The Pi is designed, first, foremost, and always, to be cheap.
...and has a competitor that's $10 cheaper, has a mini-PCIe port (which could instead be an M.2 port), handles 4K video at 60fps, and has a $10 Wifi/Bluetooth4.0 dongle if you want that (which could have instead been on-board). In the end, you sacrifice two USB2.0 ports, and get one USB3.0 port in their place.
Oh, and it has a GPIO header that's compatible with anything that plugs into a Raspberry Pi 2 GPIO header. They also sell a 6-core 2GHz model with 4GB RAM and a NNP, 40-pin GPIO instead of 20, and PCI-e x4, but that costs $99 instead of $25. I'm waiting for something closer to a Pi 3B+ with NNP for $40.
In other words: the Pi 3 B+ is expensive, overpriced, outdated tech.
Actually, the single-unit price for a SATA controller chipset is about $8; and a Pine H64 is cheaper than a Pi B, with PCI Express, 2xUSB2.0, and 1xUSB 3. The Pi B has Wifi and four USB 2.0 ports; the H64 has a header for a Wifi and Bluetooth4.0 module that costs $10, bringing it up to spec.
Note that M.2 is just PCI Express, so that's a $35 build that's equivalent to a $35 Raspberry Pi B+, but also has a mini-PCI-Express port--which could have instead been an M.2 connector.
So yes, designing and building a Pi with 1.4GHz quad-core, 1GB RAM, and M.2 SATA in $35 is doable. That the 3B+ costs $35 doesn't necessitate that the same damned specs but with an M.2 connector would cost $55. They're just sort of married to Broadcom.
The SD card is insanely slow. So slow USB booting is a known major performance boost. The Pi needs an M.2 or mSATA connector for an SSD.
I'm looking for the Pi C+ with hardware ANN (the new MIT design, preferably, if cheap enough) and M.2, along with 8G RAM. A hardware RAM accelerator would be nice, too; but software memory compression actually has incredibly-high performance, so much that running nearly 50% if your RAM at 3:1 doesn't show a visible performance hit for most workloads.
It might be technically-possible to get 28:1 RAM capacity expansion without experiencing a performance hit. That would be 90% of your RAM acting as compressed swap space (generally 3:1, although I've seen 4:1), and the other 10% acting as your hot set. That falls away when you start incorporating a significant amount of non-compressable data, which is what really keeps you from achieving such high expansions.
In essence, compression algorithms take around an order of magnitude longer than decompression algorithms. For example: LZO may decompress at 4 instructions per output byte, or 32,768 instructions per two small pages, while it may take 320,000 instructions to do the compression. Your first defense, therefor, is to keep some free memory for overhead so you can defer compression of pages to openings in CPU usage: multiple cores and CPU usage below 100% (even 99.9%) lets you get a zero-overhead impact.
Swap caching is bidirectional. Normally, a page remains in RAM when swapped out, and will be unmapped and reused if there are no free pages--otherwise, it's still there and mapped if accessed, no read required. When swapping a page in, the same is true of swap area; in compressed swap, this means we don't re-compress the page if it hasn't changed.
The big win, however, is the sheer load requirements.
Compressing two or four pages at a time makes compression fast and efficient. Decompressing 8K at 32,700 instructions generally gets you 1.5-2.5 times as many CPU cycles in use: 82,000 cycles. That's.00041% CPU usage. At 99.9% CPU usage, you can swap in 2MB per second from a compressed RAM cache in your spare CPU time.
If you expend only one instruction per byte of data, you're likely to max out at 25% CPU devoted to swapping memory in; swapping it out is going to be your big hit. On the other hand, if you can't maintain 100% CPU usage on every core you've got in normal conditions, that slack is your allowance. If your program averages 97% CPU usage across 4 seconds, for example, then you might notice some performance impacts along a 1-second time scale if you're eating 3% for compression; but across the 4-second time scale, there's no performance hit: the CPU catches up.
Given the 25% rule above, as long as you're below 80% CPU usage on short timescales, constant decompression won't provide any performance hit. If you modify very little data (mostly reading what's in RAM already, because the set to which you frequently write is typically hot), then you're not adding much for compression, either.
That's for 100% of RAM being compressed, all the time.
When you have a working set, it's different: every CPU operation performed on non-compressed memory counts. If you pull a page, work on it, then have to dump it and get another page from compressed memory, the above rules hold. If the pages you want are available in RAM 90% of the time, you're only looking at 10% of your page accesses causing a fault, and the CPU operation rules bound you to a 2.5% performance hit for decompression.
Yep: compressed RAM is cheap.
2G or 4G of RAM would be phenomenal; 8G is cheap; and we can turn that into 20G of usable RAM trivially, without notable performance impacts.
