4 ms·

From all I can see, nearly all the present AI boils down to one word -- hype. From what is really going on, nearly all of it looks like (1) in some cases a lot

by graycat 10y ago

From all I can see, nearly all the present AI boils down to one word -- hype. From what is really going on, nearly all of it looks like (1) in some cases a lot of new data, (2) some new, faster processor hardware, e.g., based on graphical processor units although the x86 processors are astoundingly fast anyway, able to manipulate the new data, (3) for what manipulations to do on the data, some tweaks of some of the work of L. Breiman and his CART -- Classification and Regression Trees, (4) fitting with S-shaped (neural) sigmoid functions, and (5) the radar, etc. engineering of autonomous vehicles. E.g., the neural networks might be able to simulate the operation of a neuron in a worm. What I don't see is (A) much progress in better methods for how to manipulate the data, that is, the basic applied math and (B) progress in working with concepts and causality -- in the history of science, progress with concepts and causality did well where the new methods would need Nevada full of disk drives of data. E.g., space flight navigation is based, first, on Newton's law of gravity and second law of motion, not on fitting massive of data via deep learning. Uh, when the AI people use classic regression analysis as in Draper and Smith, etc. and the IBM Scientific Subroutine Package, SPSS, SAS, R, etc. to find some regression coefficients, they claim that their machine learned the coefficients. Gee, I didn't see that in, say, C. Radhakrishna Rao, Linear Statistical Inference and Its Applications: Second Edition, ISBN 0-471-70823-2, John Wiley and Sons, New York, 1967. or Breiman's CART. Again, the rest I see looks like hype. E.g., it appears that there is a basic, clever, publicity, hype idea: Whenever do some technical work, give it a catchy name. Then just use the name in the hype and ignore what is really going on in any math or science. E.g., a while back I published some work in statistical hypothesis testing that is both multi-dimensional and distribution-free. The intended use was for high quality hypothesis testing of zero day problems in computer networks and server farms. Alas, I neglected to give the work a catchy name! The OP seemed surprised that a big, famous company might say things where they know better. Why not? Just assume that they are trying to sell something to some people with money. Similarly for the news media: They want eyeballs for the ad revenue. That the newsies are willing to write junk to get eyeballs goes back at least to Jefferson's remarks as in http://press-pubs.uchicago.edu/founders/documents/amendI_speechs29.html http://press-pubs.uchicago.edu/founders/documents/amendI_spe... and are not nearly new. We have long had a good filter to apply to the writing of the newsies: Does the writing meet common high school term paper writing standards for solid references and primary sources? Rarely is the answer yes. My view of the printed news is that it can't compete with Charmin, not even with cheap, house-brand paper towels. For the electronic versions, they are not useful even as fire starter, shredding for cat litter, or wrapping dead fish heads -- that is, are useless. So, don't read them. And don't debunk them, either -- debunking wastes your time and is something the newsies long since have ignored. The newsies have no shame. Ignore the newsies. They ignore the debunking efforts. My startup manipulates some data, and how is from some applied math I derived. From what I've seen of AI, my work would qualify as quite good and innovative AI -- besides the work is solid theorems and proofs. Still, I see no good reason to give my work a catchy name or call it AI. E.g., I'm not trying to fool anyone. Or, why would I want to associate my good work with a lot of hype to fool people?

