Re:More corporate BS
on
The End of Free
·
· Score: 3, Insightful
First, "Information wants to be free" is IMO clueless in itself. Information doesn't want to be free any more than your doorknob wants to be free. You could as easily say "Information wants to be paid for".
I think the right context is "information wants to be free" like "water wants to flow downhill". Sure, you can limit water's progress by building a dam, just like you can encrypt data or otherwise limit the access to information; but in the internet world, information tends to become free (pirates, cracking, etc). In that sense, information is more slippery than water, particularly on general purpose computers, but it is possible to limit it.
Then it's no longer DRM, which is basically a program that has both lock and key yet tries to hide the key from the user except in "allowed" circumstances. Personal encryption doesn't include both the lock and the key; instead, you have the key and you use it to prove to the lock program that you have (unlimited) access to the encrypted volume/whatever.
Besides, it stands to reason that what you're encrypting is meant to be private, thus, since it's not released, it doesn't fall within the domain of copyright. You're not distributing anything, so limits to distribution don't apply.
Some tidbits about the most toxic metal on the planet.
I would suspect that Polonium (you know, what the Russians used to kill Litvinenko) is a mite more toxic. As far as ingestion in particular is concerned, there is also the caffeine challenge. Granted, caffeine is not a metal, but one wouldn't usually consider it very toxic.
No, rational ignorance doesn't factor into it. If people were rational optimizers with an objective of favorably altering the outcome, they'd quickly find out that even the cost of going to the polls to vote isn't worth the extremely small chance of the increase in marginal utility. Thus they wouldn't vote at all. Yet this is not what happens in the real world, hence people aren't rational optimizers, W5.
How about an automatic sunset: a law that has 50%+1 support gets to live 5 years before it has to be passed again. A law that has 100% support gets to live 10 years before it has to be passed again. Scale linearly between the two to give some incentive to make popular laws, not just squeakers. If that would cause an overload at "pass-again day", add +/- 5% of the duration to the time until it has to be passed again so that the exact day will be sufficiently randomized.
Possible but not likely. Unless a corporation has exclusive access to a needed resource, there's always going to be competition. The reason we don't see as much now in a lot of areas is because it's a legalized monopoly, or the regulations are piled on so thick that only the big boys and play the game. Then you end up with real economic powers controlling your voted-in politicians... no matter who you voted in.
Whether or not a corporation will be able to rewrite the rules depends on whether the barriers to entry can be made high enough that the corporation can then devote itself to gaining power through first accumulating monopoly profit, and second by translating that economic power into physical power. To do the second isn't so hard to imagine: all it would have to do would be to absorb a few of the private defense contractors (in extreme libertarianism, at least). To do the first, there are many effects that could lead to a strong monopoly. For instance, the sector in which it operates might support great economies of scale, so that entrants to the market has no way of competing with the monopoly, or it may support few enough companies that collusion will occur.
The real economic powers controlling the politicians is a form of corruption or collusion in itself. It is powerful in the US because of the broken state of the democratic system. For that reason, among others, I favor an electoral system that gives third parties and independents a real chance: more candidates means that it's much harder to corrupt all of them. For that matter, if you want to be radical, pick a very large assembly randomly and let it elect from among its numbers to Congress - hard to systematically corrupt that.
One might also extend a libertarian "do it yourself" heuristic to all levels: if a decision will affect a certain group, that group (and only that group) should decide whether to do it. Thus, it's the business of nobody else what you do inside your home, but organizations that face externality or have significant power would have to take the external parties into account. The hard part, of course, is to actually implement that heuristic - I don't think Coasian bargaining would work - and some people would say every action affects everybody hence we have a rationale for a large state.
In sum, I suppose I don't really oppose a decentralized society. What I don't like is the capitalist aspect that gets entangled with it in American libertarianism, where the market is considered robust enough to keep corporations from becoming the new states, and consumer democracy good enough to keep those who have power in check. The robber baron era, as well as the case of company towns, seems to show otherwise.
At least you can (in theory) elect your representatives. A corporation has no such flaw: it's a plutocratic oligarchy through and through.
I say "in theory" because the United States implementation of democracy is broken (Gerrymandering, two-party rule, you name it). Yet that is no argument against democracy, merely one against a particular implementation of it.
Lastly... the corporations can't make up stupid laws or arrest you for making dry ice bombs.
