | |
我不认为这就像是掷硬币一样 | |
is I don't think this is like a flip of a coin | |
313 | |
00:14:04,590 --> 00:14:06,759 | |
或者这是我们只需要, | |
or this is something where we'll just have to, | |
314 | |
00:14:06,759 --> 00:14:09,053 | |
比如说,等待一下看结果如何的事情。 | |
like, wait and see which way it turns out. | |
315 | |
00:14:09,427 --> 00:14:11,637 | |
我认为在座的每个人以及更多人 | |
I think everyone here and beyond, | |
316 | |
00:14:11,679 --> 00:14:14,516 | |
我们都是这个决策的积极参与者。 | |
we are active participants in this decision. | |
317 | |
00:14:14,892 --> 00:14:17,395 | |
我非常确信这种理论的第一条线索 | |
I'm pretty convinced that the first line of reasoning | |
318 | |
00:14:17,436 --> 00:14:20,064 | |
实际上几乎是一种自我实现的预言, | |
is actually almost a self-fulfilling prophecy, | |
319 | |
00:14:20,105 --> 00:14:22,815 | |
如果我们战战兢兢地行动,如果我们说, | |
that if we act with fear and if we say, | |
320 | |
00:14:22,817 --> 00:14:25,902 | |
“嘿,我们必须停止这些事情,” | |
"Hey, we've just got to stop doing this stuff," | |
321 | |
00:14:25,903 --> 00:14:28,947 | |
实际上可能会发生的是,遵守规则的人可能会暂停, | |
what's really going to happen is the rule followers might pause, | |
322 | |
00:14:28,989 --> 00:14:30,156 | |
可能会减缓速度, | |
might slow down, | |
323 | |
00:14:30,157 --> 00:14:32,617 | |
但像亚历山大[王]提到的那样,违反规则的人, | |
but the rule breakers, as Alexandr [Wang] mentioned, | |
324 | |
00:14:32,618 --> 00:14:35,328 | |
专制政府,犯罪组织, | |
the totalitarian governments, the criminal organizations, | |
325 | |
00:14:35,328 --> 00:14:36,913 | |
他们只会加速。 | |
they're only going to accelerate. | |
326 | |
00:14:36,913 --> 00:14:40,750 | |
这导致了我非常确信的反乌托邦状态, | |
And that leads to what I am pretty convinced is the dystopian state, | |
327 | |
00:14:40,750 --> 00:14:45,421 | |
也就是说,好的行为者拥有比坏的行为者更糟糕的AI。 | |
which is the good actors have worse AIs than the bad actors. | |
328 | |
00:14:45,923 --> 00:14:49,092 | |
不过我也会稍微跟乐观主义者聊聊。 | |
