I like the general sentiment here, but I think there are a lot of details to iron out, and we would have to be careful regarding unintended consequences.
For instance, there's the matter of how to treat trade secrets, which are common in computer code. In many cases, the creator of a work doesn't even have a right to distribute source code that they've purchased a license to (say, a game engine) and have modified, so this is untenable unless you are willing to make entire business models completely flat.
I suspect on re-reading your comment that you mean the portion of a work that is distributed--in this case the game client. There are less issues with that, but still licensed assets are a fairly reasonable part of the copyrighted works market. Perhaps unlimited duplication after a lapse time would be allowed, but derivative works would not be?
It's an interesting thought. It's not going to happen, but something like it's now on my wishlist.
In fact the only mention of the word "ownership" in copyright law is in the paragraph stating all works under copyright are the inheritance of the public to own, once the copyright term has expired.
Actually, at 30 mentions of owners and ownership in Title 17, Chapter 2 alone, you are dead wrong: https://www.copyright.gov/titl...:
Read all of the laws there. You will find plenty more mentions. And in case you try to backpedal and amend your statement, since the term is used to describe the copyright itself, and not the work, you can find the term "owner of a work" and "ownership of a work" in multiple official documents associated with our government's various copyright bodies: https://www.federalregister.go... https://www.copyright.gov/docs... https://www.copyright.gov/poli...
I'm not saying I agree with US copyright law, but lets get our facts straight. Your conclusions may (or may not) be valid, but that particular argument regarding legal wording is so wrong that I have to wonder if you've even read these laws.
Bonus: Contrary to your main argument, DCMA *does* in fact prohibit actions involving circumvention of copyright--many of which are actions taken for personal use, say, displaying a legitimate copy of a video from a computer by illegally circumventing HDCP or the like. This is absurd, but that's how the law was written, and I doubt it was put there by accident.
So, one problem is that there is not always more data. In my field, we have a surplus of some sorts of data, but other data requires hundreds of thousands of hours of human input, and we only have so much of that to go around. Processing all of that is easy enough, getting more is not.
Also, by "effective", I should have made it clear that I meant "an effective overall solution to the problem", which includes all costs of training a wider, lower-precision network. This includes input data collection, storage and processing, all of the custom software to handle this odd floating point format, including FP16-specific test code and documentation, run time server costs and latency, any increased risks introduced by using code paths in training and , etc.
I'm not saying that I don't believe it's possible, I've just seen absolutely no evidence that this is a significant win in most or even a sizable fraction of cases, or that it represents a "best practice" in the field. Our own experiments have shown a severe degradation in performance when using these nets w/out a complete retraining, the software engineering costs will be nontrivial, and much of the hardware we are forced to run on does not even support this functionality.
As an analog, when we use integer based nets and switch between 16-bit and 8-bit integers, we see an unacceptable level of degradation, even though there is a modest speedup and we can use slightly larger neural nets. I'm very wary of anything with a mantissa much smaller than 16 bits for that reason--those few bits seem to make a significant difference, at least for what we're doing. We're solving a very difficult constrained optimization problem using markov chains in real time, and if the observational features are lower fidelity, the optimization search will run out of time to explore the search space effectively before the result is returned to the rest of the system. It's possible that the sensitivity of our optimization algorithm to input quality is the issue here, not the fundamental usefulness of FP16, but I'm still quite skeptical. If this were a "slam dunk", I'd expect to see it move through the literature in a wave like the Restricted Boltzmann Machine did.
Oh, and thank you for the like (great reading) and the thoughtful reply. Not always easy to find on niche topics online.
It seems like these systems are exploitative by design, even if exploitation wasn't explicitly the goal. They're designed with every possible algorithm and available data to maximize labor output at the lowest possible cost. Individual workers are operating at extreme information asymmetry and against a system which does not negotiate and only offers a take it or leave it choice.
This is by far the best comment I've ever seen regarding this sort of algorithmic labor management.
Normally I'm all for this sort of thing--my company is a client and uses it to handle large bursts of data processing quickly--but the information symmetry argument is a powerful one. Also, there doesn't seem to be a lot of competition in this space, which might otherwise ameliorate a lot of the problems induced by the "take it or leave it" bargaining approach.