In a libertarian system, eventually one of the corporations will end up with enough economic power that it can rewrite the rules of the game. When that happens, that corporation will translate the economic power into physical power and effectively make itself the new state, and then it can make whatever laws it wants.
It's possible to make a quantum repeater (it uses quantum teleportation to move the information without looking at it). Send some classical location data along the quantum encrypted stuff and have the cell tower choose which repeater to use based on the location data, and then you can have more than just point-to-point.
Aligning your laser with that of the tower will be a major pain, though, and you can't just get around it by sending more than one photon: doing so will void the theoretical uncrackability properties of quantum crypto.
Yes, but IRV branch methods usually all have the following flaws: they're not monotone (moving someone up in rank can make him lose), they're not summable (you have to send voting data for every ballot instead of a sum across candidates, unlike Plurality per-candidate counts or Condorcet matrices), and they're not Condorcet compatible.
IRV based on Borda (Baldwin and Nanson) is Condorcet compatible and I think it is summable, but the monotonicity problem remains. I don't know what Avy is, but the "twin towers" defect seems to suggest that the monotonicity problem remains there as well (Craig Carey's notation can be hard to decipher).
Because IRV has the problems I've outlined, it is true that they also exist in STV, but unlike IRV, there aren't any well known better alternatives. Besides, as the number of seats increases, STV's proportionality criterion (called the Droop proportionality criterion) limits the extent to which the problems distort the result.
You might be interested in this paper, which deals with perverse incentives in the context of education, and gives examples of how metrics are distorted outside of education as well. Their conclusion is that the use of quantitative optimization is, on the whole worth it; I don't agree, but even with the difference of opinion, their examples are illuminating.
Range reduces to Approval, since you want to give every candidate either maximum if you support him or minimum if you don't so your chance of affecting the outcome is as great as possible. Approval, in turn, requires polling and strategy so you can know where to put the cutoff.
Say the three candidate for President is Bush, Gore, and Nader. You like Nader but don't want Bush. If Nader has negligible support, you would vote for Gore and Nader, so that your vote still helps Gore keep Bush away. However, if Nader has substantial support, any vote for both will help Gore win at the expense of Nader, so you might want to reconsider and put the cutoff after Nader rather than after Gore.
I can see no reason why we should burden voters with having to vote tactically in such a manner. If you absolutely have to have Range or Approval, make a computer find out the poll and vote tactically according to the voter's sincere wishes given on a ranked (or rated) ballot. This idea is called Declared Strategy Voting. Warren himself claims that his DSV version of Range does better than Range.
Furthermore, while IRV behaves badly here, that's just a blemish (among a million others) on IRV, not on ranked voting as a whole. Condorcet methods would do the right thing: if 50% rank A above B above C and 50% rank C above B above A, then B beats A, then B ties both A and C one-on-one. A single voter preferring the broad support candidate B, would then be enough to make B win against both A and C, and so be the victor.
The best strategy in Cumulative voting is to vote plurality-style. You want to make a difference: well, the best way of doing that is pushing all your votes toward the candidate most likely to win that you like (the least of two evils) - that's pretty much what the page says.
Personally, I'd be in favor of a Condorcet method for single-winner and a proportional representation method like STV for multiple winners. The Condorcet criterion simply says that if one candidate is preferred to every other one-on-one, then that candidate should win. It's like sports: if a team beats every other team outright, it should win. The Schulze method, which is a pretty good Condorcet method, is being used by Wikimedia, the Pirate Party of Sweden, and KDE already. It's not very easy to explain, however; if that's a goal, Ranked Pairs is pretty easy and good, too.
Unlike the above, STV has actually been tried in America. New York used it in the 1930s-1940s until the established party machines abolished it by Red Scare tactics. STV's problem wasn't that it didn't work, its problem was that it worked too well. It is indeed interesting that the Republicans, who had no chance of winning pre-STV, actually opposed STV.
One should be very careful about turning the multiwinner system, STV, into a single-winner system (IRV). Some groups in the US are trying to do so, most notably FairVote, and they are linking the concept of the ranked ballot to IRV itself. IRV is not a very good method: while it is more fair than Plurality, as another post here stated, Australia has been using IRV for a very long time and still has a two-party system.
