Not really. This is just emulating the processor and supporting MacOS system calls but does not emulate any of the Mac hardware other than the processor. This works because MacOS applications do not access the hardware directly but instead use system calls and let the operating system make the changes to hardware registers. DOS Applications on the other hand usually use System calls mainly for disk access but will mostly directly manipulate the video and sound card hardware and also often do things like reading keystrokes directly from the keyboard controller, changing the timer frequency and installing their own interrupt handlers etc. For this reason DOSBox emulates quite a lot of PC hardware beside the CPU.
I'm not completely sure about that. Even with a rather small budget you can still pay very high salaries for some positions. There are also huge differences in productivity and hiring a small number of highly paid, but also highly productive engineers can be a good strategy. If you pay 200k to a highly skilled engineer who is easily performing work that normally would require 3 average engineers with 120k each, than that 200k salary is cheap and not expensive.
They will be excluded because the discriminator network of the GAN will learn to distinguish between them and real images.
The discriminator can't do magic. It can only decide based on the information contained in the training images. If the set of training images is rather small, the performance of the discriminator won't be very good. The discriminator will either overfit the training set and generate only images very similar to the ones already contained in the training set or won't be very accurate and the GAN will generate images are not plausible.
Potentially a GAN could be a way of extracting hard to describe information out of a human oncologist: Let the GAN generate lots of images and let the oncologist rate them. The plausible images could then be used to improve the GAN.
I wonder if this can actually work. If you don't have enough images to train a classifier, why would training a GAN work? And even if training a GAN works, those images won't contain any information about tumors that were not already contained in the original images.
No, if you want to make AVX wider, then you would need to introduce a new instruction set. Intel did this several times. They started floating point SIMD with SSE which is 128-bit, then introduced the new AVX instruction set to support 256-bit wide SIMD and then AVX-512 to support 512-bit wide SIMD. If they want to scale up to 1024 or 2048-bit wide SIMD in the future than again, they will have to add new instructions and still need to support the old instructions for legacy software.
With traditional SIMD, your instruction set and the maximum width of your execution units are coupled. SVE removes this coupling. You can write SVE code now for a CPU with a 512-bit wide execution unit and would get a speedup later, without recompilation, if someone builds a CPU with a 2048-bit wide execution unit.
You don't understand what "this guy" (David Patterson and Andrew Waterman) is writing. Their main complaint with regular SIMD is that it makes the instruction set grow quickly over time, as CPU performance is scaled up and SIMD units are made wider but old instructions still need to be supported for backward compatibility. They support vector instructions similar to ARM SVE, as those allow scaling up the performance by adding wide execution units without requiring the introduction of new instructions at the same time. As an additional benefit, those instructions can reduce the code size compared to regular SIMD. So by the standard of that article ARM SVE is not SIMD. The SIMD instructions they call harmful are only those older ones such as ARM Neon, SSE, AVX, AVX2,... that use a fixed SIMD size.
And at the same time, their main complaint with those instruction sets is how they evolve over time when backward compatibility is required. If you are building a CPU for a specific purpose, let's say HPC or an embedded application, where you don't have to support old code, regular SIMD with only vector size support can be a very good choice and very RISC-like choice.
The use of the x86 instruction set wasn't the big issue with Larrabee. Larrabee would have been a bad idea no matter if it used ARM or MIPS opcodes instead. Using x86 didn't help, but that was just one among many issues of that architecture. The issues of the Larrabee architecture are things such as no fixed function hardware for things such as z-buffering or rasterization, not enough hardware threads to hide the memory latency, memory interface with not that much bandwidth but expensive but not that often usefull cache coherency, etc, SIMD units were not wide enough etc. Sure: Without x86 and with a simpler decoder things would have been slightly better.
The native tribes part was for diabetes type 2 which is likely NOT an autoimmune disorder. The idea is that many autoimmune disorders happen when parts of the immune systems that were evolved to target once common germs and parasites never encounter their true target and then target similar human proteins.
"Germ phobia" or "excessive cleanliness" is not such a good description, e.g:: even without any focus on cleanliness, nearly everyone in developed countries will only drink clean water free of worms and similar parasites. They used to be very common, however.
I have nothing against targeted ads, but advertiser should be forced to reveal their target to the receiver of the ad. I think this would stop most of the abuse and would easy way to regulate this.