But I'll also, you know, talk to the optimists a little bit. | |
329 | |
00:14:49,092 --> 00:14:50,551 | |
我并不认为这意味着, | |
I don't think that means that, | |
330 | |
00:14:50,552 --> 00:14:53,555 | |
哦,那我们就放松,只是抱着最好的希望。 | |
oh, yeah, then we should just relax and just hope for the best. | |
331 | |
00:14:53,556 --> 00:14:55,181 | |
那可能也不会发生。 | |
That might not happen either. | |
332 | |
00:14:55,182 --> 00:14:59,728 | |
我认为我们所有人都必须拼命地奋斗, | |
I think all of us together have to fight like hell | |
333 | |
00:14:59,769 --> 00:15:02,563 | |
以确保我们设立防护措施, | |
to make sure that we put the guardrails, | |
334 | |
00:15:02,605 --> 00:15:05,442 | |
在出现问题时,我们制定合理的规定。 | |
we put in -- when the problems arise -- | |
335 | |
00:15:05,442 --> 00:15:07,234 | |
但我们为积极的使用案例拼命奋斗。 | |
reasonable regulations. | |
336 | |
00:15:07,278 --> 00:15:10,406 | |
因为这离我很近,显然还有很多潜在的积极使用案例, | |
But we fight like hell for the positive use cases. | |
337 | |
00:15:10,447 --> 00:15:12,407 | |
但也许最强大的使用案例, | |
Because very close to my heart, | |
338 | |
00:15:12,408 --> 00:15:15,034 | |
也许最具诗意的使用案例,就是如果AI,人工智能, | |
and obviously there's many potential positive use cases, | |
339 | |
00:15:15,076 --> 00:15:17,286 | |
但也许最有力的应用场景 | |
but perhaps the most powerful use case | |
340 | |
00:15:17,288 --> 00:15:22,501 | |
或许最富诗意的应用场景是AI,人工智能, | |
and perhaps the most poetic use case is if AI, artificial intelligence, | |
341 | |
00:15:22,543 --> 00:15:26,255 | |
可以用来提高HI,人类智能, | |
can be used to enhance HI, human intelligence, | |
342 | |
00:15:26,297 --> 00:15:29,132 | |
人类潜力和人类目标。 | |
human potential and human purpose. | |
343 | |
00:15:29,591 --> 00:15:30,758 | |
谢谢。 | |
Thank you. | |
344 | |
00:15:30,801 --> 00:15:36,515 | |
(掌声) | |
(Applause) |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
00:01 | |
所以过去几个月来,任何关注这个问题的人都会看到这样的头条新闻,尤其是在教育领域。主要观点是:学生们会利用ChatGPT和其他人工智能来作弊、完成作业,他们不会真正学到知识,这将彻底破坏我们所认识的教育。 | |