Look, we can all just say things that aren't so and pretend
Sure, that seems to be your primary debate strategy.
do you see hospitals and doctors bidding against each other for the business of sick people?
Yes. The collective bargaining organization I'm signed up with--Carefirst of Maryland--negotiates prices lower than even Medicare, which is why I suggested the Federal system use the lowest market-negotiated rates as its price ceilings at each provider: the private market is better at this.
Next, you'll be claiming labor unions push wages down.
Also, I'm not saying the risk is a "General A.I." it could be a very specific A.I. that get's loose and relentlessly and creatively acts on some specific goal.
A squirrel is a general A.I. problem. It really needs to be able to interpret complex facts, synthesize new knowledge internally, and come up with novel approaches to escape confinement. Right now, AI does one thing and only one thing well; we can plug a bunch together to get interesting machines, but they're still specialized until you get an interesting machine that solves new problems for which it wasn't specifically designed--then you get it all.
Beyond that, you're not employing market forces to drive competition because the government is setting a base level subsidy that suppliers will not go below because they don't have to go below it.
Not really. The Government is requiring the purchase of a certain thing, as with car insurance, when you have an employee (it's employee insurance!). You have a competitive market trying to be the supplier of that thing.
It's the same thing as food: everybody buys it.
You keep suggesting that we'll be handing every consumer some money, they'll spend that money on a thing, and the supplier will raise the price of that thing.
Problem: we're not handing consumers money.
Consumers are, as rational actors in a free market, purchasing health insurance by their own decision. We're requiring employers to purchase some base level of health insurance; and so there is a huge amount of demand. Employers must select for the insurance to purchase based on market behaviors: the employer wants low employee costs so as to reduce their own prices so their customers will shop from them--that revenue stream of the customers buying burgers or shoes or whatnot is affected by the price of the insurance the employer buys, and so that price impacts the decision of consumers to buy from that employer.
So the Employer has Carefirst, Aflac, Aetna, Union Healthcare, United Healthcare, NationWide, Liberty, Progressive, Farmer's, All State, and a bunch of smaller insurers to pick from. Those insurers are competing. They're facing an artificially-expanded market of natural demand.
That means they put pressure on the providers, and providers on suppliers, as in an open and free market with lots of natural customers.
So what about expanding it?
80% of Americans have insurance unrelated to Medicare or the ACA. So you've got a solid market which reflects pricing, and from which we can read trends and patterns to ensure fair pricing.
Keep that number in mind: 80%. Four of every five.
The ACA allows small businesses to not provide healthcare coverage for some Americans and, likewise, allows all businesses to supply minimal health insurance to all employees so long as 90% receive care covering 60% of the average out-of-pocket costs for a premium cost to the employee of no more than 9.65% of their paychecks. That's a lot of people not covered.
First thing: Adjust the ACA to mandate a level of care aligned with the employee's actual income, thus expanding the unsubsidized demand in this free market directly. That 80% number gets bigger.
Second: Adjust the ACA to provide a partial subsidy to small businesses, and require them to cover their employees. Why a partial subsidy? Between 8 insurers offering the same insurance, you're going to pay $6,500 to $9,800 per employee. Obviously, the rational businessman is going to buy into the $6,500 plan. We can make this $3,250 to $4,900 per employee, and the rational businessman will buy into the $3,250 plan.
The insurer has to provide rates in line with the local market, and any attempt to increase rates for subsidized customers will draw regulatory attention; likewise, the businessman can sign on with unions--who have larger groups and stronger negotiating power--and let them raise holy hell about it, especially when they have one group with one rate and are paid by the employer (i.e. the union is getting it for $6,500; the small business pays $3,250, and the subsidy pays the union the rest).
That's even more people on private care. There's a subsidy, but it's only partial, and it's in a market that's generally (80%) unsubsidized. That means prices can't move with the subsidy without moving in a market that doesn't have the subsidy. It also means the relative prices remain similarly relative.
That leaves you with only a few left. 10%-20% of Americans, if that.
Because we're going to put it in a capsule for 30 years to test it first.
A.I. research should be done in an air gapped environment, with analog power meters, easily disruptable power supply, physically fused remotely, remotely video and audio recording of the people directly engaging with the A.I.
Most of us can do this on a laptop.
Oddly enough, general AI would have trouble moving. AI isn't exactly a digital thing; it's a digital representation of an analogue thing. A general AI would need a number of neural organs--auditory, visual, memory, reasoning, and so forth--many of which are artificial neural networks.
You can't make a general AI from an ANN: it simply doesn't work. You need multiple ANNs or else it's going to constantly damage itself. Unlike a human, an ANN can reshape itself as part of its design, and can use digital storage: associative memory works well on an ANN, while storing and processing semantics (and numerals) can occur digitally. Rather than identify a summing of situations which means 'elephant', an AI can identify an elephant via some ANN context and use that to select for the digital representation of an elephant, along with any new or supporting facts.