A blood test is also considered a search that requires consent or a warrant in Germany. But there is always a judge on duty for quickly granting a warrant to allow the blood test, even if the suspect does not agree to the test. A positive breathalyzer is considered reason enough to grant a warrant.
The Second Amendment is there to defend the First. If they fall, then the Fourth and the Fifth fall shortly thereafter. And then the dark times.
It's easy to see that this is not true. Plenty of European states without something comparable to the 2nd amendment, but with constitutional rules comparable to 1th, 4th and 5th Amendment.
The EU won't survive. Brexit made matters worse. First, the EU will now have to find a way to compensate for the $13 billion net annual income coming from the UK. So either the EU will have to give less to Eastern countries, which would mean Eastern countries won't see the point in staying in the EU (particular in this post-migrant era), or it would mean even more money from Western European countries, which apart from Germany, will hurt their economy.
The western European countries will likely increase their contribution slightly. It is not just Germany, that is doing rather well and is able to increase their contribution. However, even if they don't and the EU has to reduce their subsidies significantly, this is not a big issue. The real advantage of being in the EU is the single market and not the subsidies. The EU budget is just 1% of EU GDP, while the economic effect of the single market is much larger.
But more importantly, since the economic catastrophe that Brussels predicted for the UK didn't happen, and obviously won't happen, it means the doom and gloom argument for staying in the EU won't scare anyone anymore.
Uh, the UK did not leave the EU yet, but it already dropped from the fastest growing economy in Europe to the slowest growing one. The pound dropped significantly in value and prices increased. At the same time everyone can see that the "have the cake and eat it" Brexit promised, isn't going to happen and NHS isn't going to get the 350 million pound a week promised.
Oh, and nudity doesn't have much to do with freedom of expression. What freedom of expression must protect is the expression of ideas (even those you don't like), and there's not much idea in a nipple.
Bullshit, nudity is a very powerful form of expression, able to provoke very strong emotions.
This has nothing to do with terrorism and everything to do with censorship. The EU is desperately trying to save itself.
Nope. The Brexit mess was already enough to save the EU. Popularity of the EU increased sharply in the EU27 because of that. Trying to censor content to save them, wouldn't work, but instead hurt them badly.
As a European it is also always funny to see how Americans are screaming "Censoship!!!!!111!" when terrorism promotion or holocaust denial is being removed, but just shrug when content gets removed due to a tiny bit of nudity.
I could make a fairly strong case for today's multi-core processors being fundamentally different in design and execution than the mini's and mainframes of the 60's.
Please do so. I don't think that case is going to be as strong as you think it is. After all, many of fundamental ideas behind today's multi-core CPUs are from the 60s: Out Of Execution (1967)Multi-cores and SIMD (1966)
Similarly, today's massively parallel designs in GPUs are also fundamental advances.
There is clearly a difference in scale in speed, but is there a fundamental advantage? Many of the key concepts behind GPUs were already known in the 1960s: SIMD (see above), the CDC6000 series used switching between threads like GPU do to compensate latency, vector processors also developed in 1960s also invented some of the concepts used by todays GPUs.
You don't have a clue. There are many other issues. At the moment most successful AI is using supervised learning and needs tons of labeled data in order to train the network. We still don't have a clue how to train an AI using only very small sample. Humans can easily learn from very small sets of examples, often a single example is good enough, ANNs needs tons of examples, especially the very deep and powerful ones. We don't know how the brain works yet, ANNs are only inspired by the brain, they are not a proper simulation. We still have to understand tons of things until we can build a simulation of the brain. And with semiconductor scaling slowing down, it might take really long until we get the processing power we would need even if would know what exactly needs to be simulated.
And what would we gain? Sure, you can also train a ANN to sort some rows in a spreadsheet or sum some numbers together, but it is something that conventional algorithms are already very good at, we don't needs ANNs to do that and they are not going to be efficient at it.
I don't think it is that simple. You can build an ASIC for Etherium, but different from Bitcoin it requires a good external memory interface. The memory requirements for Etherium are just too high for using only internal memory. Your ASIC architecture would likely look a lot like a GPU but remove many things that are not required for ETH mining.
