Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

 


Forgot your password?
Close
typodupeerror
×
Businesses AI

'Godmother of AI' Builds $1 Billion Startup In 4 Months (qz.com) 57

Posted by BeauHD from the tight-lipped dept.
Dr. Fei-Fei Li, the so-called "godmother of AI," is working on a startup focused on developing technology capable of human-like visual data processing and advanced reasoning. According to the Financial Times (paywalled), the startup is called World Labs and is already worth $1 billion. Quartz reports: "Curiosity urges us to create machines to see just as intelligently as we can, if not better," Li said during a Ted talk in April. "And if we want to advance AI beyond its current capabilities, we want more than AI that can see and talk. We want AI that can do." Andreessen Horowitz and the AI fund Radical Ventures are funders of World Labs.

Li is renowned for her contributions to AI. She invented ImageNet, a dataset used for advancing computer vision that many see as a catalyst for the AI boom. She consults with policymakers as they work to set up guardrails for the technology, and was named one of 12 national AI research resource task force members by the U.S. White House in 2021.
This discussion has been archived. No new comments can be posted.

'Godmother of AI' Builds $1 Billion Startup In 4 Months More

'Godmother of AI' Builds $1 Billion Startup In 4 Months

Comments Filter:

  • Godmother (Score:3)

    by phantomfive ( 622387 ) on Wednesday July 17, 2024 @08:14PM (#64633995) Journal
    She got her degree after the second AI winter. It seems she's a little young to call herself a "godmother," no matter how good she is. That's a marketing term.

    • ... is basically just saying "She scraped a bunch of images off the internet and threw some grant money at Mechanical Turk to have underpaid foreign workers label them." I mean, yeah, ImageNet was widely used, but it's a pretty weak boast. I make AI datasets; anyone want to throw me a billion dollars? ;)

      • I make AI datasets; anyone want to throw me a billion dollars? ;)

        Sounds like you need to find a sufficiently marketable nickname if you want that. Rei is just not going to cut it if for that sweet VC funding.

      • Re:Saying "She invented Imagenet"... (Score:5, Insightful)

        by sg_oneill ( 159032 ) on Wednesday July 17, 2024 @09:28PM (#64634119)

        Thats an unreasonable thing to say.

        Imagenet WAS important, but its far from her only contribution, just the one the mainstream press know about. She's got over 500 published papers (many of whom admittedly are just co-author, but in any metric thats a huge volume of research) and over quarter of a million citations. She's a giant in the research field.

        Between her work , the Transformer , and Hintons work on backpropagation , we are where we are today with visual ai. Terms like "godmother of AI" are nonsense journalist words, but to say she just scraped the web is to seriously misunderstand the progress of the field and what her role in it was.

        • I have heard the complaint "all she did was make a database" many times at conferences, always from guys who wish they were as successful. Congratulations, Rei, for breaking the mould!

          I have literally never heard the same complaint made about the PASCAL challenge (before the guy committed suicide; wouldn't expect complaints after), if anything he was generally spoken of very well.

          many of whom admittedly are just co-author

          I mean sure, you can't write that many yourself. Many are last author, which is general

          • Re: (Score:2)

            by Rei ( 128717 )

            Nobody just threw a billion dollars at Everingham and said he deserved it because he scraped a bunch of images. If they did, I would have been just as critical.

            Also, you referred to the GP as me. I wasn't the GP.

            And anyone who's worked at a large lab knows that you stick the boss's name on every paper you do even if they did nothing at all to contribute to the project. Acting all shocked that there's 500 published papers with her name on it, it would be shocking if there weren't.

            What actual innovations in

            • Even if the advisor just puts their name on the paper, it serves as a good signal, like branding, to attract attention to said papers. Unfortunately there are too many papers for the little time we have. Most of them are almost completely ignored after publication.

            • Nobody just threw a billion dollars at Everingham and said he deserved it because he scraped a bunch of images. If they did, I would have been just as critical.

              Don't be an ass. That's a blatant misrepresentation of what I was saying and you know it.