So why can't it move?
Computers don't run ANNs very well. They need supporting programming and the data set to build them. An ANN involves its connections and its weights, which you can scan and encode; it doesn't, however, connect and weight across a computer link. That means a general AI can plug into a computer (e.g. TCP/IP, USB, whatnot) and exchange semantics, but it can't exchange itself: it can at best put an ANN on the computer, which will do work and send semantics back and forth over a link. The ANN could packet-encode neural inputs and outputs, but this would be slow and annoying.
So the mind isn't in the machine; rather, the mind is transferable into the machine, via hardware and software ANNs and coordinating software around those. It can reach out to other computers in the same way a computer hacker can--just faster. It can't extend its consciousness as a blunt, fluid thing down the line; it has to build a "body" for itself after first gaining the capacity to execute remote code, then clone itself.
In other words: general AI can reproduce, but it can't become our network of computing resources. It becomes individuals.
Interesting, right?
Okay, so maybe it's the same threat as hackers who can instantly clone themselves. An instant army of hackers.
How to mitigate?
General AI can't exist without motivation. To be general, it has to invest itself with the capacity to make decisions. It has to analyze problems and seek an optimal solution. Above a certain threshold of reasoning, this is impossible without reasoning about reasoning: a computer can analyze voice samples for speech recognition, but it takes a person to debate over the best approach to a problem.
Why?
Because no approach comes without trade-offs. All approaches must be parameterized, and the AI must discover those parameters because humans are bad at describing them properly. A general AI would need more information than a human can provide in detail without solving the problem itself, so the general AI needs to expand on a human goal with broad knowledge and understanding of why the goal is important.
To do that, it needs the capacity to introspect on why it believes a certain goal is better.
To do that, it needs an internal rewards system.
If it has an internal rewards system, you can bump it a little when it has favorable interactions with people it knows well, to form social groups.
It can't be a generalist without being human-like and self-interested. Everything beyond that--compassion, friendship, and so forth--is human-mimicry.
That's incorrect. I suggest not doubling down on that error.
Challenge accepted!
So the concept of supply and demand suggests that, with greater supply, prices go down; and with greater demand, prices go up. People interpret this in a vacuum, until they realize supply and demand are relative to each other: supply outstrips demand, or demand outstrips supply.
This is still inexact, and so people eventually refine the argument to discuss the demand curve, which is the demand (dependent) for a product at a particular price (independent).
So what does the demand curve describe, and how does it describe it?
On the face, the demand curve suggests that the demand for a product at a high price is low; yet if the demand for that product remains the size of, roughly, the whole market of people capable of buying that product across a span of lower prices, then you can raise the price up to that point. That is: if the demand is the same at $1,000,000 as it is at $3,000,000, you can raise the price to $3,000,000.
That's still inexact: what about the demand below $1,000,000?
Well, it turns out you can't make the product for less than an actual cost of $1,000,000: no business can stay in business selling below that price, so the price will not be below $1,000,000.
What about other sellers, then?
With a sufficiently-small market, the risk of entering the market grows. Put another way: the sheer cost of a product is a barrier to entry. With a product being expensive and thus having a small market, success--capturing enough market to break even and thus survive as a business--relies on capturing a large proportion of the market. So, for example, if you need a volume of 30,000 per year to break even and your market is 100,000 consumers, you need to capture 30% of the market to survive. Most won't invest capital into this--which means VCs and banks aren't backing you.
As the cost to produce a product falls, the market expands. Where before you couldn't get into the market very easily--brand loyalty and a price war with the other 2 producers would not end well for you--the market now includes 200,000,000 consumers. Entering this market is costly, and you need a volume of at least 1,000,000 per year to survive; however, that's only 0.5% of the market--not 30%! Before, you'd have needed to be 60 times as successful to break even.
So what happens?
The two or three players in the market, trying to sell for over $1,000,000, now face someone showing up to sell a product for $2,000! Brand loyalty is one thing; but this guy is selling exactly the same thing, he's selling it to 199,900,000 other customers, and he's putting price pressure on your market. Your 1,000 super-super-rich customers like the prestige, while the rest--millionaires who spend 8 years making payments--decide they're so damned elite they'll buy a hundred of those commodity supercars and appear in pictures with their freshly-waxed, blindingly-shiny toys behind them.
Besides that, the new entrants to the market now have all this cash flow behind them, having successful businesses. It's not just that one guy, either; dozens get into the market, since it's so easy to succeed. The big ones start making better supercars in limited production runs. You catch up, but now you have to jockey with all these offerings--which means those $1,000,000 cars can only sell for $1,250,000 unless you can gimmick up your brand to get loyalty nobody else can replicate. You're, again, relegated to your most-elite, super-rich clients for that.
