There may be different preferences, but to what extent are those interests engineered by culture in a way that disadvantages people? We drill into kids heads that they *should* want something and coincidently as a culture we drastically make life miserable for that aspiration... There are of course two fixes, changing culture to truly make all avenues equally appealing or fix the problem where we do not compensate highly valuable responsibilities.
The problem is part of the advertised purpose of using algorithm to help govern justice is that they removed race from the equation to fix bias. So a large part of their intent is predicated an the assumption that the system has been unfair. Yet it uses an indicator that specifically just carries forward the system behavior from the previous generation. Whatever debate may be had about race issues today, there is no way someone can claim the prior generation had a fair shake.
I thought there was still a way to offset your earnings... For example: Company X is a US operation that makes $100 million dollars Company X2 is a company in a Tax Haven that isn't X, but coincidentaly owns a lot of intellectual property or whatnot that X is based upon. X2 charges X coincidentally $100 million dollars in licensing fees for use of the intellectual property, and Company X gets to write off a $100 million dollar business expense.. in reality to themselves.
Note I'd be very interested to be updated on how this is wrong, outdated, or more nuanced, but it's what I heard last time I tried to understand how tax havens worked.
I'm of the opinion this is a gimmick. 'oh we've made a terrible mistake resulting in a few thousand units being out there that will happen to become limited edition collector's editions'.
The messages are so incredibly tame that the over the top apology alongside 'yes these will be for sale, gee sorry' is just awkward.
If this is 'instead of making them watch a television screen like my opponents, they can see a nicer thing', that's one thing.
But he's going to be going to be up against people actually bothering to show up.
It may be silly, but actually getting to be in person is still considered important to prove that the candidate thinks enough of the people to show up. A lot of voters would react just like the parent post said, 'screw this elitist and his hologram' is a likely reaction among people that would normally care about a local rally happening.
Of course, evidence suggests that the arrest/conviction experience is unfair to people performing the same sorts of activities. So historical arrest/conviction/recidivsm data is not the ideal basis for an algorithm. Just because someone got arrested/convicted doesn't mean they actually did it. Just because someone didn't get arrested/convicted doesn't mean they didn't do it.
While I'm willing to believe the other parts of it, I wouldn't count out the problem of trying to fix biased systems by training algorithms. The data being fed into a machine learning strategy is going to just try its best to imitate the system it is being fed data about. Generally we lack straightforward means to 'adjust' such algorithms.
The situation you described is indeed what I was thinking of and I did oversimplify, but the core remains: the algorithm cannot measure the absolute truth, only the historical judicial consequences. If the system never manages to assert a guilty party's innocence or to note that someone got off not because they were innocent but because they had better lawyer or unfairly had more leniency from police and court. To the extent those lie on racial and economic lines and to the extent they *unfairly* lie on those lines (e.g. the really rich and really poor probably commit more crimes than the middle, the rich mainly because that's how they get rich and the poor because they have so little to lose and more desperation).
Interestingly enough in your example, even if you removed the actual gender from the data, you'd probably still have a 'biased' selection algorithm.
This came up in some other scandal where an algorithm *tried* not to be racist by excluding race and ended up still very biased in a law enforcement context. Note that the algorithm was seemingly bogus for other reasons so its not the best example, but even if it was worknig correctly it still probably would have been biased and the bias would have been undeserved. Notably they looked at arrest records of the parents as an indicator, and if a biased system caused their parents to be arrested, then the system would gladly extend that bias to a new generation.
Which all points to a key problem of playing 'whack-a-mole' with various endpoints where bias manifests when the bias problem is a bit more systemic. If a field is unfairly excluding minorities or women, then you don't just wave a wand at the employers, you have to look at education and cultural upbringing and accept that correcting a balance problem may be a generational problem. Also make sure the people you think are being slighted actually want this kind of help, rather than elevating the state of thnigs they would rather do.
The question of course is why this would be different *now* than before. From the second we had our first dual core processor, I've been hearing people proclaim the death of multi-socket designs with every core count bump. Yet here we are...
The blade v. 1u argument is a bit moot. Already dense form factors can double up the density. I've seen proposals for 'dual system' boards to gain some marginal economy of scale for the physical components but not the complexity of having the sockets interconnected.