              Also, you referred to the GP as me. I wasn't the GP.

              Were you not? You made a mean spirited jab (one I've heard countless times before at conferences to the point that it irritates me. And I don't even know Li Fei-Fei. I think we've met in passing, but that's abou

              • Re: (Score:2)

                by Rei ( 128717 )

                You know, before you went off on your "your views must be because you're sexist rant, it couldn't be because you legitimately lack respect for her work", you might have thought to ask of the gender of the person you're talking with.

                You have a chip on your shoulder. You should recognize that chip before you blow up at random strangers.

                Good grief you are obtuse. She's done a ton of stuff, much of it very good. One of them went viral. That doesn't mean she didn't do anything else.

                Did I say she "didn't do anyt

                • Re: (Score:2)

                  by Rei ( 128717 )

                  I want to be clear: I'm not saying anything bad about her. I have no ill will toward her. She has unquestionably made very useful contributions to the field. The field is better off for her work.

                  But praising ImageNet as some sort of brilliant innovation, some clever idea, is just plain stupid.

                  Many people have come up with brilliant innovations that came out of left field and led to huge revolutions in the functionality of models. She was not one of them. She didn't invent something like backpropagation or

                • You have a real problem with reading, huh. I don't know if at this point you are incapable of understanding or if you are intentionally taking the most absurdly extreme misreadings that you can for entertertainment.

                  I can't be arsed to go over it point by point, so I'll pick out my favourites.

                  you might have thought to ask of the gender of the person you're talking with.

                  I already congratulated you for breaking the mould. You might want to try reading things before replying to them rather than after/never.

                  Oh h

                  • Re: (Score:2)

                    by Rei ( 128717 )

                    And then I asked it about what look to me about the weird inconsistency in its answer and it completely fell over.

                    I don't even know what you're talking about. There was one discussion with one person - whether you or someone else - where the person made the logic problem more complicated, but not only did the AI miss one possibility in its analysis of the problem, but so did the other person (incorrectly believing that "some" could equal "none"), and so did I (the same one that the AI missed). Aka, putti

                    • I don't even know what you're talking about.

                      Clearly.

                      There was one discussion with one person

                      There was not. This is a discussion forum not a private messaging service. Anyone can weigh in at any time, so conversations can be with multiple people.

                      in a more complicated one" does not in any way equal "cannot do logic" regardless.

                      Could you give it a rest, eh? I made no such claim that they "cannot do logic". Again, that was a different person. Seriously how do you not understand the basics of a public discussio

                    • Re: (Score:2)

                      by Rei ( 128717 )

                      Could you give it a rest, eh? I made no such claim that they "cannot do logic". Again, that was a different person.

                      You're sure enjoying talking about what you're *not* claiming here, but it sure would be nice if you'd instead talk about what you *are* claiming. What exactly is your viewpoint about the capabilities of LLMs that contradicts mine?

                    • You're sure enjoying talking about what you're *not* claiming here

                      Well since you keep inventing claims I made, it's kind of you driving the conversation relentlessly in that direction.

                      What exactly is your viewpoint about the capabilities of LLMs that contradicts mine?

                      Is it worth me replying here, given you don't actually appear to read my posts, instead only replying to what you feel I wrote. Still sucker born every minute so here I go...

                      I think you are overly optimistic about their abilities in solving log

                    • Re: (Score:2)

                      by Rei ( 128717 )

                      If you don't consider "entirely fictional terms not in the dataset, shuffled, and with distraction sentences" to be "general purpose", I'd love to see what you consider general purpose.

                      It was able to filter out distractions and apply reasoning like younger implies worse. Recent bad event implies worse. Prodigy implies better. And combinations make the scores go up and down.

                      What exactly do you think you're doing?

                      We both agree that they have weaknesses. Which isn't exactly breaking news. What I don't see is

                    • If you don't consider

                      LLMs can do logic in the same way they can play chess. Which is to say they can basically splat down stuff that's close enough but beyond a certain point you deviate too far from the training data the rules add up and really matter at which point they just spew look-alike things which don't make sense.