The analysis provided by the article is absurd, but yours seems to lead to the inescapable conclusion that some kind of regulation is necessary to prevent blatant exploitation. Maybe just reducing information asymmetry in some way, or requiring transparency in reports available to the public on the website regarding effective wages paid to workers as a fraction of the minimum and average wages of employees in their respective countries. Surely someone can find an answer to this.
Accidentally posted as anonymous coward, reposting under my actual name.
So they're all excited about the lowest-precision, smallest-size floating point math in IEEE 754?
FP16 is good enough for neural nets. Do you really think the output voltage of a biological neurons has 32 bits of precision and range? For any given speed, FP16 allows you to run NNs that are wider and deeper, and/or to use bigger datasets That is way more important than the precision of individual operations.
There's a lot of rounding error with FP16. The neural networks I use are 16-bit integers, which work much, much better, at least for the work I'm doing.
Also, do you have a good citation that FP16 neural networks are, overall, more effective than FP32 networks, as you've described?
I'm pretty sure Take Two is doing just fine, even considering this. Also, nothing says "edgy" like "we got sued for something that wasn't even part of our game, because we're just that badass."
Seriously, I'm sure the execs are laughing all the way to the bank. And the shareholders, well WTF do you expect if you invest in a company that makes games where you can get your money back after fornicating with a professional via vehicular homicide?
I enjoyed playtesting a (very unfinished) version of this system last summer with a local gaming group. I'm under NDA so I cannot say much, but I would recommend that anyone serious about science fiction RPGs check this out when it's released. If nothing else, you won't find the experience boring.
I know it sounds funny, but I have a friend who has been working on a homebrew system for Jane Austin PnP RPGs. Narrativist RPGs actually lend themselves rather well to this kind of thing, especially if the emphasis is on storytelling and social maneuvering, instead of THAC0.
I read this as "Perfect for PNP RNGs." I thought that using user input from multiple users just to generate random dice throws was a bit much, but... interesting at least.
Start a google code project for this, I'm sure you'll be surprised how much help you get. email me if there's anything I can do.
They don't respond to electromagnetism, gravity, or the strong force.
IANAP (though I do have a physics degree), but you are completely wrong about neutrinos not being affected by gravity. Gravity curves spacetime itself (at least, if you accept GR) and therefore any ballistic particle passing through it will have a curved path, whether or not they have rest mass (which neutrinos seem to, as evidenced by spontaneous flavor changing). Heck, even:
Because it is an electrically neutral lepton, the neutrino interacts neither by way of the strong nor the electromagnetic force, but only through the weak force and gravity.
I'm not saying that neutrinos are easy to manipulate, but it's not as impossible as you make it sound.
Personally I think the late Dr. Thomas Gold's idea of aliens using excitation states in large nebulae such that they act as a MASER to fasciculate long-range communication is a much better solution, at least for broadcast purposes.
This is similar to a major plot device in Neal Stephenson's Interface (don't worry, no referral).
In the book the people backing the lead character's bid for the presidency have a virtual "focus group" of people across the nation that watch his speeches. They are able to make adjustments to the speeches in real time by monitoring the reactions of the focus group's vitals.
I, for one, think that truth is not only stranger than fiction, but quickly becoming creepier as well.
I'm not sure exactly... what privacy we are supposed to expect online... [snip] sending packets over open wires or open air.
Yeah, this is pretty ridiculous. At the very minimum, all wireless routers should support some standard VPN which is nearly unbreakable, web traffic should be encrypted, and addresses and DNS lookups should be encrypted as well (perhaps using one way hash functions or the like). It annoys me that individuals other than the computer I'm communicating with know that the communication is occurring, the port numbers, and potentially even the contents.
Maybe when we make the switch to IPv6 (heh!) we can visit some of these issues as well.
I hope not. The possibility that it may contaminate Earth with a Mars infection we have no immunity for is too high. Even a 1-in-a-million chance is not worth it. Would you want to take a 1-to-million gamble with all of humanity?