You should be able to implement something easy in 30 minutes even in an interview situation. And most Hackerrank are not about implementing linked lists or trees, but actually using data structures such as trees or hashes, so its perfectly fine to use the data structures (and algorithms) included the standard libraries of C++, Java or Python. Take something such as Marc's Cakewalk. It easily can be implemented using a few lines of code (e.g.: 10 lines of non-golfed python). It shows if you are able to spot the trivial greedy algorithm and are somehow able to sort a small amount of numbers, but the number is so small that even bubblesort would work well.
I disagree. That someone is doing well in competitive programming (CP) type of questions does not tell you that that person is a good programmer. However, it also doesn't measure pointless stuff and proper knowledge of data structures and algorithms is a skill every good developer should have. CS101 can sometimes be pointless because many people cannot transfer this knowledge to any other problem. Just being able to tell that Quicksort is average O(nlog n) and worst-case O(n^2) is not useful if that doesn't mean anything to you. A bigger CPUs is not going to fix that O(n^3) algorithm that worked fine when the developer unit tested it with 10 elements, but somehow struggles when trying to run it on 100k elements in the production system.
In the most popular challenges in Hackerrank speed is not an issue. You only need to be a speed champion if you are competing for the top 5% or so. Otherwise speed is not an issue but figuring out the right algorithm and being able to write it down without making tons of mistakes that require long debug sessions. Taking your time to think properly about the problem and then carefully writing correct code is actually the way to go, as debugging can easily use a lot more time than coding itself. Many of the easier tasks are just a few lines of code in C++, Python or Java without any code golfing.
Proper abstraction can actually help efficiency, e.g.: the templated c++ sort is a lot faster than the C qsort function because the compiler can optimize the code for each datatype and inline the compare function, while the C qsort has to use indirect calls to the compare function via function pointers and can't do static optimizations based on things such as size of the elements.
Efficiency is also often a question of using a proper algorithm. Most of the time that O(n) DP-algorithm coded in Python is going to be much faster than a O(2^n) bruteforce algorithm handcoded in assembly.
Sure, but if you look at the limitation of current technology it is easy to figure out that there is still a huge number of problems to solve, many of them where nobody so far has any clue how to solve them. It's likely not just a matter of a few more years of research and throwing even bigger datasets and computers at the problem. Sure, you can make up any projections about the future and no matter how crazy they seem, we won't know that they are wrong until we are in the future. But are Elon Musk's style projections about the future of AI likely? No, not really. Yann LeCun is clearly an expert in AI, while Musk is a business man. Hypeing AI helps to finance Musk business and keeps the stock price high.
I think Elon Musk is the one that has either a limited understanding of current AI technology or just hypes AI on purpose, while being fully aware that AI still has major limitations and they are unlikely to disappear within the next few years. Important and very important progress has been made, but General AI is likely still very far away. Facebook's director of AI Yann LeCun gave a very good interview to IEEE spectrum: Facebook AI Director Yann LeCun on His Quest to Unleash Deep Learning and Make Machines Smarter
You need more than a single driver per vehicle. The driver will operate 40h per week, but your bus service is likely operating something such as 7*16h=112h per week and the bus driver can't drive 8h straight without breaks. Even at minimum wage the cost for 3-4 drivers is pretty significant. Within a single year you be able to get back the extra money required for expensive sensors, compute modules and software.
Out of date also doesn't really matter as long as it can still do its job. This first generation self-driving trolley might only work within environments that are easy to handle and in a few years you might have a second generation trolley that can handle more complex environments and drive faster, but that doesn't mean that first gen. trolleys cannot continue to fulfil their limited roles.
The speed of a local bus is slow, so limit of 45 km/h is not going to make a significant difference. This could make journey time shorter by increasing the frequency of the buses and thus reducing waiting times. With self-driving buses cities can easily go use a high number of smaller trolleys instead of large buses at a low frequency.
More people would use public transport instead of their own car. While bus drivers will lose their job, new jobs will be created elsewhere, e.g.: When people save money by not owning a car, they will likely spend that money elsewhere, e.g.: eating at restaurants more often.
The issues here is the bundling of the google applications. A manufacturer can't decide that it wants to install e.g.: maps and gmail but not google search. Either full AOSP without maps, etc. or full blown google. This seems pretty similar to the Microsoft Windows / IE bundling things. On the other hand, Microsoft was charging money for windows, while google gives away Android and the apps for free. I wonder if they could actually do something like gapps+gsearch+forced chrome default->free, gapps without gsearch+chrome $20?