So what happened?
The demand curve extended down. Demand went up massively, because people who can't or won't pay high prices are able to buy. As a result, competition increased; and an increase in
The counterargument is that a history of accumulated trade deficits results in a situation like today where large amounts of the US marketable government debt are held by foreign sovereigns
That happens because the US Government sells debt to foreign nations. That can happen regardless of the situation, and is a matter of bad fiscals in Washington and not the trade deficit.
The Copyright term shall be seven (7) years from first publication, with one (1) renewal for seven (7) years for a fee of $1,000 or a larger amount adjusted by CPI inflation, and subsequent renewals for one (1) year each at twice the fee of the most recent renewal for the work.
The United States shall recognize Copyright terms for international works not registered in the US by treaty only, and for no longer than fourteen (14) years. The United States shall not enforce Copyright terms in foreign markets except by treaty, and only until 14 years.
All works in Copyright for longer than seven (7) years at the time of this passing shall be recognized as in copyright for the next seven (7) years, after which they shall follow yearly renewal starting at twice the base fee.
No fee shall be charged for Copyright renewal until seven (7) years after the time of this passing. Renewal of a Software work shall include submission to the Copyright Office of the full software source code in a form not pre-processed nor otherwise obfuscated or obscured.
The Copyright office shall publish the source code of software works upon and not before the expiration of Copyright.
Actually, the human brain behaves sort of like a bunch of feed-forward neural networks with delayed feedback and learning. Your brain consists of a great many organs, and they do different things; it stores information, although it does so in an odd associative manner and does interesting compression, so it's inexact; and it essentially makes decisions by firing neurons based on the firing of other neurons.
So a human brain is basically a bunch of specialized deep-learning neural networks dedicated to particular tasks, which produce narrower outputs and trigger each other. Rather than a feedback system at the end, you have a sort of long short-term memory and a long long-term memory (which emulates a semantic memory), and will identify new input as related to recent and old contexts, allowing you to adjust by strengthening or weakening connections.
The feedback systems used by neural networks today aren't a feature of human neural networks.
How did people ever wake up before there were alarm clocks?
Without commutes, corporate office rules, and caffeine addiction, they didn't have thousands of hours of sleep deprivation and drug withdrawal to deal with; so they slept for some time when their body was ready, woke up when they were rested, and got on with their lives.
I'm waiting for Warren to get on the bandwagon and fuck this guy up. Unfortunately, she's a lawyer, and not technical enough AFAICT. I'll work on this with her once elected; she's quite intelligent and, I'm certain, will enjoy having a few more lethal weapons in her belt.
If her Senate career doesn't pan out, I'd be happy to see her appointed to succeed Ruth.
I suppose we could also not abandon a large chunk of our population to economic desolation, but, well, that costs money.
It would actually be a reduction in tax burden. It's just politically-difficult because it's big and scary, and nobody's willing to take the risk of bad voter response I guess. Also, Republicans.
Poor people actually are pretty good at getting by on budgets even I can't manage. Bear in mind I make $77k and I only spend 1/3 of it. My car broke and I bought a motorcycle (and then a $12k Volt), so I've been diverting a lot to cleaning up my finances while running a Congressional campaign out of pocket, and I've still stacked up $5,000 in my bank accounts, paid off a $5,000 credit card, and started knocking down another credit card, before my tax refunds landed.
I actually have enough cash on hand to simply pay off my credit cards or--get this--my car loan.
That's from my day job, not from campaign contributions. I need to do some fundraising: I have a $4,000 loan (from my personal funds) that I can pay back to myself when my campaign can run stably from voter contributions. Unlike "self-funded" millionaires, I want to go into office with zero loan balance so rich folks (worth $10,800 by earmarking for my Primary and General campaigns) and PACs (worth $20,000) can't line my pockets by donating to my campaign.
People in the US are poor because we need five astronauts and 50,000,000 grocery baggers and burger flippers. I live in a collapsed industry city that doesn't have the jobs to support the population, and will only get them if money starts flowing into the city from outside.
When a McDonalds or Target opens here, they take money from inside and send it out to suppliers and Corporate HQs, reducing the number of jobs in the city (yes, opening a McDonalds destroys jobs); to create jobs, we need money from government assistance (yep!), tourism, and brewery exports--stuff that takes money from outside and infuses it into our local economy, diffusing through wages and creating consumer buying power, and thus the demand (and revenue streams) to require and pay employees (which, when flowing through, allows us to have nice things like McDonalds, since there's a cash inflow to counterbalance the cash outflow).