I agree with your sentiment about gambling with the lives of all of humankind, but is there any evidence to suggest that 1:1000000 are reasonable odds given:
Unlikelihood that there is present life on mars
Unlikelihood that it will survive the sample return mission
Unlikelihood that it will escape the lab
Unlikelihood that said life could survive in a much different atmosphere, temperature range, and populated biosphere
Unlikelihood that said life could in any way interface with and affect (especially infect) any terrestrial life forms
Unlikelihood that that terrestrial life form would happen to be human
Unlikelihood that it would actually kill every single human
I'm sorry, but I don't think we're talking million to one. I think we're a little closer to the number of grains of sand on the earth here, when all is multiplied together.
I wonder what the middle eastern religions, the trifecta judaism, christianity, and islam, will have to say about it.
Religion has survived much more "dangerous" things than finding evidence that there used to be bacteria on Mars. I would imagine they will say something along the lines of "It is the glory of God to conceal a thing: but the honour of kings is to search out a matter."
Either the universe is teaming with life, or we are the only ones.
Or it could be anywhere in between. We have no idea what the true criteria for life is, nor what effect sentient life has on the galaxy as a whole.
The only thing propping it up thus far are multi-album recording contracts.
A friend of mine moved to LA and recently got his band signed to a record label. He contends that the major factor propping them up at the moment is ringtones, of all things.
What should you see when you visit the LHC? Why not see what everyone else is hoping to see - the Higgs Boson!
If that doesn't work out, you can ask to see just about any other particle they make around there, there's certainly enough of them. But whatever you do, don't bring up Lexx...
Also I should state that I think it's a lot easier to find information about a solution than to implement it. Someone with a second or third tier mathematics or engineering degree could probably find most of the stuff that my bosses have worked on, but trust me they use their MIT degrees fully in the implementation and other judgement calls!
I agree with most of your sentiments (they were indeed the same thoughts I had when I originally posted). However, differential costs of living (i.e. outsourcing) can mitigate some factors. Also the fact that a trained searcher can find it so much faster than you that even if they get paid what you get, it's still a win.
I'm still hoping for smarter algorithms though:)
At my large defense contractor job, we do have librarians, and they are very helpful. However for ~3000 people, we have maybe two librarians. This is a far cry from a full-fledged staffed call center of maybe 1 for every fifty or hundred employees, which is what I had envisioned.
Also, the librarians do have good educations and are very intelligent people, however they are not subject matter experts. If I tell them I need, say, a set of wavelet basis functions that are orthonormal, efficient to compute, and maintai human perceptual accuracy under significant truncation when used for image encoding, I do not think they will be of much help for me. Someone who is a dedicated searcher, with a specialization in mathematics or computer graphics would far outstrip your average corporate librarian. Of course they probably aren't cheap either.
The main problem with this is that any search engine that shows you 10 results and then counts which ones you click, well, it's not getting your input on result #11, or 23, etc. So before anyone votes, items that happen to be near the top almost certainly stay at the top.
That's only the case if the algorithm naively counts each "vote" as being equal.
I'm sorry, I wasn't clear in my original post. What I meant to say was, whatever classification algorithm you use (voting, Bayesian inference, etc etc), items that are not on the front page will get much less peer review. This means that the classifier will have less information on them, and be less likely to promote them. Changes to the classification algorithm can only help this if they move enough "bad" links down to "float" the better ones to the top.
What my post proposed was a modified sampling algorithm, whereby the user is shown *mostly* results that are classified as being high quality, but also some percentage of results which have little or no "social" ranking information available. The idea is that this way diamonds in the rough have a chance to shine, at least some of the time. Sorry for not being clearer.
Sorry to reply to my own post, but I forgot a point I had, namely that one main constraint on the use of paid humans to enhance search is that if this were on an ad-supported search engine, the ad click through rates would have to be pretty insanely high to be profitable. If it were a subscription service, well, they'd have to find a niche for this - maybe people who are bad at searching - and aggressively market towards it. Thi is why I'm not really holding my breath for engines like Cha Cha.
