默契人MOQIPEOPLE INSIDER:
#AI #李開復KaiFuLee
#AISUPERPOWERS2018
Dr. 李開復 Kai-Fu Lee 给大学生的公开课：你该这样做选择
由教育部、创新工场、北京大学联合主办的DeeCamp 人工智慧训练营已经进行两周了。
作为夏令营的讲师，我上周给学生们上了一门AI课。没想到课后学生们咨询我最多的话题是：大学毕业后究竟应该工作，创业，还是读博当老师？也有很学生问AI是不是泡沫了，我们还要不要学等问题。
所以我又找时间跟夏令营的学生们做了一场职业选择问答课，帮大家答疑解惑。以下问答也许对其他学生也有帮助，分享给大家：
话题一：不支持应届生盲目创业
Q1:大学生创业应该怎么创？
我不支持应届大学生创业，在座的每一位现在拿商业计画书给我，我一个都不会收。为什么？因为创业就不是一个刚毕业的学生能做的。创新工场投资了300多个项目，只有两三个是刚毕业的学生做的，比如说旷视科技Face++。
刚毕业的时候，你会经验不足、人脉不够、不懂管理，所有我们担忧的问题，在这个过程中都会发生。所以，没有经验的创业，成功的概率微乎其微。想创业的同学不要那么急，你加入一个公司，快则两年，慢则三四年，再制定商业计画时，我们肯定会更慎重的考虑。
Q2:既然创业不能急，我们该找什么样的公司呢？
我觉得有几个选择：
首先大公司是有优势的，因为它有数据，有好的老板、好的体系，在AI时代有一个好的老板非常重要，要看他在不在乎、懂不懂AI，给不给你用他们的大数据。如果符合这些条件，我觉得国内的这些大公司都可以考虑。
但问题是，进入大公司，你就会被划分到一个个小小的部门，成为大机器的小齿轮，比如说你做技术就只能做技术，你想了解用户体验不可能；或者说你是做训练优化，你就做训练优化，你不要问数据怎么用，也不要担心产品怎么用。大公司不会让你接触到全方位的数据及产业运营。
如果要去创业公司，也可以有几种走法：
第一种，找一个已经走入我们视线的公司，比如Face++、商汤、依图等，进入这样一家做AI的公司，你也会学到很多。他们的规模可能已经几百上千人，但你仍然有机会见识到跨部门的运行。
第二种选择，是去那些非AI的公司，规模虽然不是很大，但有上升的潜力，并且已经认识到AI的重要性。
这些公司的特点是数据多、用户多，日活几千万后，突然发现他们需要AI的支援。当你选择这样一家公司时，你需要确定这个公司是懂AI并且在乎AI，比如说知乎、美图，或者每一个独角兽公司都是可选择的物件。
第三类是非常初创的公司，只有几十个人，比如我们投资的景驰。如果你更多想创业而不是找工作，在小公司里你会学得更多。公司的任何问题，无论产品、使用者、竞争、市场和技术等等，你都会接触到。
对于要创业的人来说，你要乐于接受，因为这是一个对CEO最好的培训，每一个有趣或枯燥的事你都要去做。
但是有一个底限，这家公司最少要有10个人。因为大部分三四个人的公司还是有局限性的，最好是10-50个人的公司。
最后一个建议，你在挑公司的时候，肯定是判断不了公司好或者坏的。我见过口才非常好的创始人，但是公司做的一塌糊涂，我也见到过非常内向木讷、说话无聊到听了就想睡觉的那种人，但是公司却很棒。那怎么分辨呢？
答案很简单，看投资方是谁。你们研究一下，看看有哪些国内或国际基金是可以相信的。这些基金经常出现在各种排行榜上，最好的个人或机构天使、最好的早期、或最好的AB轮早中期投资机构。你们去看看名单，看他们投了哪些项目。创新工场当然是其中之一，但值得尊重的基金至少有50个以上，数量蛮多的。
Q3：如果AI公司不幸在我进入之后倒闭了，接下来的路如何走？
这真是一个很传统但让我害怕的问题。不过你一点都不用担心，如果在座各位进入了前50名VC或天使投资的AI公司，两年之后这个公司倒闭了，我们绝对帮你找到工作，甚至帮你做出下一家公司。
在创业的路上，我们没有失败这个词，只有失业。创业的风险是存在的，但是工程师不会失败，积累的经验永远是加分项。
话题二：工作还是读博？
Q1：听说博士才能进 AI 公司，那我们不是应该先读一个博士？
千万不要相信这个，不排除有一些公司迷信博士，但你们不妨去找找那些寻找AI人才、而不是AI博士的公司，比如说创新工场人工智慧工程院。
当届毕业生，无论是博士、硕士、本科，实力可能相差不大。博士可能读了一个很差的学校，论文有可能根本没出版过，本科生也有学的比博士多的、更厉害的，所以你们要在毕业前这段期间发出光芒，让大家看到你虽然没有博士学位，但你仍然擅长于此。
Q2：如果要读博，国家、导师、学校的重要性怎么排序？
第一个顺序要选导师，第二个选学校，第三选地域。
比如，CMU（卡耐基梅隆大学）现在已经不是老师选学生，而是学生选老师了，这种机制是超级加分的。在CMU叫做marriage process，你去了以后有大概有五六十个老师给你讲他们的科研，最后你要一个个见他们。五六十个导师抢三十多个学生，学生都是很顶尖的。