There is such a thing as a type of STV that becomes a Condorcet method when only electing a single winner: Schulze STV, but it is very complex; about the only chance one would have to implement it would be if the voting population could trust the method on performance alone, like a computer or other machine (which most people don't know how works, yet use).
The distortion of standardized test scores as they are applied for optimization purposes is just another example of Campbell's law. When it becomes important to optimize the score, the score gets optimized even at the expense of what it was supposed to measure. As you say, the score may be sensible enough on its own, but optimization twists it.
How about fine-grained security? There's no reason Flash should have access to files on your system, so make the OS support (and withhold) capabilities so Flash just plain can't read your files no matter how compromised it gets. Similarly, there's no reason the image rendering component of your web browser should be able to open a server on port 1337 - or any other port for that matter.
It was done with users: an ordinary Unix user can't write to/usr/bin no matter how hard he tries (unless he escalates privileges). Why not do the same per process? It'll make coding more involved, passing security tokens back and forth, but it'll work.
But if the big bang had a slight antimatter bias, we would have called what we now call antimatter, matter, and vice versa. We'd still say the big bang would have a slight matter bias, because our planet, sun, etc, would be made of "ordinary" (common) matter -- to the eyes of our alternate selves.
Nope. If you're saying a quantum computer can solve NP-complete problems, that's generally believed not to be true (in the same sense that it's generally believed than P != NP). Even if that's not what you're saying, you would have to show that a quantum computer could break, say, AES, in polytime. Grover's algorithm can cut the number of bits required to brute-force in half, but to guard against it, just double the number of bits - 256-bit AES should still be hard to break. So private-key encryption (AES and the likes) are safe unless shown otherwise susceptible.
That leaves public-key crypto, which is probably what you're thinking about. True, with a quantum computer, one can factor numbers in polynomial time (Shor's algorithm), and thus break RSA. But RSA isn't the only public-key cryptography algorithm around. First, even in the worst of worlds short of BQP containing NP, one can make digital signatures using any sort of one-way function: the construct is called a Lamport signature. Second, there are public-key cryptosystems that seem hard to break using quantum computers - the McEliece system, which is based on error-correcting codes, is one of them.
If it turns out BQP contains NP, then there's always quantum cryptography, using the laws of physics to hide the message. However, the practical implementation of quantum cryptography is quite difficult, since if the laser emits more than a single photon for each "burst" (entangled bit), the scheme can be broken. To get over the loss rate implied by single photons, the crypto would either have to be by the use of lasers in free air, or have lots of quantum repeaters along the fiber - more than would be the case for a traditional message.
In any case, if BQP contains NP, we'd have a magic machine that can solve any puzzle that can be quickly verified. Finding optimal solutions to general engineering problems would just be a matter of churning the specs through the machine. The world would change -- very quickly and very radically -- and concerns about crypto would seem slight in comparison.
Well, hooray for a more fragmented Internet. While every keyboard can type A-Za-z, that's not true of Chinese or Arabic, so sites using those TLDs will be effectively off-limits to those that aren't "native". Sure, the sites can also register an ordinary domain name, but then why not just use that domain name to begin with?
I think the source of the term is from ropes made by the British Navy. These ropes had a red thread going through them (one of the many used to make the rope), so that whenever the rope was cut, its origin would be obvious, and so that feature would deter theft. The idea was then used in a metaphorical sense for something that was an integral part of some work ("the easily identifiable thread running through the rope"), and from there to the meaning of the idea that ties a story together (since that is the integral part of the story, without which it would not work).
We could have harnessed fusion in the 1970s, but the scheme would have been impractical. "Simply" fill up a huge underground cavern with a salt that's molten at ordinary temperature, link it to your usual turbine apparatus, and blow up H-bombs in the cavern. The explosion heats the salt, and there you go. See PACER for more information.
Bussard's Polywell fusion reactor (inertial electrostatic confinement) would also harness the energy directly instead of having to go the route through steam and turbines.
Basically, since the reactor is electrostatic, one already has to use electrodes (virtual or otherwise) to set up field potentials. It is then no big deal to add another electrode outside the main system so that the charged particles that are ejected from the fusion reaction itself slows down, turning their energy (kinetic and potential) into electricity. Direct conversion, >90% efficiency (albeit at a few megavolts).
Slashdot Mirror
First, "Information wants to be free" is IMO clueless in itself. Information doesn't want to be free any more than your doorknob wants to be free. You could as easily say "Information wants to be paid for".