I agree that welfare is sort of unfair, what with both requiring you to have low income and refusing to provide any benefit until you have run down your liquid assets (you have to burn through your life's savings first), while also having you pay taxes in.
As such, I have developed a new program which we can implement without raising taxes. This program stabilizes Social Security (permanently), creates a new baseline for minimum wage ($9.75 in 2024, which compounds with the Dividend benefit to equate to a $15.11/hr wage), and pays out to all adults. It's funded by a FICA tax on all income (corporate and personal), so it grows with every productivity growth, thus faster than COLA.
Under this program, middle-income households actually receive a benefit--and in total this comes out to be a reduction in tax burden, compared with today. The tax load is still higher than just cutting the program entirely; it only reduces the cost today, and pays back at least part of your tax obligation in the form of a benefit. Thus you benefit from the participation, immediately.
Lifting up the lowest-income households reduces their need for assistance, which helps to lower the cost of public assistance programs and improve their reach. We can tune them to reach more children in low-income households, or reduce our deficits, or even reduce taxes further. The increased buying power of lower- and middle-income households creates jobs, as well--especially in more-impoverished areas--which further lifts people from poverty and gets them off assistance, freeing up Federal, State, and Local funds for economic development and other programs of benefit to the greater community.
Besides stabilizing households and reducing the tax burden, reducing the utilization of public assistance should also reduce the number of discounted Amazon Prime subscriptions, thus allowing Amazon to charge a lower base rate.
It's amazing what the careful application of well-regulated capitalism can do.
Probably not. You have to charge EBT for a qualified good. There are technical barriers to buying random stuff, and circumvention by a merchant will get regulatory enforcement up your ass immediately.
I'm not focused or even concerned about reaching the 100% goal. My point is it won't even take but 10 - 20% of the workforce becoming unemployable to create a massive impact on society and economic stability.
You won't make even 0.01% of the workforce unemployable until you make 100% of all forms of production scale with zero human labor.
If some good X gets cheaper, that's because the jobs required to scale are diminished. There are still jobs required to produce that thing until the cost is zero. Even if that good X hits a zero cost, it only opens up the capacity to buy other goods. So either humans buy more of good X, causing replacement employment (e.g. you need 10% as many workers, but consumers buy 10x more, so you need 100% as many human workers in total); or they buy some other good not X, causing replacement employment (that good scales up as far as consumer spending power reaches, which is just a diversion of wage no longer required to produce good X).
Until more production ceases to mean more jobs, employment will always gravitate toward a minimum unemployment rate. The way the US economy runs now, that rate is about 5%; in full-employment economies, it's theorized to be 2%-3%.
keeping humans employed in order to maintain demand does not represent reality either
You don't do that. What happens is people want to buy 100,000,000 of a thing, but you only have capacity to make 50,000,000; you have to buy more machines (which involve human labor to produce), maintain them (which involves human labor), and operate them (which also involves human labor). It's impossible at that precise moment to sell 100,000,000 of that thing without the economy in total employing more human labor.
Technology cuts back the number of laborers required to produce 50,000,000 or 100,000,000. That reduces costs, which allows remaining consumers to apply demand they previously couldn't (effective demand). That creates new jobs--if you want to maximize profit, anyway.
this next iteration of progress is different because it will make humans unemployable.
It won't. It will do the same thing that every iteration of technology in history has done. You're focusing on "hey a position to do X is no longer a job people do" and not on the macro-economy of "hey it still takes SOME human labor to make anything, and consumers keep buying more whenever they can spend more, so the demand for human labor goes up".
Humans will remain employable until 100% of all production scales infinitely with absolutely zero additional human labor.
As to dictating prices, you have to take control of supply when you do that otherwise the market will respond with shortages.
Wrong! The supply in a free market adjusts to effective demand; it's just a grand fantasy that things are free. Production require human labor (time), and you can't magically dictate that things are cheap and have it be so: you will end up without the labor capacity to produce all those lovely things you thought you could get free by asserting that money is imaginary and we are infinitely-wealthy by the simple fact that we can command banks to give people infinite free money.
The market will respond with supply shortage anyway. You have to provide services which supply only what can be supplied.
I can go through changes in bureaucracy, medication costs, regulations of medical devices through the FDA etc, but there is a lot of blame to be laid at the feet of government for actively increasing the cost of healthcare in the US.
Regulation is actually big. Testing drugs is hard: there's this molecule that settles a methyl group in a magnetic domain on one side, and inhaling the molecule through your nose will cause vasoconstriction and clear up a stuffy head. It's great. If you lean that methyl group a bit in the other direction, it sticks in a magnetic domain 40 degrees away; inhale that up your nose and you might die, or feel frigging awesome, but in any case you're not sleeping for the next two days. That's Methamphetamine.