Of course, it is fair to say that if you are an application developer with a cpu/memory hungry behavior and you *don't* support multiple-systems... One day a competitor will displace you that will span multiple systems. It's an increasingly rare segment and the vendors are charging ever more for systems to accomodate that appetite. A competitor that allows your customers to do the same thing but with 4x cheaper hardware can have an easy time of it.
Note that this was said the second we had dual core processors, and is generally repeated ever so often.
We now can have a single-socket system that is a 32-way server, and yet there is still push for 8 socket and beyond. Parts of the industry just stubbornly don't want to let go.
Some of it is adherence to tradition, but at least some of it is due to some memory capacity intensive workflows that will eat every byte of memory you can feed it.
A strong argument for the relatively few applications that do this is that they've been derelict in moving beyond single system and if they do not migrate to multi-system model, then they will be replaced by a competitor that will.
Another regretable situation I've seen is some people thinking that's the way to do virtualization. Thankfully this is no longer a majority opinion as far as I can tell.
The difference being a human that sees lane markers leading into active oncoming traffic will decide there are shenigans and not follow.
It points to a big gap in machine learning strategies in general: Training generally happens focused on positive correlations and not a lot of injection of maliciously designed data. So a well trained model is dumb and just says 'training says always follow lines' and follows it right head on into traffic.
This is also a sign of likely problems in road construction, where markings are frequently very messed up.
This is not 'a machine can be fooled like a human', it's a reminder that the machine is still a *lot* dumber than a human.
There have been vendors chasing the 'high' end, but all but Marvell have bowed out. AMD has cancelled future ARM product. Qualcomm seems to have failed to bring their offering to market. Cavium got bought by Marvell. Broadcom's rumored Xeon competitor is nowhere to be seen.
So Cavium ThunderX is the only platform to get the PCIe and bunch of DIMMs. In fact the *one* benchmark Marvell can show as compelling is the memory bandwidth compared to Xeon. It also can be seen with fairly normal 'pc-like' firmware.
Problem is that performance and performance/watt actually suck compared to Intel. To the extent this is due to hardware limitations or just lack of software investment is unclear, but it is the reality they are faced with at the moment. AMD not being able to compete using ARM but can compete using x86 at least *suggests* that Intel's massive software investment pays off in performance for x86 platforms.
. It's not like a lens-adjustment mechanism is going to be break the bank,
Well there's the lense adjustment, but more critically those headsets have two smaller panels instead of one big panel, which is presumably more expensive than the single panel. Of course a Quest is going to have two panels and a 'flagship' mobile processor and still be under $400....
At this point I'm ignoring Oculus software. Not only is it lockin, it is lockin to an ecosystem that *explicitly* doesn't want a high end device. If they don't want to do the better hardware, they should just open up support for industry headsets. Of course their problem there is complete lack of differentiation from Valve, but they are now only differentiated in bad ways.
Inference of adequate recognition models is not so impossibly demanding that you need an expensive cluster of machines. In fact a raspberry pi can do inference of a voice recognition model.
An argument can be made about how 'connected' a device is to your data and how easy it is for a vendor to have to worry about a simplistic frontend in the edge device and thus be able to freely update the backend... But saying that you need gigantic clusters to do simple voice recognition is not accurate.
Of course, schools also like the whole 'can't do crap locally' so that the device state can't be screwed up by a student. Though if they had to pay a single penny more for that, then they wouldn't do it.
With respect to duration, yes music is a distraction, but as work goes on does the impact to morale offset the distraction?
Also, if the music is serving to filter out other distractions (e.g. open landscape office area), is the music less distracting than office noice (conversations and such)?
Of course self-reporting is likely to be incorrect: -When it's purely academic, you assume yes, but then you actually look at your budget or perhaps notice that Hulu is cheaper, maybe you don't get netfilx -A policy change to clamp down on moochers creates negative PR, souring people's opinion of netflix -There are probably some people that don't care about Netflix that much, but still pay for it because they have a friend/family member mooching and it's worth the small amount of money to not inconvenience that friend/family member. So there may be some losses of those not surveyed.
I will confess I was trying to find some defense for a blockchain role in supply chain, but the more I think about it, having traditionally signed transactions just makes so much more sense there.... Well, I tried...
I don't think the 'pitch' includes making your blockchain available to people not involved in the transaction, as more often than not these companies would rather obscure the supply chain.
This is one reason why it takes some effort to justify blockchain for this, as the number of parties you want to be reading the data is very limited, and why signed transactions in a non-blockchain media is probably a better fit...