nfd 10y ago
You'd do it to make money. Boatloads of delicious, steaming venture capital. Mmm.
- graycat 10y ago
  I can say with high confidence that there is not even one venture capital person in the US who would invest even 10 cents in my work, call it AI or not, before they see usage significant and growing rapidly, and then they are not investing in the math or the AI but just the traction and its rate of growth. Besides, I'm a solo founder with a meager burn rate so that by the time I have the traction the VCs want, I will be nicely profitable with plenty of cash for growth just from retained earnings. Or, my back of the envelope arithmetic is that with common ad rates from ad networks, a $1000 server kept half busy 24 by 7 would generate $250,000 a month in revenue. For just one server, a cheap Internet connection, just one employee, that's a heck of a profit margin and plenty of cash for 10 more servers. Half fill those, say, in two spare bedrooms I have, with some window A/C units, some emergency power supplies and an emergency generator, and I will have annual revenue more than a VC seed or Series A equity check. And, "Look, Ma, 100% owner of just an LLC and no BoD!". Sure, once I have the traction, VCs will call me, and then I will check and tell them about all the times I sent them e-mail they ignored and explain that my plane has already left the ground, has altitude, and is climbing quickly and it's too late to buy a ticket. Sure, I typed in all the 25,000 programming language statements myself and implemented my math derivations in 100,000 lines of typing -- lots of in-line comments! The code is all in just Microsoft's Visual Basic .NET with ADO.NET for getting to SQL Server and ASP.NET for the Web pages with IIS for the Web server plus a little open source C called with Microsoft's platform invoke. I wrote my own Web page session state store using two collection classes and some TCP/IP sockets with class de/serialization -- sure, could have used REDIS, but my code is so short and simple writing my own was likely easier. Besides, now I'm about to copy that code, rip out most of it, and get a log file server that I will like a lot better than what I'm using now from Microsoft. Otherwise the code looks ready for at least first production, to, say, well past $250,000 a month in revenue. At IBM's Watson lab, I wrote AI code that shipped as a commercial IBM Program Product, and what I've written for my startup is more solid (got an IBM award for some of the code I wrote in an all night session -- let one of our programmers be done in the next afternoon instead of two weeks and got a MUCH nicer result for the customers -- trick was to do some things with entry variables to keep some run-time code on the stack of dynamic descendancy). My project needs data, and I have a lot but need to get more. Curiously, there's some good news: My development computer was crashing about five times a day, apparently a hardware problem, maybe on the motherboard. I did mud wrestling with it and, then, shopping for parts for a new computer. But my old computer now, for no good reason, is no longer crashing! The computer has plenty of free disk space for some more data. So, for now I get to set aside all the system management mud wrestling of getting a new computer and getting all the software moved to it and running, can review the most critical parts of my code a third time, write the log server, write some code to make working with SQL Server easier, get some more data, and do an alpha test, a beta test, and get some publicity. Maybe I will even go live with my development computer, get some revenue, and get a really nice first server. The problem is important; the math is solid; the code is solid; I suspect a lot, at least enough, people will like the results (it's intended to please essentially everyone on the Internet), etc., but no VC in the country wants anything to do with my work now. Nothing. Zip, zilch, zero. Lesson: To VCs, nothing but nothing matters but traction. So, the flip side of that lesson is an Opportunity: Be a solo founder where the traction the VCs want is enough for profitability and plenty of cash for organic growth. The founder of Plenty of Fish was a solo founder who eventually sold out for $500+ million. Some old remarks by A16Z confirm the possibility of a one engineer unicorn -- at http://a16z.com/2014/07/30/the-happy-demise-of-the-10x-engineer/ http://a16z.com/2014/07/30/the-happy-demise-of-the-10x-engin... with in part "This is the new normal: fewer engineers and dollars to ship code to more users than ever before. The potential impact of the lone software engineer is soaring. How long before we have a billion-dollar acquisition offer for a one-engineer startup? " I agree! I saw the need -- obvious enough. I cooked up a new solution with a new UI, UX, and some new data to be just what some new math I derived needed. Easy enough -- I've worked harder single exercises in Rudin, Principles of Mathematical Analysis. The work was easier than my Master's paper, my Ph.D., and any of the papers I've published. I wrote the code for the math and checked it various ways including with programming those calculations again in another language to check. I designed the data base tables, the Web pages, and wrote the code for the Web pages. Then the code for the session state store. Then I got sick, then got well, then my computer got sick and got well, now I'm back to progress. So far, being a solo founder, nothing particularly difficult. No reason to need a co-founder.

by graycat 10y ago

nfd 10y ago

You'd do it to make money. Boatloads of delicious, steaming venture capital. Mmm.