                      You can get them to play chess successfully for quite a few moves, interspersing irrelevant chitchat too. But they don't know how to play and eventually make illegal moves or resurrect pi

                    • Re: (Score:2)

                      by Rei ( 128717 )

                      I have to say, you seem awfully angry and confrontational with me over a topic that I increasingly think that we largely agree on.

                      (MoE?)

                      Surely you're not talking about this topic without understanding what a MoE is (Mixture of Experts). Link [wikipedia.org]

                      There are existing techniques to identify or prevent hallucination, generally involving many runs with different starting conditions or other sources of randomness, and then comparing outputs or hidden states. This is, obviously, very slow. MoEs however run multiple e

                    • Re: (Score:2)

                      by Rei ( 128717 )

                      (I actually modified the Mixtral code at one point to do this, but I don't exactly have the grant funding to train my own foundational model. Might try with a small toy model at some point..)

                    • You know, I think the reason you find LLMs impressive is they compare so favorably to you. Yes I am being rude, and yes you thoroughly deserve it.

                      You keep on hallucinating opinions that I don't hold but are in your training set (I note you have never once admitted a mistake or apologised for claiming I said something I did not). You don't appear to understand how public discussion forums work. You have a about as much problem following reasoning about probabilities as LLMs. And you take bizarrely extreme mi

                    • Re: (Score:2)

                      by Rei ( 128717 )

                      That is a standard acronym in the field, it is extremely common, the acronym is widely used in the literature, and the fact that you don't know it speaks volumes as to how (not) familiar you are with the current state of the field.

                      You've also continuously been an arse, spent post after post refusing to state what you actually believe but damning me for not guessing it based on ancient conversations, and I have better things to do with my life than engage with a person like you. End of thread.

                    • Coolio, see you on the next thread.

      • Why didn't you make the dataset that started the deep learning era if it was so easy? (hindsight is 20-20)

        • Re: (Score:2)

          by Rei ( 128717 )

          I dunno, it might have had something to do with... and I'm just speculating wildly here - not having an AI research position at Princeton and extensive grant money in the mid 2000s?

      • ... is basically just saying "She scraped a bunch of images off the internet and threw some grant money at Mechanical Turk to have underpaid foreign workers label them." I mean, yeah, ImageNet was widely used, but it's a pretty weak boast. I make AI datasets; anyone want to throw me a billion dollars? ;)

        You would be worth a lot of money now if you had the vision to start creating ImageNet 18 years ago. Sure, it's not a big deal now, but back then it wasn't obvious. And throw in the challenges of creating a good benchmark. It's easy to create any old benchmark but hard to create one that creates a distribution of results with sufficient challenge for all candidates. Oh, and also have the organizational and marketing ability to get that benchmark noticed and to create a contest that people care enough to

    • She got her degree after the second AI winter. It seems she's a little young to call herself a "godmother," no matter how good she is. That's a marketing term.

      How dare you, she's a strong independent woman just like Elizabeth Holmes.

    • But in tech's cutting edge, you're considered a geezer at 35.

    • Well, technically one can be a godmother or godfather at sixteen years of age (Roman Catholic Church). A grandmother or grandfather on the other hand...
    • You just don't know Fei-Fei Li. She is an important force in the current era of AI, almost as notable as Hinton and LeCun. She was behind ImageNet - the dataset that kickstarted deep learning, she directs the Stanford AI Lab and was Andrej Karpathy's teacher.
      • No shit. Glad you read Wikipedia. She's still not the godmother of a field she is much, much younger than. That's a hype name.

      • I don't have a problem saying they made significant contributions. My problem is when they call them godfather/ godmother of AI. Ai existed a long time before so did neural nets. A lot of people have taken AI to where it is today and it's kind of saying ai didn't exist before

    • She got her degree after the second AI winter. It seems she's a little young to call herself a "godmother," no matter how good she is. That's a marketing term.