如果没有这种制度，只是老师选学生的，那你们可以做的是什么？给老师们发Email呀。不要写太多，因为老师时间宝贵，你要写一页就能写完的。在Email里告诉老师你对他有关注，读过他的哪些论文，最后提出你的疑惑或者观点。
发邮件的前提是你真的读懂了文章。因为老师通常会给你回邮件解释，然后你可以回复，“噢，你的解释让我突然茅塞顿开，正好我明年要申请博士班……”。
如果你坚定要出国读AI的博士，美国、英国和加拿的几所大学可以优先考虑。对于大部分的学科而言，英国和加拿大是远远落后美国，但在机器学习方面，加拿大和英国有几所很不错的学校。
Q3：如果进不了好学校，还要读博士吗？
想像一下，你本科毕业后要花三四年的时间在一个三流的学校，写一些没有人读的论文，毕业之后也找不到一个好的教职，那还有什么意义？所以当你们找不到一个好的学校时，我建议你可以先在某个学校读个硕士。 （美国读博不一定要先取得硕士学位）
假设可以进美国前四十的学校，那就先读硕士，再申请前三十名的博士，宁可多花点时间也要读到好学校的博士学位。
还有同学说我读不了好的博士，我读博士后可不可以？我告诉各位，世界上没有博士后这个学位。博士后在美国是什么职位呢？就是找不到工作的博士生帮导师打工的职位，一般顶级博士毕业之后都开始做教职了，很少做博士后的，当然如果是在世界排名第一的学校做博士后，我们另说。
如果你真的进不了好学校怎么办？也不要沮丧，在这些排名略差的学校里面，找找看有没有顶级学校毕业的老师。因为你还要面对的一个现实是，即便你进了一个前十名的学校读博士，恐怕也找不到一个毕业于前十名学校的教师。
我有个非常要好的朋友就是在美国排名前五的大学毕业的，毕业之后想在美国找一个 Tenure EE professor（电子工程终身教职）多难你知道吗？每年申请CMU、MIT、Berkeley教职的有1000多人，但职位只有几个。所以你即便拿到前十名大学的毕业学位，如果你真的想去教书的话，你最好做好最坏的心理准备。
那如果你到一个排名五十名后的学校做教职，那你这一生会很不幸，因为大概也做不出什么成果。你们有谁看过哪一个美国第五十名学校做出什么伟大的研究呢？即便他是美国前五名大学毕业的，这个概率也微乎其微。为什么？因为老师的成果大多来自于学生，你在第五十名的学校能找到最好的学生吗？
但是如果反过来就有机会了，如果你进了前25名的学校，有一个老师他是三年前刚从一个前三名的学校毕业的，跟着这个老师做研究还是非常有希望的，这就是读博士的方向，希望对大家有帮助。
话题三：未来工作与AI
Q1：请问区块链跟AI到底哪个比较好？比较值得投资？
AI确定是会改变世界的，但区块链是一个有比较大的可能性，在未来的一段时间里面改变这个世界。我都比较看好，但很难把两者作比较，因为一个是已知的、一个是未知的。
Q2：请问人工智慧的泡沫期是不是到来了？
人工智慧会在未来15到25年不断成长、改变世界，它的整个产业链也会从CV（电脑视觉）走进NLP（自然语言处理），走向新的领域。
就像25年前，网路刚刚兴起，也是几经起伏，最后被广泛的应用。回溯一下网路的发展，经过了几波浪潮，从之前是PC上网为主，然后流览器带来一波浪潮，接下来是门户网站的机会，之后是搜寻引擎带来了广告、社交，而后的电商，移动互联网出现，而且逐渐成为主力。
所以网路在25年前是泡沫吗？在过去的25年中，我们可以列出12波、或是15波的网路发展，所以无论你就业是在25年前或者是去年，你都有可能赶上下一波浪潮。互联网的机会都还没有结束，人工智慧怎么会结束呢？所以泡沫的说法是不存在的。
那是不是表示所有AI公司都没有泡沫？也不是，现在AI公司很多估值过高，非AI公司也拿着AI的光环去忽悠投资人的钱，还有很多不懂的投资人真的投了，这些现象也是存在的。这些现象累积久了，肯定需要泄一泄气，让它回归到正常的状态，才能再往下走。所以才会看到接二连三的跌宕起伏，但从长期来说，未来的25年肯定是乐观的趋势。
你们做AI，也要有一个与时俱进的学习态度，就像如果25年前你进入网路，你说你一辈子就都做流览器，那后来就没工作咯！只要你与时俱进，不断反覆运算、不断学习新知识，任何一个行业都是没有泡沫的。比如当时做UC流览器的何小鹏，人家现在在做什么？人家造车了，对不对？但是对他来说汽车就是互联网的延伸啊！
再稍微做一点广告，你们再过30年后再看，我们创新工场所投资的AI公司，成功率一定是业界最高的，当然失败的肯定也会有。所以一个真正懂行的VC，他是可以看得清清楚楚的。
最后还要发布一则预告：
继DeeCamp之后，8月底创新工场联合搜狗、美团点评、美图发起的AI Challenger2018全球AI挑战赛也将拉开帷幕。从9月上旬开始，我们将在全国多所大学展开“校园行”活动，届时会有AI行业的顶级大咖亲临学校，与学生们现场互动；世界各地也将有参赛选手和爱好者组织的AI技术交流活动哦。具体日程资讯，本月内会公布。
Via Dr. 李開復 Kai-Fu Lee
Public Opening for college students: you should choose like this
The Deecamp Artificial Intelligence Training Camp, co-hosted by the ministry of education, innovation workshop, Beijing University, has been in place for two weeks.
As a lecturer in summer camp, I had an ai class for students last week. I didn’t expect the students to consult me on the most topic after the class: should I work, Entrepreneurship, or read the bo teacher after college graduation? There are also students who ask ai is not foam, we still have to don’t learn to wait.
So I’m looking for time to make a career choice and answer class with summer camp students to help everyone answer questions. The following quiz may also help other students, share it to everyone:
Topic one: do not support fresh birth blind entrepreneurship
Q1: how should college students be created?
I don’t support fresh college students entrepreneurship, every one of them is now getting a business plan book for me, and I won’t collect one. Why? Because Entrepreneurship is not a student who just graduated. Innovative workshops invest more than 300 projects, only two or three are made by students who just graduated, for example, kuang tech face++.
When just graduated, you will have insufficient experience, not enough connections, not understand management, all our concerns, will happen in this process. So, without experience, the odds of success are negligible. Classmates who want to start their business, don’t be so anxious, you join a company, two years fast, three or four years, and we will definitely consider it more carefully when we make a business plan.
Q2: since entrepreneurship can’t rush, what kind of company should we look for?
I think there are several options:
First big company is advantageous because it has data, there are good boss, good system, there is a good boss in ai era very important to see he is not care, understand not understand ai, give not for you Their big data. If these conditions are met, I think these big companies in the country can be considered.
But the problem is, enter the big company, you will be divided into a small and small sector and become a small gear for big machines, for example, you do technology just do technology, you want to know the user experience impossible; or say you Is to do training optimization, you do training optimization, you don’t ask the data how to use, also don’t worry about the product how to use. The big company won’t let you contact the full range of data and industrial operations.
If going to startup company, there can be a few ways to go:
First, looking for a company that has gone into our sight, such as face++, aliexpress soup, under, etc, enter such a company to do ai, you will also learn a lot. Their scale may have been hundreds of thousands, but you still have a chance to see the cross-sector run.
The second option is to go to the companies that are non-AI, the scale though not big, but with the potential of rising, and already recognize the importance of ai.
These companies are characterized by a lot of data, a lot of users, and a few million days after the day, suddenly found that they need ai’s support. When you choose a company like this, you need to be sure that this company is understand ai and cares about ai, for example, the United, the United, or every independence company is an optional item.
The third class is very startups, only dozens of people, such as the king we invest. If you more want to entrepreneurship instead of looking for work, in the small company you will learn more. Any questions of the company, regardless of products, users, competition, market and technology, and so on, you will be contacted.
For those who are going to entrepreneurship, you have to be happy to accept because it is the best training for the CEO, every funny or boring thing you are going to do.
But there is a bottom limit, the company is at least 10 people. Because most of the three-four-person companies are still limited, preferably 10-50 people’s companies.
Last advice, when you’re picking the company, it’s definitely not to judge the company good or bad. I’ve seen a very good founder, but the company has done a mess, and I have also met the kind of people who are very within, talking boring to the people who want to sleep, but the company is awesome. So how to tell?
The answer is simple, see who the investor is. You guys study and see which domestic or international funds can be believed. These funds often appear on various charts, the best individual or agency angel, the best early, or the best ab wheel early mid-term investment agency. You guys go check out the list and see what items they voted for. Innovative workshops are of course one of them, but the funds worth respect are at least 50 more and a lot.
Q3: if ai company unfortunately collapses after i enter, how is the next road going?
It’s really a tradition but scares me. But you don’t have to worry about it, if you guys enter the first 50 VC or angel investment ai company, after two years this company is closed, we definitely help you find a job and even help you make the next home The company.
On the way to entrepreneurship, we didn’t fail the word, only unemployment. The risk of entrepreneurship exists, but engineers do not fail, and the accumulated experience is always added.
Topic Two: work or read bo?
Q1: heard Dr. can enter ai company, then shouldn’t we read a Phd first?
Don’t believe this, don’t rule out some company superstition Dr, but you guys may want to find the company that looks for ai talent, not Dr. Ai, for example, the innovative workshop artificial intelligence engineering.
When Graduate Graduates, whether Phd, Master, undergraduate, strength may vary. Dr. May read a very poor school, and the paper is probably not published at all, undergraduate students have more than Dr. Learn and more powerful, so you guys have to make a glow during the period before graduation, let everyone see You don’t have a Phd, but you’re still good at it.
Q2: if to read the importance of bo, country, mentor, school, how to sort?
First order to pick up mentor, second pick school, third pick geographical.
For example, CMU (Card University of card) is now not a teacher to choose a student, but a student elected teacher, and this mechanism is super plus. In cmu, it is called marriage process, and after you go, there are probably teachers who give you their research, and finally you have to see them all. Fifty-six mentors robbed more than students, and students are very top.
If there is no such system, just teacher pick students, then what can you guys do? Email the teachers. Don’t write too much, because teacher time is precious, you have to write a page to finish it. Tell the teacher in email that you have a concern about him, read his thesis, and finally raise your doubts or views.
The premise of email is that you really read the article. Because teachers usually give you back to the mail to explain, then you can reply,” awww your explanation makes me suddenly ever, exactly I’m going to apply for the Phd class next year……”.
If you are determined to go abroad to read ai’s Phd, several universities in the United States, UK and Canada can prioritize. For most disciplines, the UK and Canada are far behind the us, but in machine learning, there are several very nice schools in Canada and Britain.
Q3: if not in good school, still have to read Dr.?
Imagine, you have to spend three or four years after your undergraduate graduation in a three-stream school, write some papers that have no read, and after graduation, can’t find a good teaching job, then what’s the point? So when you guys can’t find a good school, I suggest you can read a master in a school first. (American reading poh doesn’t necessarily have to get a master’s degree first)
Assuming that we can enter the first schools in the United States, then read the master first, and apply for the first doctors, rather take a little more time to read the Phd degree in the good school.