graycat 10y ago
I can say with high confidence that there is not even one venture capital person in the US who would invest even 10 cents in my work, call it AI or not, before they see usage significant and growing rapidly, and then they are not investing in the math or the AI but just the traction and its rate of growth. Besides, I'm a solo founder with a meager burn rate so that by the time I have the traction the VCs want, I will be nicely profitable with plenty of cash for growth just from retained earnings. Or, my back of the envelope arithmetic is that with common ad rates from ad networks, a $1000 server kept half busy 24 by 7 would generate $250,000 a month in revenue. For just one server, a cheap Internet connection, just one employee, that's a heck of a profit margin and plenty of cash for 10 more servers. Half fill those, say, in two spare bedrooms I have, with some window A/C units, some emergency power supplies and an emergency generator, and I will have annual revenue more than a VC seed or Series A equity check. And, "Look, Ma, 100% owner of just an LLC and no BoD!". Sure, once I have the traction, VCs will call me, and then I will check and tell them about all the times I sent them e-mail they ignored and explain that my plane has already left the ground, has altitude, and is climbing quickly and it's too late to buy a ticket. Sure, I typed in all the 25,000 programming language statements myself and implemented my math derivations in 100,000 lines of typing -- lots of in-line comments! The code is all in just Microsoft's Visual Basic .NET with ADO.NET for getting to SQL Server and ASP.NET for the Web pages with IIS for the Web server plus a little open source C called with Microsoft's platform invoke. I wrote my own Web page session state store using two collection classes and some TCP/IP sockets with class de/serialization -- sure, could have used REDIS, but my code is so short and simple writing my own was likely easier. Besides, now I'm about to copy that code, rip out most of it, and get a log file server that I will like a lot better than what I'm using now from Microsoft. Otherwise the code looks ready for at least first production, to, say, well past $250,000 a month in revenue. At IBM's Watson lab, I wrote AI code that shipped as a commercial IBM Program Product, and what I've written for my startup is more solid (got an IBM award for some of the code I wrote in an all night session -- let one of our programmers be done in the next afternoon instead of two weeks and got a MUCH nicer result for the customers -- trick was to do some things with entry variables to keep some run-time code on the stack of dynamic descendancy). My project needs data, and I have a lot but need to get more. Curiously, there's some good news: My development computer was crashing about five times a day, apparently a hardware problem, maybe on the motherboard. I did mud wrestling with it and, then, shopping for parts for a new computer. But my old computer now, for no good reason, is no longer crashing! The computer has plenty of free disk space for some more data. So, for now I get to set aside all the system management mud wrestling of getting a new computer and getting all the software moved to it and running, can review the most critical parts of my code a third time, write the log server, write some code to make working with SQL Server easier, get some more data, and do an alpha test, a beta test, and get some publicity. Maybe I will even go live with my development computer, get some revenue, and get a really nice first server. The problem is important; the math is solid; the code is solid; I suspect a lot, at least enough, people will like the results (it's intended to please essentially everyone on the Internet), etc., but no VC in the country wants anything to do with my work now. Nothing. Zip, zilch, zero. Lesson: To VCs, nothing but nothing matters but traction. So, the flip side of that lesson is an Opportunity: Be a solo founder where the traction the VCs want is enough for profitability and plenty of cash for organic growth. The founder of Plenty of Fish was a solo founder who eventually sold out for $500+ million. Some old remarks by A16Z confirm the possibility of a one engineer unicorn -- at http://a16z.com/2014/07/30/the-happy-demise-of-the-10x-engineer/ http://a16z.com/2014/07/30/the-happy-demise-of-the-10x-engin... with in part "This is the new normal: fewer engineers and dollars to ship code to more users than ever before. The potential impact of the lone software engineer is soaring. How long before we have a billion-dollar acquisition offer for a one-engineer startup? " I agree! I saw the need -- obvious enough. I cooked up a new solution with a new UI, UX, and some new data to be just what some new math I derived needed. Easy enough -- I've worked harder single exercises in Rudin, Principles of Mathematical Analysis. The work was easier than my Master's paper, my Ph.D., and any of the papers I've published. I wrote the code for the math and checked it various ways including with programming those calculations again in another language to check. I designed the data base tables, the Web pages, and wrote the code for the Web pages. Then the code for the session state store. Then I got sick, then got well, then my computer got sick and got well, now I'm back to progress. So far, being a solo founder, nothing particularly difficult. No reason to need a co-founder.