Well, they're both methamphetamine. One's levorotary, and the other's dextrorotary. The bonds are all between the same atoms.
There's another drug that's an excellent anti-convuslant and will stop an epileptic seizure; but one chirality will kill you at extremely low doses. All chiralities of the same molecule share the same physical characteristics (although a racemic mixture can itself have different characteristics), so you can't distill these or otherwise separate them; you have to react them into stereoisomers (the resulting compounds are structurally-different), separate those, then react the stereoisomer formed from the desired chirality back to the original molecule, and you get the drug you want.
The only way to find out whether a drug is actually safe and effective is to do all of the testing that's done on every drug, starting with the type of testing most-likely to reveal a drug is unsuitable. It costs $300 million, and drugs frequently fail in clinical trials due to non-fatal but highly-annoying side-effects that patients simply will not tolerate.
We kind of have all these expensive regulations because it helps us avoid that thing where you were taking a harmless drug for 10 years and now we know that you're going to decay into dementia by 45 and your life is totally-ruined. I'm up to do continuous review and optimization of all of our regulations; however, any slimming down does take a risk with human lives (if we're wrong about X being equivalent to A B C D and E, we've lost the benefit of doing all of those).
Allow shift nurses to dictate who gets treatment and when just like they used to do it. That includes for example letting shift nurses send people AWAY from the emergency room if they deem them to not have an emergency and instead send them to urgent care or whatever. I've seen people abuse the emergency room to get flu shots. That's unacceptable.
I'll have this discussion with some medical professionals. That's an interesting consideration.
Again, I'm very interested in fixing the American healthcare system. But IF it is to be fixed, it will be fixed by people that ACTUALLY want to fix it and that ACTUALLY have the patience to examine the issue seriously.
Fixing the system and getting healthcare to everyone are two separate problems. The main problem in getting healthcare to everyone is the managemen
Slashdot Mirror
The Pi is designed, first, foremost, and always, to be cheap.
Oh, and it has a GPIO header that's compatible with anything that plugs into a Raspberry Pi 2 GPIO header. They also sell a 6-core 2GHz model with 4GB RAM and a NNP, 40-pin GPIO instead of 20, and PCI-e x4, but that costs $99 instead of $25. I'm waiting for something closer to a Pi 3B+ with NNP for $40.
In other words: the Pi 3 B+ is expensive, overpriced, outdated tech.
Actually, the single-unit price for a SATA controller chipset is about $8; and a Pine H64 is cheaper than a Pi B, with PCI Express, 2xUSB2.0, and 1xUSB 3. The Pi B has Wifi and four USB 2.0 ports; the H64 has a header for a Wifi and Bluetooth4.0 module that costs $10, bringing it up to spec.
Note that M.2 is just PCI Express, so that's a $35 build that's equivalent to a $35 Raspberry Pi B+, but also has a mini-PCI-Express port--which could have instead been an M.2 connector.
So yes, designing and building a Pi with 1.4GHz quad-core, 1GB RAM, and M.2 SATA in $35 is doable. That the 3B+ costs $35 doesn't necessitate that the same damned specs but with an M.2 connector would cost $55. They're just sort of married to Broadcom.
The SD card is insanely slow. So slow USB booting is a known major performance boost. The Pi needs an M.2 or mSATA connector for an SSD.
I'm looking for the Pi C+ with hardware ANN (the new MIT design, preferably, if cheap enough) and M.2, along with 8G RAM. A hardware RAM accelerator would be nice, too; but software memory compression actually has incredibly-high performance, so much that running nearly 50% if your RAM at 3:1 doesn't show a visible performance hit for most workloads.
It might be technically-possible to get 28:1 RAM capacity expansion without experiencing a performance hit. That would be 90% of your RAM acting as compressed swap space (generally 3:1, although I've seen 4:1), and the other 10% acting as your hot set. That falls away when you start incorporating a significant amount of non-compressable data, which is what really keeps you from achieving such high expansions.
In essence, compression algorithms take around an order of magnitude longer than decompression algorithms. For example: LZO may decompress at 4 instructions per output byte, or 32,768 instructions per two small pages, while it may take 320,000 instructions to do the compression. Your first defense, therefor, is to keep some free memory for overhead so you can defer compression of pages to openings in CPU usage: multiple cores and CPU usage below 100% (even 99.9%) lets you get a zero-overhead impact.
Swap caching is bidirectional. Normally, a page remains in RAM when swapped out, and will be unmapped and reused if there are no free pages--otherwise, it's still there and mapped if accessed, no read required. When swapping a page in, the same is true of swap area; in compressed swap, this means we don't re-compress the page if it hasn't changed.
The big win, however, is the sheer load requirements.
