Video Of The Week: Matthew Zeiler at Code Commerce

Matthew is the founder and CEO of our portfolio company Clarifai.

Here is his talk (11mins) from last week’s Code Commerce conference in which he talks about how to use AI to grow your business.

#machine learning#Uncategorized

Comments (Archived):

  1. creative group

    FRED:The best AI says you would be a better owner of the NY Knicks than the current inept James Dolan. On behalf of KnickNation we plead you to assemble an ownership group together to bid 2.5 Billion with related Madison Square Garden properties and rights for the NBA New York Knickerbockers.Thanks you on behalf of KnickNation!Captain Obvious

    1. fredwilson

      What makes you think he would sell?

      1. LE…Before being caught up in the future downward trend, Dolan could sell the Knicks before the market for NBA teams collapses. The future value of television rights for NBA games could drop as people move toward streaming services and away from traditional bundle packages. Eventually, that could erode the value of NBA franchises, like the Knicks. James Dolan may be a bad team owner, but he is not a dumb businessman.

        1. sigmaalgebra

          IIRC, Dolan sold his cable business to some European company.Then, IMHO, now he is just pursuing personal interests! There’s a history of that, e.g., from Steve Ballmer, Paul Allen, Ted Turner, Mark Cuban, maybe more!

          1. LE

            I say men get tired of their toys. Also typical toys don’t open you up to everybody constantly complaining and criticizing your strategy when from what I read it’s a losing one and the complaints are universal.

          2. sigmaalgebra

            Ah, “The only difference between men and boys is the price of their toys!”Life is a competition, and “the one who ends with the most toys wins” — IIRC from one of the Forbes boys!

          3. creative group

            LE:When any owner hires a HOF Player in Isaiah Thomas and HOF Coach in Phil Jackson to destroy the entire organization who have never held an executive position on any level he doesn’t deserve criticism but the team sport taken away.

        2. PhilipSugar

          I agree completely.If tech companies get their wish of a la carte cable the value of NBA teams will plummet.ESPN who made a big bet on the NBA already has plummeted.I love sports. If you told me I had to pay one penny to watch an NBA game? Nope.If you told me I could get rid of ESPN and discount my cable bill by the subscriber fees? Done.

          1. sigmaalgebra

            A penny? I’d spring for a penny IF I could download HD video with expert commentary and could play the crucial seconds in slow motion down to one frame at a time, with zoom in — all so that finally I could see what they heck they were doing instead of just the blur. And for the announcers, have a separate track for the fictional drama majors, a track I could cut off!!!!

        3. creative group

          LE:Many would beg to differ on that opinion of James Dolan’s business acumen. His father was the scion that was a respected business man who started from the bottom. James Dolan only capitalized off his dad’s hard work. (A fact)

          1. LE

            A guy like Dolan gets to spend a great deal of time and be friends with people that a regular guy can’t. As such he is getting a great education and advice from people that many could only dream of being able to deal with. So while he might not be a great intuitive business person and was able to start with a large nest egg he does have a great deal of advantages that would allow him to make the right decision.I would rather be able to move in the right circles than have 15 times the bank account. Would be easier to make money that way, all else equal.

          2. creative group

            LE:Your point you made you unmade it with the response.If an idiot hangs out with mathematicians it doesn’t make the idiot good in math because he hears mathematical equations bring discussed.

      2. Vendita Auto

        Just learned how to get a top VC’s attention : )

        1. creative group

          JaredMermey:Our request isn’t based upon conjecture and doesn’t rely upon which actually uses sources some of us know personally. There is no need to name drop. We leave that for the jock huggers.

      3. creative group

        FRED:Dolan is more interested in becoming a famous traveling blues player. The right ownership group offer would be entertained. The majority who have the wherewithal to make a qualified offer will not make one because they are waiting for Dolan to offer the Properties for sale. Dolan would not offer them because he doesn’t need the money. An offer needs to be presented by the right ownership group.Fred would be part of the right ownership group that Dolan would respect.