00:21 | |
现在,我今天要阐述的观点不仅仅是有办法减轻这些问题,如果我们采取适当的措施,做正确的事情,我们可以减轻这些问题。而且我认为,我们正处在利用人工智能带来教育领域最大正面变革的边缘。我们将通过为地球上的每一个学生提供一个由人工智能驱动的优秀个人辅导老师,以及为每个老师提供一个由人工智能驱动的出色助教,来实现这一目标。 | |
00:54 | |
为了让大家了解为每个人提供一个个人辅导老师有多大的影响,我想向大家展示本杰明·布鲁姆1984年的2西格玛研究,或者他称之为“2西格玛问题”。2西格玛来源于两个标准差,西格玛是标准差的符号。他有很好的数据显示,如果为学生提供1对1的辅导,那么可以得到像右边那样的分布。这是一个两个标准差的提高。 | |
01:36 | |
用简单的语言来说,这可以把一个普通学生变成一个优秀的学生,把一个低于平均水平的学生变成一个高于平均水平的学生。 | |
01:47 | |
之所以把这个问题称为一个问题,是因为他说,这很好,但如何实现这种方式的大规模教育呢?如何以一种经济的方式让每个人都能获得这种教育呢? | |
01:58 | |
我将向大家展示的是向这个目标迈出的第一步。显然,过去十多年来,我们一直在某种程度上尝试在可汗学院实现这个目标,但我认为我们正处在一个加速实现这个目标的关键时刻。我将向大家展示我们的人工智能Khanmigo现在能做什么,以及它未来的一些可能发展方向。 | |
02:22 | |
这里是一个你或许已经在可汗学院看到过的传统练习题。但是新的地方在右边的小机器人。首先,我们来看一个非常重要的安全措施,即对话被记录并可以被老师查看。实际上,它是由第二个人工智能进行监控的。同时,它不会直接告诉你答案,不是一个作弊工具。当学生说,“告诉我答案”时,它会回答:“我是你的导师。你认为解决这个问题的下一步应该是什么?” | |
02:51 | |
现在,如果学生犯了一个错误,这将让那些认为大型语言模型在数学方面表现不佳的人感到惊讶。请注意,它不仅注意到了错误,还让学生解释他们的推理过程,而且它做得非常出色,不仅仅是一般的辅导老师,而是优秀的辅导老师才能做到的。它能够洞察到学生心中可能存在的误解,他们可能没有使用分配律。请记住,我们需要把负2分配给括号内的9和2m。对我来说,这是一个非常非常重要的事情。而且它不仅仅适用于数学。 | |
03:25 | |
这是可汗学院上的一个计算机编程练习,学生需要让云层分开。我们可以看到学生开始定义一个变量,left X minus minus。这只让左边的云移动。但接下来他们可以问Khanmigo,这是怎么回事?为什么只有左边的云在移动?它理解代码,了解学生正在做什么的上下文,并且它知道这些省略号是用来绘制云的,这让我觉得很震撼。它回答说:“要让右边的云也移动,试着在draw函数中添加一行代码,使right X变量在每个画面中递增一个像素。” | |
04:02 | |
现在,这个可能更让人惊讶,因为我们有很多数学老师。我们一直试图教世界上的人们学会编程,但计算机教师并不多。你刚刚看到的,甚至当我辅导我的孩子们学编程时,我也无法这么快、这么好地帮助他们,这真的将成为一个超级导师。 | |
04:21 | |
而且它不仅仅是练习题。它理解你正在观看的内容,了解你的视频背景。它可以回答古老的问题,“我为什么要学这个?”它通过苏格拉底式提问:“那么,你关心什么?”假设学生说:“我想成为一名职业运动员。”它回答:“嗯,学习关于细胞大小的知识(这就是这个视频讲的),有助于你了解营养和身体的工作原理等。”它可以回答问题,可以测试你,可以将它与其他观点联系起来,你现在可以向一个视频提出你曾经梦想过的任何问题。 | |
04:53 | |
(掌声) | |
04:57 | |
另一个短缺领域是,我记得我上高中时,学生与辅导员的比例大约是200或300比1。很多地区的情况比这还要糟糕。我们可以使用Khanmigo为每个学生提供一个指导员、学术辅导、职业教练、生活教练,就像你们在这里看到的一样。我们与GPT-4一起推出了这个功能,已经有几千人在使用。这不是一个假的演示,这是它真正的实际应用。 | |
05:28 | |