As the author states, a lot of the supply chain stuff is garbage.
The only thing that might be useful is that if at any point where the product changes hands neither party trusts the other, a double-entry accounting of transactions can be useful. It can't prove what happened, but it can prevent a party from going back and changing what they had previously agreed happened.
I have 10 barrels of product I am handing over to a truck from a warehouse.
Let's say this is a traditional 'boring old database' and I agree that the truck took 10 barrels and the truck agrees it took 10 barrels and this goes into a database.
Now let's say that the trucking company controlled the database. They could decide to go back and change it to 'only 8 barrels was picked up', and then steal 2 barrels.
If the warehouse company controls the database, then they go back and change it to retroactively claim the truck picked up 12 barrels and accuse them of stealing 2 barrels.
Blockchain is *a* strategy where: -Warehouse would authenticate a transaction where they provided 10 barrels -Truck authenticates a transaction where they received 10 barrels
Any attempt by either party to 'rewind' and modify their transaction would leave them unable to produce a blockchain that has the other party agreeing to the new proposed way it went.
Of course a traditional database with signatures *could* be used as well, but in practice it just isn't done.
While I'm willing to believe the rewrite was a debacle, I would guess that the same team that did a crappy rewrite would have also failed at extending the existing codebase as well. In fact I'd be surprised if they hadn't spent some non-trivial time failing to evolve the codebase before resorting to rewrite.
Rewrites are frequently the 'end of the road', but I think they are usually a symptom rather than a cause. When the product starts failing, a rewrite is a likely 'hail mary' move to hope for the best. As such it can get the blame even though things were already set before that move was even attempted.
Slashdot Mirror
There may be different preferences, but to what extent are those interests engineered by culture in a way that disadvantages people? We drill into kids heads that they *should* want something and coincidently as a culture we drastically make life miserable for that aspiration... There are of course two fixes, changing culture to truly make all avenues equally appealing or fix the problem where we do not compensate highly valuable responsibilities.
The problem is part of the advertised purpose of using algorithm to help govern justice is that they removed race from the equation to fix bias. So a large part of their intent is predicated an the assumption that the system has been unfair. Yet it uses an indicator that specifically just carries forward the system behavior from the previous generation. Whatever debate may be had about race issues today, there is no way someone can claim the prior generation had a fair shake.
I thought there was still a way to offset your earnings... For example:
Company X is a US operation that makes $100 million dollars
Company X2 is a company in a Tax Haven that isn't X, but coincidentaly owns a lot of intellectual property or whatnot that X is based upon.
X2 charges X coincidentally $100 million dollars in licensing fees for use of the intellectual property, and Company X gets to write off a $100 million dollar business expense.. in reality to themselves.
Note I'd be very interested to be updated on how this is wrong, outdated, or more nuanced, but it's what I heard last time I tried to understand how tax havens worked.
I'm of the opinion this is a gimmick. 'oh we've made a terrible mistake resulting in a few thousand units being out there that will happen to become limited edition collector's editions'.
The messages are so incredibly tame that the over the top apology alongside 'yes these will be for sale, gee sorry' is just awkward.
I'm skeptical as well, but I don't think UBI would be 'free ride' but rather 'barely survivable when augmented by some sort of job or meager savings'.
If this is 'instead of making them watch a television screen like my opponents, they can see a nicer thing', that's one thing.
But he's going to be going to be up against people actually bothering to show up.
It may be silly, but actually getting to be in person is still considered important to prove that the candidate thinks enough of the people to show up. A lot of voters would react just like the parent post said, 'screw this elitist and his hologram' is a likely reaction among people that would normally care about a local rally happening.
Of course, evidence suggests that the arrest/conviction experience is unfair to people performing the same sorts of activities. So historical arrest/conviction/recidivsm data is not the ideal basis for an algorithm. Just because someone got arrested/convicted doesn't mean they actually did it. Just because someone didn't get arrested/convicted doesn't mean they didn't do it.
While I'm willing to believe the other parts of it, I wouldn't count out the problem of trying to fix biased systems by training algorithms. The data being fed into a machine learning strategy is going to just try its best to imitate the system it is being fed data about. Generally we lack straightforward means to 'adjust' such algorithms.