      She didn't call herself the "godmother of AI." That's a term created by others. In fact, she has voiced [princeton.edu] discomfort with the label: "Somewhere along the path of her pioneering career, experts in her field began to refer to her as the godmother of AI as a tip of the hat to her groundbreaking contributions. It is a title she didn’t immediately embrace. 'I would never call myself that,' she told CBS News last year. 'I don’t know how to balance my personal discomfort with the fact that, throughout

  • > She consults with policymakers as they work to set up guardrails for the technology

    So she helps congress build a better mousetrap, then goes out and builds a better mouse. I love it.

  • Who fucking cares
  • A sufficiently wealthy group of people have selected someone to stand in for them as the face of a new AI company. They're marketing her as the 'Godmother of AI' and have started the venture with a whopping one biiiiilion dollars.
  • Human-Centered Artificial Intelligence at Stanford with a mission to "advance AI research, education, policy and practice to improve the human condition". If anyone would build a humanity-benefiting AI it would be her. You can read her autobiographical book "The Worlds I See", https://www.amazon.com/Worlds-... [amazon.com] yet the 2-star review there is worth a read,

    The fact that the culmination of her work led to corporate interests at Google and Microsoft is no indication that - beyond her personal upbringing - she

  • Raising $1bn in 4 months (Score:5, Insightful)

    by Rosco P. Coltrane ( 209368 ) on Wednesday July 17, 2024 @10:57PM (#64634255)

    If that doesn't scream "bubble", I don't know what does.

    It also means when the bubble bursts, the company will lose those $1bn in 4 hours.

    • No bubble, or not more than the dot-com bubble, which was quickly followed by unprecedented online growth and major profits. LLMs are too useful to be just a fad. I use them everyday both for work and fun.

    • Re: (Score:2)

      by Bumbul ( 7920730 )

      If that doesn't scream "bubble", I don't know what does.

      It also means when the bubble bursts, the company will lose those $1bn in 4 hours.

      No, the company itself doesn't lose anything. That valuation bubble is with its owners, who would lose those imaginary funds should the valuation drop.

      • Yes, exactly.

        In 2001 I was working for a startup that was at the tail end of the dot-com bubble. It raised $300M before the company folded. Some of that money came from regular people, like a coworker who had saved his money and paid off his house. He sold his house to put the money into the company. It was completely wiped out.

        I don't care about the losses for the investors who have more money than they know what to do with. It's the little guys like my coworker, that really get hurt.

    • The 1 billion dollars never existed. See also This video [youtube.com]

  • This crap is now a megabubble and a real threat to world economy.

  • It seems significant that Fei-Fei Li is looking to build a new generation of AI tools that do something different than everyone else. Labeling data was a relatively tiny thing before she set up legions of Mechanical Turkers to create Imagenet, which planted the seeds for real breakthroughs in new CNNs for image processing like Alexnet and started opening people's eyes to the potential of scaling more automated AI infrastructure. She also comes with this from a sense of ethics, which seems sadly lacking the

    • Indeed. AlexNert felt like a key tipping point. The state of the art was infinte variations on SIFT and SVMs with the pyramid match kernel and neural nets were deeply unfashionable old and busted things without fancy maths from the 80s.

      AlexNet blasted onto the scene and showed definitively that ANNs were back and the old 3-LP was dead and convnets were where it was at. But of course that couldn't have happened without ImageNET.

      And for all the "oh it's just a dataset" crowd: well why didn't you do it. She al

  • I raised 3 Billion !! (Score:3)

    by TTL0 ( 546351 ) on Thursday July 18, 2024 @02:30AM (#64634483)

    I built an AI App based on Blockchain running on a Beowulf cluster

  • This is [aminoapps.com] Faye Faye [youtube.com].

  • As a pre-check in for a medical office visit on website, a box had whether I would allow AI to handle my info. NO, hell no !!!
  • No one can build machines or write softrware that is capable of 'reasoning' because we have no idea how that even works on a fundamental level.

Slashdot Top Deals

Children begin by loving their parents. After a time they judge them. Rarely, if ever, do they forgive them. - Oscar Wilde

Close