Slashdot Mirror
Not really. This is just emulating the processor and supporting MacOS system calls but does not emulate any of the Mac hardware other than the processor. This works because MacOS applications do not access the hardware directly but instead use system calls and let the operating system make the changes to hardware registers. DOS Applications on the other hand usually use System calls mainly for disk access but will mostly directly manipulate the video and sound card hardware and also often do things like reading keystrokes directly from the keyboard controller, changing the timer frequency and installing their own interrupt handlers etc. For this reason DOSBox emulates quite a lot of PC hardware beside the CPU.
I'm not completely sure about that. Even with a rather small budget you can still pay very high salaries for some positions. There are also huge differences in productivity and hiring a small number of highly paid, but also highly productive engineers can be a good strategy. If you pay 200k to a highly skilled engineer who is easily performing work that normally would require 3 average engineers with 120k each, than that 200k salary is cheap and not expensive.
They will be excluded because the discriminator network of the GAN will learn to distinguish between them and real images.
The discriminator can't do magic. It can only decide based on the information contained in the training images. If the set of training images is rather small, the performance of the discriminator won't be very good. The discriminator will either overfit the training set and generate only images very similar to the ones already contained in the training set or won't be very accurate and the GAN will generate images are not plausible.
Potentially a GAN could be a way of extracting hard to describe information out of a human oncologist:
Let the GAN generate lots of images and let the oncologist rate them. The plausible images could then be used to improve the GAN.
I wonder if this can actually work. If you don't have enough images to train a classifier, why would training a GAN work? And even if training a GAN works, those images won't contain any information about tumors that were not already contained in the original images.
AVX can get wider too if it makes sense.
No, if you want to make AVX wider, then you would need to introduce a new instruction set. Intel did this several times. They started floating point SIMD with SSE which is 128-bit, then introduced the new AVX instruction set to support 256-bit wide SIMD and then AVX-512 to support 512-bit wide SIMD. If they want to scale up to 1024 or 2048-bit wide SIMD in the future than again, they will have to add new instructions and still need to support the old instructions for legacy software.
With traditional SIMD, your instruction set and the maximum width of your execution units are coupled. SVE removes this coupling. You can write SVE code now for a CPU with a 512-bit wide execution unit and would get a speedup later, without recompilation, if someone builds a CPU with a 2048-bit wide execution unit.
You don't understand what "this guy" (David Patterson and Andrew Waterman) is writing. Their main complaint with regular SIMD is that it makes the instruction set grow quickly over time, as CPU performance is scaled up and SIMD units are made wider but old instructions still need to be supported for backward compatibility. They support vector instructions similar to ARM SVE, as those allow scaling up the performance by adding wide execution units without requiring the introduction of new instructions at the same time. As an additional benefit, those instructions can reduce the code size compared to regular SIMD. So by the standard of that article ARM SVE is not SIMD. The SIMD instructions they call harmful are only those older ones such as ARM Neon, SSE, AVX, AVX2, ... that use a fixed SIMD size.
And at the same time, their main complaint with those instruction sets is how they evolve over time when backward compatibility is required. If you are building a CPU for a specific purpose, let's say HPC or an embedded application, where you don't have to support old code, regular SIMD with only vector size support can be a very good choice and very RISC-like choice.
The use of the x86 instruction set wasn't the big issue with Larrabee. Larrabee would have been a bad idea no matter if it used ARM or MIPS opcodes instead. Using x86 didn't help, but that was just one among many issues of that architecture. The issues of the Larrabee architecture are things such as no fixed function hardware for things such as z-buffering or rasterization, not enough hardware threads to hide the memory latency, memory interface with not that much bandwidth but expensive but not that often usefull cache coherency, etc, SIMD units were not wide enough etc. Sure: Without x86 and with a simpler decoder things would have been slightly better.
The native tribes part was for diabetes type 2 which is likely NOT an autoimmune disorder. The idea is that many autoimmune disorders happen when parts of the immune systems that were evolved to target once common germs and parasites never encounter their true target and then target similar human proteins.
"Germ phobia" or "excessive cleanliness" is not such a good description, e.g:: even without any focus on cleanliness, nearly everyone in developed countries will only drink clean water free of worms and similar parasites. They used to be very common, however.
I have nothing against targeted ads, but advertiser should be forced to reveal their target to the receiver of the ad. I think this would stop most of the abuse and would easy way to regulate this.