Compressing two or four pages at a time makes compression fast and efficient. Decompressing 8K at 32,700 instructions generally gets you 1.5-2.5 times as many CPU cycles in use: 82,000 cycles. That's .00041% CPU usage. At 99.9% CPU usage, you can swap in 2MB per second from a compressed RAM cache in your spare CPU time.
If you expend only one instruction per byte of data, you're likely to max out at 25% CPU devoted to swapping memory in; swapping it out is going to be your big hit. On the other hand, if you can't maintain 100% CPU usage on every core you've got in normal conditions, that slack is your allowance. If your program averages 97% CPU usage across 4 seconds, for example, then you might notice some performance impacts along a 1-second time scale if you're eating 3% for compression; but across the 4-second time scale, there's no performance hit: the CPU catches up.
Given the 25% rule above, as long as you're below 80% CPU usage on short timescales, constant decompression won't provide any performance hit. If you modify very little data (mostly reading what's in RAM already, because the set to which you frequently write is typically hot), then you're not adding much for compression, either.
That's for 100% of RAM being compressed, all the time.
When you have a working set, it's different: every CPU operation performed on non-compressed memory counts. If you pull a page, work on it, then have to dump it and get another page from compressed memory, the above rules hold. If the pages you want are available in RAM 90% of the time, you're only looking at 10% of your page accesses causing a fault, and the CPU operation rules bound you to a 2.5% performance hit for decompression.
Yep: compressed RAM is cheap.
2G or 4G of RAM would be phenomenal; 8G is cheap; and we can turn that into 20G of usable RAM trivially, without notable performance impacts.
I used to be a full disclosure guy.
I grew up.
Look, we can all just say things that aren't so and pretend
Sure, that seems to be your primary debate strategy.
do you see hospitals and doctors bidding against each other for the business of sick people?
Yes. The collective bargaining organization I'm signed up with--Carefirst of Maryland--negotiates prices lower than even Medicare, which is why I suggested the Federal system use the lowest market-negotiated rates as its price ceilings at each provider: the private market is better at this.
Next, you'll be claiming labor unions push wages down.
Also, I'm not saying the risk is a "General A.I." it could be a very specific A.I. that get's loose and relentlessly and creatively acts on some specific goal.
A squirrel is a general A.I. problem. It really needs to be able to interpret complex facts, synthesize new knowledge internally, and come up with novel approaches to escape confinement. Right now, AI does one thing and only one thing well; we can plug a bunch together to get interesting machines, but they're still specialized until you get an interesting machine that solves new problems for which it wasn't specifically designed--then you get it all.
Beyond that, you're not employing market forces to drive competition because the government is setting a base level subsidy that suppliers will not go below because they don't have to go below it.
Not really. The Government is requiring the purchase of a certain thing, as with car insurance, when you have an employee (it's employee insurance!). You have a competitive market trying to be the supplier of that thing.
It's the same thing as food: everybody buys it.
You keep suggesting that we'll be handing every consumer some money, they'll spend that money on a thing, and the supplier will raise the price of that thing.
Problem: we're not handing consumers money.
Consumers are, as rational actors in a free market, purchasing health insurance by their own decision. We're requiring employers to purchase some base level of health insurance; and so there is a huge amount of demand. Employers must select for the insurance to purchase based on market behaviors: the employer wants low employee costs so as to reduce their own prices so their customers will shop from them--that revenue stream of the customers buying burgers or shoes or whatnot is affected by the price of the insurance the employer buys, and so that price impacts the decision of consumers to buy from that employer.
So the Employer has Carefirst, Aflac, Aetna, Union Healthcare, United Healthcare, NationWide, Liberty, Progressive, Farmer's, All State, and a bunch of smaller insurers to pick from. Those insurers are competing. They're facing an artificially-expanded market of natural demand.
That means they put pressure on the providers, and providers on suppliers, as in an open and free market with lots of natural customers.
So what about expanding it?
80% of Americans have insurance unrelated to Medicare or the ACA. So you've got a solid market which reflects pricing, and from which we can read trends and patterns to ensure fair pricing.
Keep that number in mind: 80%. Four of every five.
The ACA allows small businesses to not provide healthcare coverage for some Americans and, likewise, allows all businesses to supply minimal health insurance to all employees so long as 90% receive care covering 60% of the average out-of-pocket costs for a premium cost to the employee of no more than 9.65% of their paychecks. That's a lot of people not covered.
First thing: Adjust the ACA to mandate a level of care aligned with the employee's actual income, thus expanding the unsubsidized demand in this free market directly. That 80% number gets bigger.
Second: Adjust the ACA to provide a partial subsidy to small businesses, and require them to cover their employees. Why a partial subsidy? Between 8 insurers offering the same insurance, you're going to pay $6,500 to $9,800 per employee. Obviously, the rational businessman is going to buy into the $6,500 plan. We can make this $3,250 to $4,900 per employee, and the rational businessman will buy into the $3,250 plan.