FYI, my original source was not Wikipedia. I asked a police officer that I happen I know about this, and he said reasonable suspicion is all it takes to pull someone over. Considering that this is his job, and he does this on a daily basis, I figured he just might know what he was talking about, and so I only linked to Wikipedia as a summary/quick reference to confirm his information, because I assumed that if you wanted better information than that, that you could do your own query work. Most article content on Wikipedia is poor but the references are often quite good.
I think you will find that a simple reading of the U.S. Constitution is almost never an adequate measure to understand even a basic legal concept under U.S. Law. Most rights and freedoms come directly from the tradition of common law, established long before our country, and many "constitutional" rights are derived almost entirely from jurisprudence surrounding the constitution, and not the text itself.
I can understand how you might have made the assumption that because I was linking to Wikipedia, that that was somehow where I got my information. I have legal discussions on a daily basis with friends who are lawyers, policemen, trustees and medical workers, and while I am the first to admit that I do not have a rich, deep understanding of case law in the United States, I do understand that there is a little more to these kinds of tricky situations than just reading the constitution and applying "common sense".
Also, I should make the disclaimer that I know that I've been trolled, but what can I say, I have a weak spot for trolling.
Slashdot Mirror
For instance, there's the matter of how to treat trade secrets, which are common in computer code. In many cases, the creator of a work doesn't even have a right to distribute source code that they've purchased a license to (say, a game engine) and have modified, so this is untenable unless you are willing to make entire business models completely flat.
I suspect on re-reading your comment that you mean the portion of a work that is distributed--in this case the game client. There are less issues with that, but still licensed assets are a fairly reasonable part of the copyrighted works market. Perhaps unlimited duplication after a lapse time would be allowed, but derivative works would not be?
It's an interesting thought. It's not going to happen, but something like it's now on my wishlist.
In fact the only mention of the word "ownership" in copyright law is in the paragraph stating all works under copyright are the inheritance of the public to own, once the copyright term has expired.
Actually, at 30 mentions of owners and ownership in Title 17, Chapter 2 alone, you are dead wrong:
https://www.copyright.gov/titl...:
Read all of the laws there. You will find plenty more mentions. And in case you try to backpedal and amend your statement, since the term is used to describe the copyright itself, and not the work, you can find the term "owner of a work" and "ownership of a work" in multiple official documents associated with our government's various copyright bodies:
https://www.federalregister.go...
https://www.copyright.gov/docs...
https://www.copyright.gov/poli...
I'm not saying I agree with US copyright law, but lets get our facts straight. Your conclusions may (or may not) be valid, but that particular argument regarding legal wording is so wrong that I have to wonder if you've even read these laws.
Bonus: Contrary to your main argument, DCMA *does* in fact prohibit actions involving circumvention of copyright--many of which are actions taken for personal use, say, displaying a legitimate copy of a video from a computer by illegally circumventing HDCP or the like. This is absurd, but that's how the law was written, and I doubt it was put there by accident.
So, one problem is that there is not always more data. In my field, we have a surplus of some sorts of data, but other data requires hundreds of thousands of hours of human input, and we only have so much of that to go around. Processing all of that is easy enough, getting more is not.
Also, by "effective", I should have made it clear that I meant "an effective overall solution to the problem", which includes all costs of training a wider, lower-precision network. This includes input data collection, storage and processing, all of the custom software to handle this odd floating point format, including FP16-specific test code and documentation, run time server costs and latency, any increased risks introduced by using code paths in training and , etc.
I'm not saying that I don't believe it's possible, I've just seen absolutely no evidence that this is a significant win in most or even a sizable fraction of cases, or that it represents a "best practice" in the field. Our own experiments have shown a severe degradation in performance when using these nets w/out a complete retraining, the software engineering costs will be nontrivial, and much of the hardware we are forced to run on does not even support this functionality.
As an analog, when we use integer based nets and switch between 16-bit and 8-bit integers, we see an unacceptable level of degradation, even though there is a modest speedup and we can use slightly larger neural nets. I'm very wary of anything with a mantissa much smaller than 16 bits for that reason--those few bits seem to make a significant difference, at least for what we're doing. We're solving a very difficult constrained optimization problem using markov chains in real time, and if the observational features are lower fidelity, the optimization search will run out of time to explore the search space effectively before the result is returned to the rest of the system. It's possible that the sensitivity of our optimization algorithm to input quality is the issue here, not the fundamental usefulness of FP16, but I'm still quite skeptical. If this were a "slam dunk", I'd expect to see it move through the literature in a wave like the Restricted Boltzmann Machine did.