I think the right context is "information wants to be free" like "water wants to flow downhill". Sure, you can limit water's progress by building a dam, just like you can encrypt data or otherwise limit the access to information; but in the internet world, information tends to become free (pirates, cracking, etc). In that sense, information is more slippery than water, particularly on general purpose computers, but it is possible to limit it.
Like maybe, personal encryption?
Then it's no longer DRM, which is basically a program that has both lock and key yet tries to hide the key from the user except in "allowed" circumstances. Personal encryption doesn't include both the lock and the key; instead, you have the key and you use it to prove to the lock program that you have (unlimited) access to the encrypted volume/whatever.
Besides, it stands to reason that what you're encrypting is meant to be private, thus, since it's not released, it doesn't fall within the domain of copyright. You're not distributing anything, so limits to distribution don't apply.
Do you know anything about Plutonium?
Some tidbits about the most toxic metal on the planet.
I would suspect that Polonium (you know, what the Russians used to kill Litvinenko) is a mite more toxic.
As far as ingestion in particular is concerned, there is also the caffeine challenge. Granted, caffeine is not a metal, but one wouldn't usually consider it very toxic.
No, rational ignorance doesn't factor into it. If people were rational optimizers with an objective of favorably altering the outcome, they'd quickly find out that even the cost of going to the polls to vote isn't worth the extremely small chance of the increase in marginal utility. Thus they wouldn't vote at all. Yet this is not what happens in the real world, hence people aren't rational optimizers, W5.
How about an automatic sunset: a law that has 50%+1 support gets to live 5 years before it has to be passed again. A law that has 100% support gets to live 10 years before it has to be passed again. Scale linearly between the two to give some incentive to make popular laws, not just squeakers. If that would cause an overload at "pass-again day", add +/- 5% of the duration to the time until it has to be passed again so that the exact day will be sufficiently randomized.
Possible but not likely. Unless a corporation has exclusive access to a needed resource, there's always going to be competition. The reason we don't see as much now in a lot of areas is because it's a legalized monopoly, or the regulations are piled on so thick that only the big boys and play the game. Then you end up with real economic powers controlling your voted-in politicians... no matter who you voted in.
Whether or not a corporation will be able to rewrite the rules depends on whether the barriers to entry can be made high enough that the corporation can then devote itself to gaining power through first accumulating monopoly profit, and second by translating that economic power into physical power. To do the second isn't so hard to imagine: all it would have to do would be to absorb a few of the private defense contractors (in extreme libertarianism, at least). To do the first, there are many effects that could lead to a strong monopoly. For instance, the sector in which it operates might support great economies of scale, so that entrants to the market has no way of competing with the monopoly, or it may support few enough companies that collusion will occur.
The real economic powers controlling the politicians is a form of corruption or collusion in itself. It is powerful in the US because of the broken state of the democratic system. For that reason, among others, I favor an electoral system that gives third parties and independents a real chance: more candidates means that it's much harder to corrupt all of them. For that matter, if you want to be radical, pick a very large assembly randomly and let it elect from among its numbers to Congress - hard to systematically corrupt that.
One might also extend a libertarian "do it yourself" heuristic to all levels: if a decision will affect a certain group, that group (and only that group) should decide whether to do it. Thus, it's the business of nobody else what you do inside your home, but organizations that face externality or have significant power would have to take the external parties into account. The hard part, of course, is to actually implement that heuristic - I don't think Coasian bargaining would work - and some people would say every action affects everybody hence we have a rationale for a large state.
In sum, I suppose I don't really oppose a decentralized society. What I don't like is the capitalist aspect that gets entangled with it in American libertarianism, where the market is considered robust enough to keep corporations from becoming the new states, and consumer democracy good enough to keep those who have power in check. The robber baron era, as well as the case of company towns, seems to show otherwise.
At least you can (in theory) elect your representatives. A corporation has no such flaw: it's a plutocratic oligarchy through and through.
I say "in theory" because the United States implementation of democracy is broken (Gerrymandering, two-party rule, you name it). Yet that is no argument against democracy, merely one against a particular implementation of it.
Lastly... the corporations can't make up stupid laws or arrest you for making dry ice bombs.
In a libertarian system, eventually one of the corporations will end up with enough economic power that it can rewrite the rules of the game. When that happens, that corporation will translate the economic power into physical power and effectively make itself the new state, and then it can make whatever laws it wants.