The situation you described is indeed what I was thinking of and I did oversimplify, but the core remains: the algorithm cannot measure the absolute truth, only the historical judicial consequences. If the system never manages to assert a guilty party's innocence or to note that someone got off not because they were innocent but because they had better lawyer or unfairly had more leniency from police and court. To the extent those lie on racial and economic lines and to the extent they *unfairly* lie on those lines (e.g. the really rich and really poor probably commit more crimes than the middle, the rich mainly because that's how they get rich and the poor because they have so little to lose and more desperation).
Interestingly enough in your example, even if you removed the actual gender from the data, you'd probably still have a 'biased' selection algorithm.
This came up in some other scandal where an algorithm *tried* not to be racist by excluding race and ended up still very biased in a law enforcement context. Note that the algorithm was seemingly bogus for other reasons so its not the best example, but even if it was worknig correctly it still probably would have been biased and the bias would have been undeserved. Notably they looked at arrest records of the parents as an indicator, and if a biased system caused their parents to be arrested, then the system would gladly extend that bias to a new generation.
Which all points to a key problem of playing 'whack-a-mole' with various endpoints where bias manifests when the bias problem is a bit more systemic. If a field is unfairly excluding minorities or women, then you don't just wave a wand at the employers, you have to look at education and cultural upbringing and accept that correcting a balance problem may be a generational problem. Also make sure the people you think are being slighted actually want this kind of help, rather than elevating the state of thnigs they would rather do.
The question of course is why this would be different *now* than before. From the second we had our first dual core processor, I've been hearing people proclaim the death of multi-socket designs with every core count bump. Yet here we are...
The blade v. 1u argument is a bit moot. Already dense form factors can double up the density. I've seen proposals for 'dual system' boards to gain some marginal economy of scale for the physical components but not the complexity of having the sockets interconnected.
Of course, it is fair to say that if you are an application developer with a cpu/memory hungry behavior and you *don't* support multiple-systems... One day a competitor will displace you that will span multiple systems. It's an increasingly rare segment and the vendors are charging ever more for systems to accomodate that appetite. A competitor that allows your customers to do the same thing but with 4x cheaper hardware can have an easy time of it.
Note that this was said the second we had dual core processors, and is generally repeated ever so often.
We now can have a single-socket system that is a 32-way server, and yet there is still push for 8 socket and beyond. Parts of the industry just stubbornly don't want to let go.
Some of it is adherence to tradition, but at least some of it is due to some memory capacity intensive workflows that will eat every byte of memory you can feed it.
A strong argument for the relatively few applications that do this is that they've been derelict in moving beyond single system and if they do not migrate to multi-system model, then they will be replaced by a competitor that will.
Another regretable situation I've seen is some people thinking that's the way to do virtualization. Thankfully this is no longer a majority opinion as far as I can tell.
The difference being a human that sees lane markers leading into active oncoming traffic will decide there are shenigans and not follow.
It points to a big gap in machine learning strategies in general: Training generally happens focused on positive correlations and not a lot of injection of maliciously designed data. So a well trained model is dumb and just says 'training says always follow lines' and follows it right head on into traffic.
This is also a sign of likely problems in road construction, where markings are frequently very messed up.
This is not 'a machine can be fooled like a human', it's a reminder that the machine is still a *lot* dumber than a human.
There have been vendors chasing the 'high' end, but all but Marvell have bowed out. AMD has cancelled future ARM product. Qualcomm seems to have failed to bring their offering to market. Cavium got bought by Marvell. Broadcom's rumored Xeon competitor is nowhere to be seen.
So Cavium ThunderX is the only platform to get the PCIe and bunch of DIMMs. In fact the *one* benchmark Marvell can show as compelling is the memory bandwidth compared to Xeon. It also can be seen with fairly normal 'pc-like' firmware.
Problem is that performance and performance/watt actually suck compared to Intel. To the extent this is due to hardware limitations or just lack of software investment is unclear, but it is the reality they are faced with at the moment. AMD not being able to compete using ARM but can compete using x86 at least *suggests* that Intel's massive software investment pays off in performance for x86 platforms.
. It's not like a lens-adjustment mechanism is going to be break the bank,
Well there's the lense adjustment, but more critically those headsets have two smaller panels instead of one big panel, which is presumably more expensive than the single panel. Of course a Quest is going to have two panels and a 'flagship' mobile processor and still be under $400....
At this point I'm ignoring Oculus software. Not only is it lockin, it is lockin to an ecosystem that *explicitly* doesn't want a high end device. If they don't want to do the better hardware, they should just open up support for industry headsets. Of course their problem there is complete lack of differentiation from Valve, but they are now only differentiated in bad ways.
Inference of adequate recognition models is not so impossibly demanding that you need an expensive cluster of machines. In fact a raspberry pi can do inference of a voice recognition model.
An argument can be made about how 'connected' a device is to your data and how easy it is for a vendor to have to worry about a simplistic frontend in the edge device and thus be able to freely update the backend... But saying that you need gigantic clusters to do simple voice recognition is not accurate.
Of course, schools also like the whole 'can't do crap locally' so that the device state can't be screwed up by a student. Though if they had to pay a single penny more for that, then they wouldn't do it.
It's goofy if it is indeed linked to 'two screen' devices.
To the extent chromebooks succeed in practice, it is because they are cheap laptop-alikes, with screens and keyboards.
A dual screen device would not have a keyboard, and would further cost more because a second screen is more expensive than a keyboard.
It doesn't make sense to try to conflate 'intel wants to drive a new form factor' and 'microsoft wants to compete with chromebooks' into one thing.
I presume he is talking about the representatives, where those states have badly gerrymandered districts.
Not particularly relevant to senate though.
With respect to duration, yes music is a distraction, but as work goes on does the impact to morale offset the distraction?
Also, if the music is serving to filter out other distractions (e.g. open landscape office area), is the music less distracting than office noice (conversations and such)?
This is true and should be recognized.
Of course self-reporting is likely to be incorrect:
-When it's purely academic, you assume yes, but then you actually look at your budget or perhaps notice that Hulu is cheaper, maybe you don't get netfilx
-A policy change to clamp down on moochers creates negative PR, souring people's opinion of netflix
-There are probably some people that don't care about Netflix that much, but still pay for it because they have a friend/family member mooching and it's worth the small amount of money to not inconvenience that friend/family member. So there may be some losses of those not surveyed.
This might apply for keyboard key presses
Well, I doubt you'll find a keyboard that is DMAing, so it wouldn't apply there. Even if it did, no one types fast enough for this to even be a blip.
15% cost will matter a great deal to some, not at all to others.
Note that thunderbolt is a common feature of non-Apple PCs as well.
One facet is that it supports 4x PCIe, so it can provide a much better performing connection than usb-c by itself.
I will confess I was trying to find some defense for a blockchain role in supply chain, but the more I think about it, having traditionally signed transactions just makes so much more sense there.... Well, I tried...
I don't think the 'pitch' includes making your blockchain available to people not involved in the transaction, as more often than not these companies would rather obscure the supply chain.
This is one reason why it takes some effort to justify blockchain for this, as the number of parties you want to be reading the data is very limited, and why signed transactions in a non-blockchain media is probably a better fit...
As the author states, a lot of the supply chain stuff is garbage.
The only thing that might be useful is that if at any point where the product changes hands neither party trusts the other, a double-entry accounting of transactions can be useful. It can't prove what happened, but it can prevent a party from going back and changing what they had previously agreed happened.
I have 10 barrels of product I am handing over to a truck from a warehouse.
Let's say this is a traditional 'boring old database' and I agree that the truck took 10 barrels and the truck agrees it took 10 barrels and this goes into a database.
Now let's say that the trucking company controlled the database. They could decide to go back and change it to 'only 8 barrels was picked up', and then steal 2 barrels.
If the warehouse company controls the database, then they go back and change it to retroactively claim the truck picked up 12 barrels and accuse them of stealing 2 barrels.
Blockchain is *a* strategy where:
-Warehouse would authenticate a transaction where they provided 10 barrels
-Truck authenticates a transaction where they received 10 barrels
Any attempt by either party to 'rewind' and modify their transaction would leave them unable to produce a blockchain that has the other party agreeing to the new proposed way it went.
Of course a traditional database with signatures *could* be used as well, but in practice it just isn't done.
While I'm willing to believe the rewrite was a debacle, I would guess that the same team that did a crappy rewrite would have also failed at extending the existing codebase as well. In fact I'd be surprised if they hadn't spent some non-trivial time failing to evolve the codebase before resorting to rewrite.
Rewrites are frequently the 'end of the road', but I think they are usually a symptom rather than a cause. When the product starts failing, a rewrite is a likely 'hail mary' move to hope for the best. As such it can get the blame even though things were already set before that move was even attempted.