A blood test is also considered a search that requires consent or a warrant in Germany. But there is always a judge on duty for quickly granting a warrant to allow the blood test, even if the suspect does not agree to the test. A positive breathalyzer is considered reason enough to grant a warrant.
There should be even earlier examples of prior art, e.g.
a gamepad for a PalmOS PDA from 2000.
The Second Amendment is there to defend the First. If they fall, then the Fourth and the Fifth fall shortly thereafter. And then the dark times.
It's easy to see that this is not true. Plenty of European states without something comparable to the 2nd amendment, but with constitutional rules comparable to 1th, 4th and 5th Amendment.
The EU won't survive. Brexit made matters worse. First, the EU will now have to find a way to compensate for the $13 billion net annual income coming from the UK. So either the EU will have to give less to Eastern countries, which would mean Eastern countries won't see the point in staying in the EU (particular in this post-migrant era), or it would mean even more money from Western European countries, which apart from Germany, will hurt their economy.
The western European countries will likely increase their contribution slightly. It is not just Germany, that is doing rather well and is able to increase their contribution. However, even if they don't and the EU has to reduce their subsidies significantly, this is not a big issue. The real advantage of being in the EU is the single market and not the subsidies. The EU budget is just 1% of EU GDP, while the economic effect of the single market is much larger.
But more importantly, since the economic catastrophe that Brussels predicted for the UK didn't happen, and obviously won't happen, it means the doom and gloom argument for staying in the EU won't scare anyone anymore.
Uh, the UK did not leave the EU yet, but it already dropped from the fastest growing economy in Europe to the slowest growing one. The pound dropped significantly in value and prices increased. At the same time everyone can see that the "have the cake and eat it" Brexit promised, isn't going to happen and NHS isn't going to get the 350 million pound a week promised.
Oh, and nudity doesn't have much to do with freedom of expression. What freedom of expression must protect is the expression of ideas (even those you don't like), and there's not much idea in a nipple.
Bullshit, nudity is a very powerful form of expression, able to provoke very strong emotions.
This has nothing to do with terrorism and everything to do with censorship. The EU is desperately trying to save itself.
Nope. The Brexit mess was already enough to save the EU. Popularity of the EU increased sharply in the EU27 because of that. Trying to censor content to save them, wouldn't work, but instead hurt them badly.
As a European it is also always funny to see how Americans are screaming "Censoship!!!!!111!" when terrorism promotion or holocaust denial is being removed, but just shrug when content gets removed due to a tiny bit of nudity.
I could make a fairly strong case for today's multi-core processors being fundamentally different in design and execution than the mini's and mainframes of the 60's.
Please do so. I don't think that case is going to be as strong as you think it is. After all, many of fundamental ideas behind today's multi-core CPUs are from the 60s: Out Of Execution (1967) Multi-cores and SIMD (1966)
Similarly, today's massively parallel designs in GPUs are also fundamental advances.
There is clearly a difference in scale in speed, but is there a fundamental advantage? Many of the key concepts behind GPUs were already known in the 1960s: SIMD (see above), the CDC6000 series used switching between threads like GPU do to compensate latency, vector processors also developed in 1960s also invented some of the concepts used by todays GPUs.
You don't have a clue. There are many other issues. At the moment most successful AI is using supervised learning and needs tons of labeled data in order to train the network. We still don't have a clue how to train an AI using only very small sample. Humans can easily learn from very small sets of examples, often a single example is good enough, ANNs needs tons of examples, especially the very deep and powerful ones. We don't know how the brain works yet, ANNs are only inspired by the brain, they are not a proper simulation. We still have to understand tons of things until we can build a simulation of the brain. And with semiconductor scaling slowing down, it might take really long until we get the processing power we would need even if would know what exactly needs to be simulated.
And what would we gain? Sure, you can also train a ANN to sort some rows in a spreadsheet or sum some numbers together, but it is something that conventional algorithms are already very good at, we don't needs ANNs to do that and they are not going to be efficient at it.
I don't think it is that simple. You can build an ASIC for Etherium, but different from Bitcoin it requires a good external memory interface. The memory requirements for Etherium are just too high for using only internal memory. Your ASIC architecture would likely look a lot like a GPU but remove many things that are not required for ETH mining.