    2. Tom Labus

      Nets better buy and are forsale

      1. creative group

        Tom Labus:Nets maybe a better financial buy but far from what is considered Marque. The LA Clippers were overpriced.The Brooklyn Nets will never be the New York Knicks. (It is what is called history in NYC)

    3. jason wright

      knickerknocker 🙂

    4. JaredMermey

      Would cost way more than $2.5B

      1. creative group

        JaredMermey:The way the Knicks, Rangers and MSG would be sold is unknown. In part or as a whole. Knicks and Rangers would be a packaged deal for sure. MSG and network would be another story. 2.5B would be the starting bid.

  2. Joe Lazarus

    Clarifai is so cool. Better results than nearly all their competitors in my experience, aside from Google for certain images. Impressive considering their competitors are some of the biggest companies in the world with stellar AI talent.

    1. Lawrence Brass

      Clarifai advantage is focus.

  3. Vendita Auto

    Nice simple presentation,Reminded of the recent comment: (AI) “Geoffrey Hinton said that, to push materially ahead, entirely new methods will probably have to be invented. “Max Planck said, ‘Science progresses one funeral at a time.’ The future depends on some graduate student who is deeply suspicious of everything I have said.”

    1. sigmaalgebra

      Yes, there can be some utility in a combination of (1) learning the old stuff so are able both to appraise and use it and then (2) having enough contempt for the old stuff to push, also building on some of the old stuff, for some new stuff.But using any such stuff has a tricky step. The secret obstacle is that can’t just “use” the stuff. Instead, have to regard the stuff as just tools, make a list of problems to be solved and tools available and, then, find a PAIR of problem and tool that work — and work in all relevant respects. Quite seriously, it’s the pair that really matters.Sometimes can relax that pairing a little and find a pair of a problem and some original research can do to find a tool so that the pair works.

      1. Vendita Auto

        True, but so often second mouse gets the cheese

        1. sigmaalgebra

          That’s why need some good barriers to entry! E.g., at times original research, regarded as secret sauce, trade secret, proprietary intellectual property, can provide that.

      2. Lawrence Brass

        So true and simple yet so hard to understand and practice.A cloud of abstractions can make you blind.You can also build a new tool to solve an old problem or just build a tool and see if other people pairs it with a problem.Taken to the extreme, you can also build a problem and then come out with a proprietary solution.

    2. sigmaalgebra

      Ah, in Planck’s time, a lot of physics profs were convinced that the world was based just on billiard balls and water waves.Then thinking about the photoelectric effect would ruin more than one good dinner with wine, beer, whatever! Planck black body radiation — real gut twister! Then thinking about Young’s double slit would ruin more — need the liter size bottles of Pepto-Bismol! Hearing that the experiment also works the same with just electrons sent one at a time — “has to be some mistake!”. Generalize that just to antenna theory and could get even more confused! No way do they want to think about the Michelson-Morley experiment — could mess up a whole week thinking about that! Even thinking about just molecular spectroscopy could ruin good times. Then, take some hydrogen atoms and weigh them and then weigh some carbon atoms, count the electrons, neutrons, and protons, and try to see where the extra mass went? Bad afternoon! By the time of the EPR (Einstein-Podolsky-Rosen) paper, time to give up on physics! Where that 3 degree K background radiation came from and how it could be so uniform — just ignore that evidence! Gamma ray bursts — gotta be just a noise signal from some arc welding shop down the block! And those GPS satellites with clocks that tick a little too slowly — again, “gotta be some mistake!”. Then hearing about Bell’s inequality, they’d be glad they retired from physics!

  4. LE

    Interesting video. I think Matthew needs to lead with the fun product demo that he ended with. Just 30 seconds to draw people in and create curiosity. This is what TV has been doing forever. (They also do it before a commercial so that you don’t change channels).Here it is:

    1. PhilipSugar

      The authority on this is Peter Cohan:…Now it is just a simple “don’t burry the lead, start at the end”But it is also why it is tough to give good demo’s if you understand the technology too much.People don’t care about that they care about the end.Now do not get me wrong. There is nothing that makes me more angry than hearing, I assume the technology just works, that is table stakes. See how many people have lost their job doing that.