Oh, and thank you for the like (great reading) and the thoughtful reply. Not always easy to find on niche topics online.
It seems like these systems are exploitative by design, even if exploitation wasn't explicitly the goal. They're designed with every possible algorithm and available data to maximize labor output at the lowest possible cost. Individual workers are operating at extreme information asymmetry and against a system which does not negotiate and only offers a take it or leave it choice.
This is by far the best comment I've ever seen regarding this sort of algorithmic labor management.
Normally I'm all for this sort of thing--my company is a client and uses it to handle large bursts of data processing quickly--but the information symmetry argument is a powerful one. Also, there doesn't seem to be a lot of competition in this space, which might otherwise ameliorate a lot of the problems induced by the "take it or leave it" bargaining approach.
The analysis provided by the article is absurd, but yours seems to lead to the inescapable conclusion that some kind of regulation is necessary to prevent blatant exploitation. Maybe just reducing information asymmetry in some way, or requiring transparency in reports available to the public on the website regarding effective wages paid to workers as a fraction of the minimum and average wages of employees in their respective countries. Surely someone can find an answer to this.
So they're all excited about the lowest-precision, smallest-size floating point math in IEEE 754?
FP16 is good enough for neural nets. Do you really think the output voltage of a biological neurons has 32 bits of precision and range? For any given speed, FP16 allows you to run NNs that are wider and deeper, and/or to use bigger datasets That is way more important than the precision of individual operations.
There's a lot of rounding error with FP16. The neural networks I use are 16-bit integers, which work much, much better, at least for the work I'm doing. Also, do you have a good citation that FP16 neural networks are, overall, more effective than FP32 networks, as you've described?
I'm pretty sure Take Two is doing just fine, even considering this. Also, nothing says "edgy" like "we got sued for something that wasn't even part of our game, because we're just that badass."
Seriously, I'm sure the execs are laughing all the way to the bank. And the shareholders, well WTF do you expect if you invest in a company that makes games where you can get your money back after fornicating with a professional via vehicular homicide?
I enjoyed playtesting a (very unfinished) version of this system last summer with a local gaming group. I'm under NDA so I cannot say much, but I would recommend that anyone serious about science fiction RPGs check this out when it's released. If nothing else, you won't find the experience boring.
I know it sounds funny, but I have a friend who has been working on a homebrew system for Jane Austin PnP RPGs. Narrativist RPGs actually lend themselves rather well to this kind of thing, especially if the emphasis is on storytelling and social maneuvering, instead of THAC0.
I read this as "Perfect for PNP RNGs." I thought that using user input from multiple users just to generate random dice throws was a bit much, but... interesting at least.
Start a google code project for this, I'm sure you'll be surprised how much help you get. email me if there's anything I can do.
IANAP (though I do have a physics degree), but you are completely wrong about neutrinos not being affected by gravity. Gravity curves spacetime itself (at least, if you accept GR) and therefore any ballistic particle passing through it will have a curved path, whether or not they have rest mass (which neutrinos seem to, as evidenced by spontaneous flavor changing). Heck, even
I'm not saying that neutrinos are easy to manipulate, but it's not as impossible as you make it sound.
Personally I think the late Dr. Thomas Gold's idea of aliens using excitation states in large nebulae such that they act as a MASER to fasciculate long-range communication is a much better solution, at least for broadcast purposes.
This is similar to a major plot device in Neal Stephenson's Interface (don't worry, no referral).
In the book the people backing the lead character's bid for the presidency have a virtual "focus group" of people across the nation that watch his speeches. They are able to make adjustments to the speeches in real time by monitoring the reactions of the focus group's vitals.
I, for one, think that truth is not only stranger than fiction, but quickly becoming creepier as well.
Maybe when we make the switch to IPv6 (heh!) we can visit some of these issues as well.