You should be able to implement something easy in 30 minutes even in an interview situation. And most Hackerrank are not about implementing linked lists or trees, but actually using data structures such as trees or hashes, so its perfectly fine to use the data structures (and algorithms) included the standard libraries of C++, Java or Python. Take something such as Marc's Cakewalk. It easily can be implemented using a few lines of code (e.g.: 10 lines of non-golfed python). It shows if you are able to spot the trivial greedy algorithm and are somehow able to sort a small amount of numbers, but the number is so small that even bubblesort would work well.
I disagree. That someone is doing well in competitive programming (CP) type of questions does not tell you that that person is a good programmer. However, it also doesn't measure pointless stuff and proper knowledge of data structures and algorithms is a skill every good developer should have. CS101 can sometimes be pointless because many people cannot transfer this knowledge to any other problem. Just being able to tell that Quicksort is average O(nlog n) and worst-case O(n^2) is not useful if that doesn't mean anything to you. A bigger CPUs is not going to fix that O(n^3) algorithm that worked fine when the developer unit tested it with 10 elements, but somehow struggles when trying to run it on 100k elements in the production system.
In the most popular challenges in Hackerrank speed is not an issue. You only need to be a speed champion if you are competing for the top 5% or so. Otherwise speed is not an issue but figuring out the right algorithm and being able to write it down without making tons of mistakes that require long debug sessions. Taking your time to think properly about the problem and then carefully writing correct code is actually the way to go, as debugging can easily use a lot more time than coding itself. Many of the easier tasks are just a few lines of code in C++, Python or Java without any code golfing.
Proper abstraction can actually help efficiency, e.g.: the templated c++ sort is a lot faster than the C qsort function because the compiler can optimize the code for each datatype and inline the compare function, while the C qsort has to use indirect calls to the compare function via function pointers and can't do static optimizations based on things such as size of the elements.
Efficiency is also often a question of using a proper algorithm. Most of the time that O(n) DP-algorithm coded in Python is going to be much faster than a O(2^n) bruteforce algorithm handcoded in assembly.
Sure, but if you look at the limitation of current technology it is easy to figure out that there is still a huge number of problems to solve, many of them where nobody so far has any clue how to solve them. It's likely not just a matter of a few more years of research and throwing even bigger datasets and computers at the problem. Sure, you can make up any projections about the future and no matter how crazy they seem, we won't know that they are wrong until we are in the future. But are Elon Musk's style projections about the future of AI likely? No, not really. Yann LeCun is clearly an expert in AI, while Musk is a business man. Hypeing AI helps to finance Musk business and keeps the stock price high.
I think Elon Musk is the one that has either a limited understanding of current AI technology or just hypes AI on purpose, while being fully aware that AI still has major limitations and they are unlikely to disappear within the next few years. Important and very important progress has been made, but General AI is likely still very far away.
Facebook's director of AI Yann LeCun gave a very good interview to IEEE spectrum: Facebook AI Director Yann LeCun on His Quest to Unleash Deep Learning and Make Machines Smarter
You need more than a single driver per vehicle. The driver will operate 40h per week, but your bus service is likely operating something such as 7*16h=112h per week and the bus driver can't drive 8h straight without breaks. Even at minimum wage the cost for 3-4 drivers is pretty significant. Within a single year you be able to get back the extra money required for expensive sensors, compute modules and software.
Out of date also doesn't really matter as long as it can still do its job. This first generation self-driving trolley might only work within environments that are easy to handle and in a few years you might have a second generation trolley that can handle more complex environments and drive faster, but that doesn't mean that first gen. trolleys cannot continue to fulfil their limited roles.
The speed of a local bus is slow, so limit of 45 km/h is not going to make a significant difference. This could make journey time shorter by increasing the frequency of the buses and thus reducing waiting times. With self-driving buses cities can easily go use a high number of smaller trolleys instead of large buses at a low frequency.
More people would use public transport instead of their own car. While bus drivers will lose their job, new jobs will be created elsewhere, e.g.: When people save money by not owning a car, they will likely spend that money elsewhere, e.g.: eating at restaurants more often.
The issues here is the bundling of the google applications. A manufacturer can't decide that it wants to install e.g.: maps and gmail but not google search. Either full AOSP without maps, etc. or full blown google. This seems pretty similar to the Microsoft Windows / IE bundling things. On the other hand, Microsoft was charging money for windows, while google gives away Android and the apps for free. I wonder if they could actually do something like gapps+gsearch+forced chrome default->free, gapps without gsearch+chrome $20?