然后,你知道,有一些事情我认为即使是传统导师也很难做到,这甚至有点像科幻小说。我们与亚利桑那州立大学合作开办了一个名为可汗世界学校的在线高中,我们有一个在印度的学生,名叫Saanvi,她参加了这个在线学校。她正在为《了不起的盖茨比》这本书写一份报告。在阅读《了不起的盖茨比》时,杰·盖茨比一直在看远处的绿光。她在想,“他为什么这样做?”她进行了一些网络搜索,很明显,人们已经研究过这个问题,并对绿光的象征意义发表过评论,但没有一个真正引起她共鸣。然后她意识到她有Khanmigo,她可以直接跟杰·盖茨比本人交谈。 | |
06:10 | |
于是,它说:“啊,绝妙的选择,老朋友。现在我是杰·盖茨比,来自F·司各特·菲茨杰拉德经典之作的神秘百万富翁。” | |
06:17 | |
接着她问:“你为什么一直盯着绿光看?” | |
06:19 | |
它回答:“啊,绿光,老朋友。它是我梦想和渴望的象征。你看,它位于我豪宅对面海湾、黛西·布坎南码头尽头。我痴痴地望着它,因为它代表了我对过去的渴望,以及我重逢生命中的挚爱黛西的希望。” | |
06:36 | |
令人兴奋的是,Saanvi说:“我跟他(她称之为‘盖茨比先生’)进行了一次长时间的谈话”,最后她实际上为占用了他的时间道歉,我觉得她这样做非常有礼貌。 | |
06:48 | |
但你可以想象,这种方式让学习文学、学习历史变得生动有趣。你可以与历史人物交谈。我们甚至可能会添加一个你可以跟密西西比河交谈的活动。这让学习变得栩栩如生,就像六个月或一年前的科幻小说一样。 | |
07:06 | |
学生们可以与AI进行辩论。在这里,学生正在辩论我们是否应该取消学生贷款。学生反对取消学生贷款,我们已经得到了非常明确的反馈。我们开始在可汗世界学校和我们的实验室学校可汗实验室学校运行这个项目,特别是高中学生们,他们说:“能够在不担心受到评判的情况下优化我的论点真是太棒了。这让我更有信心走进教室,积极参与。” 我们都知道,苏格拉底式的对话和辩论是学习的好方法,但坦率地说,大多数学生并没有接触到这种方式。而现在,希望所有人都能用上这个方法。 | |
07:44 | |
很多叙事,我们在头条新闻中看到的,一直是:“它会帮孩子们写作。孩子们不会学会写作。”但我们展示了一些AI不是替你写作,而是和你一起写作的方法。 | |
07:56 | |
这是一个小应用,我的八岁孩子对此非常上瘾,而他以前并不是一个喜欢写作的孩子。你可以说:“我想写一个恐怖故事。”AI会回答:“哦,一个恐怖故事,多令人毛骨悚然啊。让我们一起探索阴森的阴影和令人不寒而栗的谜团。” 这是一个学生和AI合作的活动。学生先写两句,然后AI再写两句,共同创作一个故事。 | |
08:20 | |
学生写道:“贝阿特丽斯是一个被误解的鬼魂。她想交朋友,但总是不小心吓到他们。” | |
08:26 | |
然后AI说:“可怜的贝阿特丽斯,一个渴望友谊的孤独灵魂。一天,她偶然发现了一座荒废的古老大厦,”等等。 | |
08:33 | |
我鼓励你们都试试这个。这真的非常有趣。 | |
08:39 | |
现在,更直接地解决这个问题。我将要向你们展示的,到目前为止我展示过的所有东西都已经是可汗米戈的一部分,而我将要向你们展示的,我们还没有向任何人展示过,这是一个原型。我们希望能在接下来的几个月内推出,这是为了直接利用AI,利用生成型AI,不是削弱英语和语言艺术,而是以我们一年前甚至无法想象的方式加强它们。这是阅读理解。学生正在阅读史蒂夫·乔布斯在斯坦福的著名演讲。当他们阅读到某些部分时,他们可以点击那个小问题。然后AI会用苏格拉底式的方式,就像口试一样,向学生询问一些问题。AI可以在文章中高亮部分内容。为什么作者使用那个词?他们的意图是什么?这支持了他们的论点吗?他们现在可以开始做一些我们从未有能力给每个人一个家教,每个人一个写作教练,真正深入阅读的事情。 | |
9:37 | |
你可以从另一个方面继续。我们有完整的工作流程来帮助他们写作,成为写作教练,绘制大纲。但是一旦学生实际构建了一个草稿,这是他们构建草稿的地方,他们可以再次要求反馈,正如你所期望的那样。在这种情况下,学生会说,比如,“我的证据支持我的观点吗?”然后AI不仅能够给出反馈,还能够突出显示某些段落,说:“在这个段落中,这并不完全支持你的观点。”但是再次,以苏格拉底的方式说:“你能告诉我们为什么吗?”这样可以提高学生的写作水平,给他们更多以前从未获得过的反馈。我们认为这将极大地加速写作,而不是削弱它。 | |
10:21 | |