- Unlikelihood that there is present life on mars
- Unlikelihood that it will survive the sample return mission
- Unlikelihood that it will escape the lab
- Unlikelihood that said life could survive in a much different atmosphere, temperature range, and populated biosphere
- Unlikelihood that said life could in any way interface with and affect (especially infect) any terrestrial life forms
- Unlikelihood that that terrestrial life form would happen to be human
- Unlikelihood that it would actually kill every single human
I'm sorry, but I don't think we're talking million to one. I think we're a little closer to the number of grains of sand on the earth here, when all is multiplied together.Hey Geoffrey, long time no see! Do you still hang around the Boston area? That's where I moved after I got done with the JPL and grad school stuff.
Do they still have you doing MER work? Will you be on MSL in any capacity?
What should you see when you visit the LHC? Why not see what everyone else is hoping to see - the Higgs Boson!
If that doesn't work out, you can ask to see just about any other particle they make around there, there's certainly enough of them. But whatever you do, don't bring up Lexx...
Also I should state that I think it's a lot easier to find information about a solution than to implement it. Someone with a second or third tier mathematics or engineering degree could probably find most of the stuff that my bosses have worked on, but trust me they use their MIT degrees fully in the implementation and other judgement calls!
I agree with most of your sentiments (they were indeed the same thoughts I had when I originally posted). However, differential costs of living (i.e. outsourcing) can mitigate some factors. Also the fact that a trained searcher can find it so much faster than you that even if they get paid what you get, it's still a win. I'm still hoping for smarter algorithms though :)
And kissing just got even sexier!
I certainly don't mind if those drones are flying around me - I just hope they do not look directly into my laser with their remaining eye... *grin*
At my large defense contractor job, we do have librarians, and they are very helpful. However for ~3000 people, we have maybe two librarians. This is a far cry from a full-fledged staffed call center of maybe 1 for every fifty or hundred employees, which is what I had envisioned.
Also, the librarians do have good educations and are very intelligent people, however they are not subject matter experts. If I tell them I need, say, a set of wavelet basis functions that are orthonormal, efficient to compute, and maintai human perceptual accuracy under significant truncation when used for image encoding, I do not think they will be of much help for me. Someone who is a dedicated searcher, with a specialization in mathematics or computer graphics would far outstrip your average corporate librarian. Of course they probably aren't cheap either.
What my post proposed was a modified sampling algorithm, whereby the user is shown *mostly* results that are classified as being high quality, but also some percentage of results which have little or no "social" ranking information available. The idea is that this way diamonds in the rough have a chance to shine, at least some of the time. Sorry for not being clearer.
Sorry to reply to my own post, but I forgot a point I had, namely that one main constraint on the use of paid humans to enhance search is that if this were on an ad-supported search engine, the ad click through rates would have to be pretty insanely high to be profitable. If it were a subscription service, well, they'd have to find a niche for this - maybe people who are bad at searching - and aggressively market towards it. Thi is why I'm not really holding my breath for engines like Cha Cha.
FYI, my original source was not Wikipedia. I asked a police officer that I happen I know about this, and he said reasonable suspicion is all it takes to pull someone over. Considering that this is his job, and he does this on a daily basis, I figured he just might know what he was talking about, and so I only linked to Wikipedia as a summary/quick reference to confirm his information, because I assumed that if you wanted better information than that, that you could do your own query work. Most article content on Wikipedia is poor but the references are often quite good.
I think you will find that a simple reading of the U.S. Constitution is almost never an adequate measure to understand even a basic legal concept under U.S. Law. Most rights and freedoms come directly from the tradition of common law, established long before our country, and many "constitutional" rights are derived almost entirely from jurisprudence surrounding the constitution, and not the text itself.
I can understand how you might have made the assumption that because I was linking to Wikipedia, that that was somehow where I got my information. I have legal discussions on a daily basis with friends who are lawyers, policemen, trustees and medical workers, and while I am the first to admit that I do not have a rich, deep understanding of case law in the United States, I do understand that there is a little more to these kinds of tricky situations than just reading the constitution and applying "common sense".
Also, I should make the disclaimer that I know that I've been trolled, but what can I say, I have a weak spot for trolling.