      1. someone

        true, though I found it refreshing that the speaker actually had an idea of what he was talking about.and, given that his point was not “AI is cool” but this-is-too-hard-for-you-to-do-in-house, I thought the lead-in was appropriate

      2. sigmaalgebra

        It is strongly in the interest of the MSM to slight, denigrate, omit, etc. anything like solid evidence, e.g., as is rock solidly standard in nearly all actually important work, e.g., pure/applied math, physical science, bio-medical science, engineering, high end finance, US national security, and more.We should not let the sick-o norms of the MSM leak into startup communications.E.g., the times I got the startup advice on the Internet on how to communication, I noticed howdifferent their suggestions were from the college, grad school, research seminars I’d heard, the math and science texts I’d studied, the research papers I’d published, and the college and grad school courses I’d taught!

  5. sigmaalgebra

    So, his ML can recognize 11,000 objects. Hmm …. Maybe there is something useful for that to do, but (1) for the world 11,000 is not very many, (2) as I understand his ML, to recognize a new object takes a lot of input data, (3) for the applications he has in mind, e.g., helping and understanding people who are shopping, I’m not impressed by what such ML image recognition can do, and (4) I’d be concerned about his revenue model.For the applications he has in mind, there are other approaches that do not involve image recognition at all. E.g., IMHO for nearly all of shopping, keyword/phrase search is much more promising. E.g., yesterday I did a lot of searching with keywordsDDR3 1333MHz ECC unbuffered 8GBI’ve done a lot of shopping on the Internet, and I can’t think of anything where pictures as my inputs would be better than keywords/phrases.Yes, for research on image recognition, I can believe that there is progress to be made there. E.g., yesterday in the refrigerator I found a little plastic freezer bag with some broken pieces of slices of American cheese. So I put them out on the back porch, and the next time I looked all the cheese pieces were gone. So, some small animals, birds or squirrels, did some fantastic image recognition with, since likely they’d never seen American cheese before, no “training” at all! How’d they do that? I don’t believe that anyone knows! It’d be nice to know!To me the Clarifai image recognition looks like a solution looking for a problem and from the lecture has yet to find the right problem(s).I’d suggest:(1) He go to DC and chat with some of the US national security people who need to analyze images, e.g., recognize objects in images. There, sure, for the quality of his recognition, he will learn about the classic Neyman-Pearson result. If he wants a quite general proof, I can send him the one I worked out using the Hahn decomposition from the Radon-Nikodym result with the cute von Neumann proof! IIRC, at one time, Raytheon was interested in the Neyman-Pearson result! Uh, for a little more Clarifai might report both detection rate and false alarm rate and how to adjust the trade-off between the two!(2) Chat with the TSA about recognizing images.(3) Chat with people who have millions of objects — peanuts, apples, oranges, tomatoes, parts from casting, stamping, forging, or 3D printing, from mining, waste processing, bio-medical samples — e.g., maybe moving down a conveyor belt, and need to recognize and extract certain ones of the objects.IMHO, search, recommendation, and discovery via users entering keywords/phrases is terrific, beautifully polished, highly effective, with 1+ billion good users, the basis of Google, Bing, Amazon, and more yet is only for about 1/3rd of the problem and is at best poor in the other 2/3rds. For the other 2/3rds, the crucial, central key is the meaning of what the user wants. So far, AI and ML have no idea how to work with meaning, and to me object recognition in images seems too far away from the needed meaning for any useful step toward meaning.Are there useful, practical ways to work with such meaning? Yup. Based on ML, AI, neural stuff (like in the human brain)? Nope!

  6. creative group

    CONTRIBUTORS:The New York Knickerbockers trade Carmelo Anthony to the OKC Thunder. Hope he does well there. Let the rebuilt begin.