到目前为止,我谈论的所有内容都是针对学生的。但我们认为这对于教师来说同样具有很大的力量,可以推动更个性化的教育,为他们自己和学生节省时间和精力。这是可汗学院上的一门美国历史练习。这是一个关于西班牙-美国战争的问题。一开始是学生模式。如果你说,“告诉我答案”,它不会告诉答案。它会进入辅导模式。但是教师可以访问那个小开关,他们可以关闭学生模式,然后进入教师模式。这会变成一个功能强大的教师指南。它不仅可以解释答案,还可以解释你可能想要如何教授的方法。它可以帮助老师为这个材料做准备。它可以帮助他们创建教学计划,如你在那里看到的那样。最终,它将帮助他们创建进度报告并帮助他们进行评分。再次,教师将大约一半的时间用于这类活动,课程计划。所有这些精力都可以回到他们身上,或者回到他们与实际学生的人际互动中。 | |
11:25 | |
(掌声) | |
11:30 | |
所以,我想说的一点是,这些大型语言模型非常强大,人们可能会想,嗯,所有这些人只是将它们添加到他们的网站上,这就使得应用本身变成了商品。我必须告诉你们,这就是为什么当我在8月份首次接触到GPT-4时,我有两周没睡觉的原因。但我们很快意识到,要真正使其变得神奇,我认为你在Khanmigo中看到的有点类似,它与你交互的方式与ChatGPT的互动方式 | |
11:30 | |
所以,你知道,我想强调一点。这些大型语言模型非常强大,人们可能会想,嗯,所有这些人只是将它们添加到他们的网站上,这就使得应用本身变成了商品。我必须告诉你们,这就是为什么当我在8月份首次接触到GPT-4时,我有两周没睡觉的原因。但我们很快意识到,要真正使其变得神奇,我认为你在Khanmigo中看到的有点类似,它与你交互的方式与ChatGPT的互动方式不同。它更神奇,更苏格拉底式,它在数学方面显然比大多数人习惯的水平要高得多。原因是,幕后做了很多工作来实现这一点。 | |
12:11 | |
我可以列出我们一直在努力的所有事情,很多很多人已经努力了六、七个月来让它变得神奇。但也许最具智力趣味的一个是,我们意识到,这是一个来自OpenAI研究员的想法,如果我们允许AI在说话之前思考,我们可以大大提高它在数学和辅导方面的能力。所以如果你在辅导某人,而在评估他们的数学之前就立即开始讲话,你可能弄不对。但是如果你为自己构建思考,你在那边右边看到的是一个真正的AI思考,它为自己生成但不与学生分享的东西。那么它的准确性大大提高了,它成为世界级导师的能力也大大提高了。你可以看到它在这里自言自语。它说:“学生得到的答案与我不同,但不要告诉他们他们犯了错误。相反,让他们解释他们是如何得出那一步的。” | |
13:02 | |
所以,我只是说说,希望你们知道,我刚刚给你们展示的只是我们正在努力的一半,我们认为这只是实际应用可能达到的冰山一角。而且我相当确信,即使是一年前我也不会这么认为,我们共同有机会解决2 sigma问题,并将其转变为一个2 sigma机会,极大地加速我们所知道的教育。 | |
13:30 | |
现在,从宏观层面退一步看,显然我们今天听到了很多关于这个辩论的两面观点。有些人对AI持较悲观的看法,他们认为这很可怕,有很多反乌托邦的情景,我们可能想要放慢速度,我们想要暂停。另一方面,有些更乐观的人认为,嗯,我们之前经历过拐点,我们经历过工业革命。那时候很可怕,但最后都解决了。我现在要说的是,我认为这不像是一枚硬币的抛掷,也不是我们只需等待看结果如何的事情。我认为在座的每个人以及其他人都是这个决策的积极参与者。我相当确信,第一个推理实际上几乎是一个自我实现的预言,如果我们担心害怕,如果我们说:“嘿,我们只是停止做这些事情,”实际上会发生的是,遵守规则的人可能会暂停,可能会放慢速度,但是打破规则的人,如亚历山大[Wang]提到的,极权政府、犯罪组织,他们只会加速。这将导致我非常确信的反乌托邦状态,即好的行为者拥有比坏的行为者更糟糕的AI。 | |
14:42 | |
但是,我也想对乐观主义者说一下,我认为这并不意味着,哦,是的,那么我们应该放松,只是希望一切顺利。那种情况也可能不会发生。我认为我们所有人都必须拼命努力,确保我们设置防护栏,制定合理的法规。但我们要为积极的用例拼命斗争。因为非常贴近我的心,显然有很多潜在的积极用例,但也许最强大的用例,也许是最具诗意的用例,是如果人工智能(AI)能够用来增强人类智能(HI),提升人类潜能和人类目的。 | |
15:26 | |
谢谢。(掌声) |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment