How AI Startups Must Compete with Google - Dr Fei-Fei Li + Mike Abbott
ฝัง
- เผยแพร่เมื่อ 4 ส.ค. 2024
- SUBSCRIBE: / startupgrind
STARTUP GRIND: www.startupgrind.com/
GLOBAL CONFERENCE: www.startupgrind.com/conference/
Mike Abbott, Partner of Kleiner Perkins Caufield & Byers, sits with Dr. Fei Fei Li, Associate Professor in the Computer Science Department at Stanford, and the Director of the Stanford Artificial Intelligence Lab and the Stanford Vision Lab..
Dr Li's main research areas are in machine learning, deep learning, computer vision and cognitive and computational neuroscience. She has published more than 150 scientific articles in top-tier journals and conferences and invented ImageNet and the ImageNet Challenge, a critical large-scale dataset and benchmarking effort that has contributed to the latest developments in deep learning and AI. As of January 2017, Fei-Fei Li is spending her sabbatical at Google Cloud as Chief Scientist AI/ML.
EUROPE CONFERENCE: www.startupgrind.com/europe/
TWITTER: / startupgrind
FACEBOOK: / startupgrind
INSTAGRAM: / startup - วิทยาศาสตร์และเทคโนโลยี
You democratise AI if and only if you democratise data.
Lots of good data streams in the public domain - what sort of data are you hoping? Google search library?
My comment was not meant to point directly to a "data gap" but rather to the obvious. That the "complexity" of ML techniques is not the entry barrier, data availability is. While I'm very grateful for Prof. Li democratising knowledge, I claim that it is not knowledge but data that needs democratisation.
And Prof Li has already done that image-net.org/ before you ask.
Post information age, it's not a problem. You can stick a camera out a window and collect buttloads of data.
I wouldn't look a gift horse in the mouth. ML requires not just data and model / algorithm knowledge, but also compute. Not many people can afford 16 GB gpus to train big transformer models. Google Colab is a godsend. Tons of opportunity out there for the taking.
I like that youtube shows these interviews.
One of the best talk I have ever seen!!
Thank you
Probably the best interview I’ve watched this year
The issue about automation taking jobs is business thinking has changed since 50/100 years ago when machines or ATMs were introduced. Companies actively look to automate and reduce headcount in a world of unregulated inflation and instability.
Superb content thank you for posting
This is a fantastic talk. Thank you!
Awesome video bro, keep it up
nice discussion
Word of advice .. .watch on 1.25x! Also, notice the subtle sales pitch - google makes money from selling ML api services. So what fei is really saying is don't compete, just pay us money to do it for you. That's fine - google has to make money - but it doesn't touch upon the issue that these API's aren't really that great and robust. Furthermore, they will be misused immensely since the people using them won't have a clue about statistics and machine learning. The API's are just passthroughs to pre-trained ML models. ML models need to be retrained on YOUR data and no 2 datasets are the same, so that is why API's are doomed to failure.
Just want to add that Fei-Fei Li was "only" on a sabbatical visit into the industry. She is a researcher by heart and care more about science and development than profit. Second, she is behind the team which made ImageNet a reality. It's the worlds biggest image data set which is FREE for everyone to use. So please don't try to present her as a Google-disguised data devil ;).
+Simon Buus Jensen.
But she is. She is an employee of Google.
@Coco Cat
Love it!
I like it that +Startup Grind (there's no tagging in youtube) is active in the comments, way to go :)
We do what we can :)
It all comes down to finding the right market but even once you get to settle in, if you want to scale you get into the difficult part which is finding the talent. Fun Times!
AI/ML talent is SUPER hard to find outside of major tech corps now, too. Hard to get an edge here.
I greatly enjoy the Hitchhiker's Guide to the Galaxy series. In the first book, Deep Thought was given the task to find the answer to the ultimate question of life, the universe, and everything and after 7 and a half million years of calculation gives an answer of 42. But it turns out Deep Thought doesn't know the ultimate question and offers to design Earth, the computer; to calculate it. One of the ideas that really fascinates me here is the idea of finding(computing) and answer without knowing the question. And in a sense in machine learning, we are trying to find an answer without specifying the exact question. The thing that is really interesting to me is, how can we design two Machine Learning systems so that while one is doing the work, the other can OBSERVE(whatever that means?) the former while it is at work. So in a sense, a second Machine Learning system would learn the learning behavior of the former and build an estimate for the question. Using such a pair of two machine learning systems, how close are we to asking the first one to say solve a differential equation(without telling it how to solve it), it uses learning mechanism to solve the differential equation, and based on the behavior of the first one, the second one tries to configure the steps in solving the problem, here the differential equation. In summary, how close are we to designing machine learning algorithms that write other algorithms or programs? I think this is really exciting and is followed by a load of questions about AI and society.
thank you for this video. great talk 👌
Any favorite parts?
Yes!
I liked her particular focus on the emotional responsibility that data and ai scientists must bear with regards to the human (or more broadly put, "non-artificial") aspects of their work. We must empathetically frame our solutions to the world's problems to avoid the "doom-and-gloom" scenarios put out by skeptics. You can tell that her talents are not just built on a deep (😉) knowledge of the field, but also upon a genuine concern for the world and a desire to understand "why" -- not just "how" -- things work.
Thank you for this content, it was truly appreciated.
smart scientist... proud of you!
I am a ML enthusiast but I think that there is huge lack of labeled data that prevent ML from being mainstream and within reach of every scientist
True. But it is a direct result of the limitations of the ML algorithms as such. One can use multiple models of reinforment/adverserial learning to self-label the remaining data thus improving. Some would prompt humans to do the labelling but it will be the last 0.01%. Some companies has been using human to label video clips for some time (ie paid). Also, LeCun said, developers ought to develop interpretability into the learning models to ensure we can understand the way the layers bottlenect solution determination in its final result.
I trust AI, but I don't trust military AI.. Hopefully more people will work for the benefit of humanity.
Yes to that!
Military AI Is pretty insane. You can have flying drones (even small qyrocopters) today that operate machine guns. This could easily be used as low-risk&low-cost terrorism.
"Gun not kill people, people do". Isn't that is very similar idea?
I don't trust ANY AI (or any technology for that matter). I don't trust the military, I don't trust the government, and I don't even trust the public sector. The only answer is constant vigilance and the ability to articulate the warnings into the court of public opinion (subject to the constraints of government). Our government agencies have shown that they will bypass legal procedures and will intercept private communications to use against their "enemies". Google has shown that they will discriminate against their employees for their views. There is no longer a rule of law: we are now in the age of rule by political will.
ya and maybe video game companies will give up on microtransactions too.. ummm probably not
I like how people think that AI will not crush the number of workplaces. I mean come on, banks may have increased the number of customer service staff, but that is probably because many more people use banks. In my country, one bank let go 3/4 of its staff due to automation. They were making profit before and now that profit is increasing.
How can start-ups manage to collect the data? It would be nice if someone talks about this.
Never really discussed in depth the topic in the headline "How AI startups must compete with Google" because she really did not much to offer to such startups.
How Google and other tech giants buy AI start-ups.
dude talks to much, let the genius finish her thoughts
Eh...There is an increase in female science nobel prize winners these years.
Should hear her talk at her own event. She is super nice and humble. She lets others who don’t know much talk more and interrupt because she sees it as a way to flow in conversation.
We already have big data and algorithms to create AI. But an AI that can respond instantaneously requires the CPU speed at least twice of the Watson to make it efficient and practical in the real world. The Moore's law is running out and there's about 10 other ways to create a faster chip. So, it's the speed of the chip now.
? Sounds like a lot of buzz words.
dr fei fe you will be a thought leader in the future / similar to jack ma and jeff bezos. awesome story ! but what i am seeing is the chinese immigrants are too into the gaming and frizolous focus. awesome and inspiring ! thx you for your background !!
if you can't beat em join em
No doesn't mean that
more like -fight em until they offer you a truckload of money
Good way to get (a) an inside track on where the opportunities like and (b) meet all the talent to start your next Co.
All Machine learning experts would tell you that post ML we can focus on more creative tasks rather than repetitive ones. How about a single ML researcher replacing jobs of say 50 feature engineering guys who specialized in their domains? So this is not just about replacing repetitive jobs. Also if all the heavy lifting is done by Google, the benefits to those making products on top of Google platforms would be minimal coz of DEMOCRATIZATION.
what in the world are you talking about????
If a single ML researcher can replace 50 engineers, a lot of engineers will be put out of work and ML researchers will be in demand. I'm not sure where this fairy tale of less jobs leading to greater distributed wealth comes from.
Being a picky person, I can't help but notice that she didn't exactly answer the first question. But still an informative talk.
Political training, maybe?
These people are the ones taking humanity forward
Google didnt need to democratize their AI. They are light years ahead of everyone else
But they thought of the greater good of mankind
That is why I love Google.
Any other parts of their work you wish they'd democratize?
I dont have sufficient knowledge about all Google products to answer that question
Intellectual Property Theft: NOW ON LIFE SUPPORT: There is intensive mind control taking place using AI maliciously through satellite and biochemical warfare by supremacists, terrorists and communists. Please report to the highest level of governing bodies and intelligence agencies. I have been poisoned, harassed physically and mentally, raped by lesbians and ignored by the police, agencies and churches nationally as have my children. I and am under constant attack from my government and international WOKE military
这哥们坐姿有点。。。
Progress Is a marketing snake, it's an apple that is necessary evil you cannot do without.
Without you are gates or Einstein ....the world is a playground.
Sitting cross-legged on a couch onstage just looks weird on so many levels
You mean the man?
It may not be the best idea to democratize AI, maybe the controlled usage, but not the entire AI. Just imagine how dangerous it could be if everyone could have access to the full power of an AI.
News flash - everyone already has access to AI in the cloud, everything from pre-trained models to Image/Text/Speech recognition as a service. And it is not all that expensive either - couple hundred bucks for a moderate workload.Then there are open-source platforms that let you run deep learning models on your laptop or cluster of high-performance GPU nodes. AI Hammer already exists and is for rent, sale or plain-ol-free. You just have to learn how to swing it.
Among the two people in the discussion, I found the non-native speaker (dr fei-fei; my assumption) was easier to understand than Mike (likely a native speaker). He mumbles too much and swallows words.
I'm (almost?) a native speaker of English from India, just for context :)
as a non native it is normal to understand non natives better.
not even a minute into the video and I was already annoyed by how the guy dragged his words, I thought I was the only one then I saw the comments :)
so whats your level if you understand both? AI level?
yea. i did not read the comments yet, but what in the world is that guy doing? He does not complete his sentence. He breaks it and make another. He is rambling. He should just hide behind the computer.
Who is native speaker of English in India ? ... It's not an Indian Language ....😀😁😂
Yeah yeah google google, we have to start talking about BAIDU
What should we know about what they're working on?
Everything... Considering this is a platform powered by Google ( ͡° ͜ʖ ͡°)
This isn't exactly a critical peek at the difficulties inherent in A.I./M.L. It's rather a rosy story that I've heard before.
Intended for a general audience, rather than a technical one, yeah.
Really I shouldn't have said anything, but I was disappointed. It's really hard to find interesting material on current A.I. I've looked at the Stamford lectures which are great, played with speech recognition, perused the Watson API, etc. It's not really all that relevant, however,l and imo, with the sort of general a.i. that interests me.
For real? Speaking of democratizing AI by using google's cloud? Giving all our data to google for free?
Eugen Barbula yeah, and who knows what they are going to do with it in the future...... eek
Let the women talk dammit
MS.Lee should have been put into investigation regarding what her role and what have she done of helping US enemy China in AI. STANFORD should have stopped suspects to touch any research and helped US government for the investigation to prevent stealing of US technology and put the traitor into jail if proved
It is extremely irresponsible and close minded for her to suggest that every student become a professor and press them into this career.
Chinese girl had fallen so much from last 30 years. What a shame to see such tragic happen in today's time. Science without humanity is the worse.
why am I watching this crap!
Does she really want to democratize AI coming from China ?
Why all AI experts are Asians?
Do they count as a minority? I think colleges get more money for having a lot of minorities.. Plus a lot of American Asians work (study) hard.
It is hard and required a lot of work
Asians are pushed towards excellence with no room for settling for good enough. Thats why they excel, there is always more to do.
They aren't.
Jure Kralj you're racist af.
Is this woman a CCP plant? I think Peter Schweitzer thinks so. She certainly has ties to the CCP.
Her approach is to use a brute force method .... again. They are running out of ideas as to how to approach the problem. In her other talk she feeds enough picture of cats and hope it covers every perspective and when a picture is presented in front of the AI-camera it will say "CAET". Prof. Li, place your hand in front of the AI-Camera and meow .... your program fails. What? you say it is about computer vision only? Well, you mentioned a child goes through million of images by age 5 ... right. Well the child brain synthesizes smells, hears and see ... so what you have done is taken a child who cannot hear, nor smell, nor touch. You can place bunch of high achieving academics who can only achieve this if they give up on the World and expect them to produce something that is a hit with people in real world is wrong. I think revolution will be out of US not within US so GOOGLE can bust its behind develop enough golf balls and it will not be a diamond.