It's possible to make a quantum repeater (it uses quantum teleportation to move the information without looking at it). Send some classical location data along the quantum encrypted stuff and have the cell tower choose which repeater to use based on the location data, and then you can have more than just point-to-point.
Aligning your laser with that of the tower will be a major pain, though, and you can't just get around it by sending more than one photon: doing so will void the theoretical uncrackability properties of quantum crypto.
Trusted Computing is like a zombie: we keep killing it and it absolutely will not die.
Yes, but IRV branch methods usually all have the following flaws: they're not monotone (moving someone up in rank can make him lose), they're not summable (you have to send voting data for every ballot instead of a sum across candidates, unlike Plurality per-candidate counts or Condorcet matrices), and they're not Condorcet compatible.
IRV based on Borda (Baldwin and Nanson) is Condorcet compatible and I think it is summable, but the monotonicity problem remains. I don't know what Avy is, but the "twin towers" defect seems to suggest that the monotonicity problem remains there as well (Craig Carey's notation can be hard to decipher).
Because IRV has the problems I've outlined, it is true that they also exist in STV, but unlike IRV, there aren't any well known better alternatives. Besides, as the number of seats increases, STV's proportionality criterion (called the Droop proportionality criterion) limits the extent to which the problems distort the result.
You might be interested in this paper, which deals with perverse incentives in the context of education, and gives examples of how metrics are distorted outside of education as well. Their conclusion is that the use of quantitative optimization is, on the whole worth it; I don't agree, but even with the difference of opinion, their examples are illuminating.
Range reduces to Approval, since you want to give every candidate either maximum if you support him or minimum if you don't so your chance of affecting the outcome is as great as possible. Approval, in turn, requires polling and strategy so you can know where to put the cutoff.
Say the three candidate for President is Bush, Gore, and Nader. You like Nader but don't want Bush. If Nader has negligible support, you would vote for Gore and Nader, so that your vote still helps Gore keep Bush away. However, if Nader has substantial support, any vote for both will help Gore win at the expense of Nader, so you might want to reconsider and put the cutoff after Nader rather than after Gore.
I can see no reason why we should burden voters with having to vote tactically in such a manner. If you absolutely have to have Range or Approval, make a computer find out the poll and vote tactically according to the voter's sincere wishes given on a ranked (or rated) ballot. This idea is called Declared Strategy Voting. Warren himself claims that his DSV version of Range does better than Range.
Furthermore, while IRV behaves badly here, that's just a blemish (among a million others) on IRV, not on ranked voting as a whole. Condorcet methods would do the right thing: if 50% rank A above B above C and 50% rank C above B above A, then B beats A, then B ties both A and C one-on-one. A single voter preferring the broad support candidate B, would then be enough to make B win against both A and C, and so be the victor.
The best strategy in Cumulative voting is to vote plurality-style. You want to make a difference: well, the best way of doing that is pushing all your votes toward the candidate most likely to win that you like (the least of two evils) - that's pretty much what the page says.
Personally, I'd be in favor of a Condorcet method for single-winner and a proportional representation method like STV for multiple winners. The Condorcet criterion simply says that if one candidate is preferred to every other one-on-one, then that candidate should win. It's like sports: if a team beats every other team outright, it should win. The Schulze method, which is a pretty good Condorcet method, is being used by Wikimedia, the Pirate Party of Sweden, and KDE already. It's not very easy to explain, however; if that's a goal, Ranked Pairs is pretty easy and good, too.
Unlike the above, STV has actually been tried in America. New York used it in the 1930s-1940s until the established party machines abolished it by Red Scare tactics. STV's problem wasn't that it didn't work, its problem was that it worked too well. It is indeed interesting that the Republicans, who had no chance of winning pre-STV, actually opposed STV.
One should be very careful about turning the multiwinner system, STV, into a single-winner system (IRV). Some groups in the US are trying to do so, most notably FairVote, and they are linking the concept of the ranked ballot to IRV itself. IRV is not a very good method: while it is more fair than Plurality, as another post here stated, Australia has been using IRV for a very long time and still has a two-party system.
There is such a thing as a type of STV that becomes a Condorcet method when only electing a single winner: Schulze STV, but it is very complex; about the only chance one would have to implement it would be if the voting population could trust the method on performance alone, like a computer or other machine (which most people don't know how works, yet use).
The distortion of standardized test scores as they are applied for optimization purposes is just another example of Campbell's law. When it becomes important to optimize the score, the score gets optimized even at the expense of what it was supposed to measure. As you say, the score may be sensible enough on its own, but optimization twists it.
Come on, all these posts and no reference to Minority Report? For shame.
How about fine-grained security? There's no reason Flash should have access to files on your system, so make the OS support (and withhold) capabilities so Flash just plain can't read your files no matter how compromised it gets. Similarly, there's no reason the image rendering component of your web browser should be able to open a server on port 1337 - or any other port for that matter.
/usr/bin no matter how hard he tries (unless he escalates privileges). Why not do the same per process? It'll make coding more involved, passing security tokens back and forth, but it'll work.
It was done with users: an ordinary Unix user can't write to
But if the big bang had a slight antimatter bias, we would have called what we now call antimatter, matter, and vice versa. We'd still say the big bang would have a slight matter bias, because our planet, sun, etc, would be made of "ordinary" (common) matter -- to the eyes of our alternate selves.
And so the wheel of reincarnation turns another notch...
Nope. If you're saying a quantum computer can solve NP-complete problems, that's generally believed not to be true (in the same sense that it's generally believed than P != NP). Even if that's not what you're saying, you would have to show that a quantum computer could break, say, AES, in polytime. Grover's algorithm can cut the number of bits required to brute-force in half, but to guard against it, just double the number of bits - 256-bit AES should still be hard to break. So private-key encryption (AES and the likes) are safe unless shown otherwise susceptible.
That leaves public-key crypto, which is probably what you're thinking about. True, with a quantum computer, one can factor numbers in polynomial time (Shor's algorithm), and thus break RSA. But RSA isn't the only public-key cryptography algorithm around. First, even in the worst of worlds short of BQP containing NP, one can make digital signatures using any sort of one-way function: the construct is called a Lamport signature. Second, there are public-key cryptosystems that seem hard to break using quantum computers - the McEliece system, which is based on error-correcting codes, is one of them.
If it turns out BQP contains NP, then there's always quantum cryptography, using the laws of physics to hide the message. However, the practical implementation of quantum cryptography is quite difficult, since if the laser emits more than a single photon for each "burst" (entangled bit), the scheme can be broken. To get over the loss rate implied by single photons, the crypto would either have to be by the use of lasers in free air, or have lots of quantum repeaters along the fiber - more than would be the case for a traditional message.
In any case, if BQP contains NP, we'd have a magic machine that can solve any puzzle that can be quickly verified. Finding optimal solutions to general engineering problems would just be a matter of churning the specs through the machine. The world would change -- very quickly and very radically -- and concerns about crypto would seem slight in comparison.
You're not allowed to try to eat your own shoes, either - you might be the next shoe bomber!
Good luck doing MITM through SSL.
Well, hooray for a more fragmented Internet. While every keyboard can type A-Za-z, that's not true of Chinese or Arabic, so sites using those TLDs will be effectively off-limits to those that aren't "native". Sure, the sites can also register an ordinary domain name, but then why not just use that domain name to begin with?
I think the source of the term is from ropes made by the British Navy. These ropes had a red thread going through them (one of the many used to make the rope), so that whenever the rope was cut, its origin would be obvious, and so that feature would deter theft. The idea was then used in a metaphorical sense for something that was an integral part of some work ("the easily identifiable thread running through the rope"), and from there to the meaning of the idea that ties a story together (since that is the integral part of the story, without which it would not work).
We could have harnessed fusion in the 1970s, but the scheme would have been impractical. "Simply" fill up a huge underground cavern with a salt that's molten at ordinary temperature, link it to your usual turbine apparatus, and blow up H-bombs in the cavern. The explosion heats the salt, and there you go. See PACER for more information.
Bussard's Polywell fusion reactor (inertial electrostatic confinement) would also harness the energy directly instead of having to go the route through steam and turbines.
Basically, since the reactor is electrostatic, one already has to use electrodes (virtual or otherwise) to set up field potentials. It is then no big deal to add another electrode outside the main system so that the charged particles that are ejected from the fusion reaction itself slows down, turning their energy (kinetic and potential) into electricity. Direct conversion, >90% efficiency (albeit at a few megavolts).