Also classmate said I can’t read good Dr. I read postdoctoral can not? I’m telling you, there’s no postdoctoral in the world. What position is postdoctoral in America? It is the job of a Phd student who can’t find a job. The General Top Phd has started to do the teaching job after graduation, and rarely do the post-Doctoral, of course, if it is in the world ranked first school to do the post-Doctoral, we
What if you really can’t get in the good school? Also don’t be depressed, in these ranking schools, look for teachers who have no top school graduation. Because the reality you still have to face is that even if you get into a top ten school reading dr, I’m afraid I can’t find a teacher who graduated from the top ten schools.
I have a very good friend who graduated from the top five college in the United States, and after graduation, wanted to find a tenure ee professor (E-engineering lifelong teaching post) in the us. How hard do you know? Every year, there are more than 1000 people applying for cmu, MIT, Berkeley, but there are only a few positions. So even if you get the top ten college graduation degrees, if you really want to go to teach, you better do the worst psychological preparation.
Then if you come to a school in a rank of fifty -, then you will be unlucky in your life, because probably don’t make any results. Have any of you ever seen which great study of the th school in America? Even if he was graduated from the top five universities in the United States, the odds were negligible. Why? Because teacher results are mostly from students, can you find the best students in the th school?
But if in turn there is a chance, if you get into the top 25 school, there’s a teacher who just graduated from a top three school three years ago, follow this teacher for research still very promising The of, that’s the way to read Dr., hope to help everyone.
Topic three: future work with ai
Q1: what is the comparison between the block chain and ai? Compare Worth Investment?
Ai sure is going to change the world, but the block chain is a relatively big possibility to change the world in the future. I’m all better, but it’s hard to compare both, because one is known and one is unknown.
Q2: May I ask if the bubble period of artificial intelligence is coming?
Artificial intelligence will grow in the next 15 to 25 years and change the world, and its entire chain of industry will also go from CV (computer visual) TO NLP (Natural Language processing), towards new areas.
Like 25 years ago, the network just came up, and it was also a lot of twists and downs, and it was finally Retrace the development of the network, after a few wave waves, from before it was pc online, then the flow brought a wave of wave, followed by the portal chance, after the search engine brought advertising, social, and then electricity Aliexpress, mobile internet appeared and gradually became the main.
So the network is foam 25 years ago? In the last 25 years, we can list 12 wave or 15 wave of network development, so whether you employment is 25 years ago or last year, you are all likely to catch up with the next wave tide. The Internet opportunities are not over, how can artificial intelligence end? So the statement of foam is not available.
Does that mean all ai companies have no foam? Also not, now ai company many valuation too high, non ai company also take ai’s aura to fool investors money, and many not understand investors really voted, these phenomena also exist. These phenomena are tired of jī jiǔ and definitely need to be leaked and let it return to normal state before going down. That’s why it will be seen after a successive ups and downs, but in the long term, the next 25 years will definitely be an optimistic trend.
You guys do ai, also have to have a learning attitude with the times, like if you enter the network 25 years ago, you said you’ve been doing the streamer for a lifetime, then it’s not working later! As long as you get in with the times, keep anti-Review, keep learning new knowledge, any industry is without foam. For example, ho peng, who did the UC flow at that time, what are people doing now? People make cars, right? But for him, the car is an extension of the internet!
To make a little more ad, you will see again after another 30 years, and the ai company that we have invested in the innovation workshop must be the highest in the industry, and of course the failure will definitely have. So a really knowledgeable vc, he can see clearly.
Finally to post a trailer:
After Deecamp, at the end of August, the innovative workshop joint search dog, the us review, the ai challenger2018 global ai challenge, launched by the United States, will also pull the curtain. Starting from early September, we will launch the “campus line” event at many universities across the country, which will have the top big curry in the ai industry to come to the school, interact with the students; there will also be an ai of contestants and lovers in all parts of the world. Technology exchange event oh. Specific schedule information will be announced this month.
Via Dr. 李開復 Kai-Fu Lee