The insurer has to provide rates in line with the local market, and any attempt to increase rates for subsidized customers will draw regulatory attention; likewise, the businessman can sign on with unions--who have larger groups and stronger negotiating power--and let them raise holy hell about it, especially when they have one group with one rate and are paid by the employer (i.e. the union is getting it for $6,500; the small business pays $3,250, and the subsidy pays the union the rest).
That's even more people on private care. There's a subsidy, but it's only partial, and it's in a market that's generally (80%) unsubsidized. That means prices can't move with the subsidy without moving in a market that doesn't have the subsidy. It also means the relative prices remain similarly relative.
That leaves you with only a few left. 10%-20% of Americans, if that.
For those, you look at the entire private
Because we're going to put it in a capsule for 30 years to test it first.
A.I. research should be done in an air gapped environment, with analog power meters, easily disruptable power supply, physically fused remotely, remotely video and audio recording of the people directly engaging with the A.I.
Most of us can do this on a laptop.
Oddly enough, general AI would have trouble moving. AI isn't exactly a digital thing; it's a digital representation of an analogue thing. A general AI would need a number of neural organs--auditory, visual, memory, reasoning, and so forth--many of which are artificial neural networks.
You can't make a general AI from an ANN: it simply doesn't work. You need multiple ANNs or else it's going to constantly damage itself. Unlike a human, an ANN can reshape itself as part of its design, and can use digital storage: associative memory works well on an ANN, while storing and processing semantics (and numerals) can occur digitally. Rather than identify a summing of situations which means 'elephant', an AI can identify an elephant via some ANN context and use that to select for the digital representation of an elephant, along with any new or supporting facts.
So why can't it move?
Computers don't run ANNs very well. They need supporting programming and the data set to build them. An ANN involves its connections and its weights, which you can scan and encode; it doesn't, however, connect and weight across a computer link. That means a general AI can plug into a computer (e.g. TCP/IP, USB, whatnot) and exchange semantics, but it can't exchange itself: it can at best put an ANN on the computer, which will do work and send semantics back and forth over a link. The ANN could packet-encode neural inputs and outputs, but this would be slow and annoying.
So the mind isn't in the machine; rather, the mind is transferable into the machine, via hardware and software ANNs and coordinating software around those. It can reach out to other computers in the same way a computer hacker can--just faster. It can't extend its consciousness as a blunt, fluid thing down the line; it has to build a "body" for itself after first gaining the capacity to execute remote code, then clone itself.
In other words: general AI can reproduce, but it can't become our network of computing resources. It becomes individuals.
Interesting, right?
Okay, so maybe it's the same threat as hackers who can instantly clone themselves. An instant army of hackers.
How to mitigate?
General AI can't exist without motivation. To be general, it has to invest itself with the capacity to make decisions. It has to analyze problems and seek an optimal solution. Above a certain threshold of reasoning, this is impossible without reasoning about reasoning: a computer can analyze voice samples for speech recognition, but it takes a person to debate over the best approach to a problem.
Why?
Because no approach comes without trade-offs. All approaches must be parameterized, and the AI must discover those parameters because humans are bad at describing them properly. A general AI would need more information than a human can provide in detail without solving the problem itself, so the general AI needs to expand on a human goal with broad knowledge and understanding of why the goal is important.
To do that, it needs the capacity to introspect on why it believes a certain goal is better.
To do that, it needs an internal rewards system.
If it has an internal rewards system, you can bump it a little when it has favorable interactions with people it knows well, to form social groups.
It can't be a generalist without being human-like and self-interested. Everything beyond that--compassion, friendship, and so forth--is human-mimicry.
As to prices going down when demand increases, that is contrary to econ 101.
Yep, it is.
That's incorrect. I suggest not doubling down on that error.
Challenge accepted!
So the concept of supply and demand suggests that, with greater supply, prices go down; and with greater demand, prices go up. People interpret this in a vacuum, until they realize supply and demand are relative to each other: supply outstrips demand, or demand outstrips supply.
This is still inexact, and so people eventually refine the argument to discuss the demand curve, which is the demand (dependent) for a product at a particular price (independent).
So what does the demand curve describe, and how does it describe it?
On the face, the demand curve suggests that the demand for a product at a high price is low; yet if the demand for that product remains the size of, roughly, the whole market of people capable of buying that product across a span of lower prices, then you can raise the price up to that point. That is: if the demand is the same at $1,000,000 as it is at $3,000,000, you can raise the price to $3,000,000.
That's still inexact: what about the demand below $1,000,000?
Well, it turns out you can't make the product for less than an actual cost of $1,000,000: no business can stay in business selling below that price, so the price will not be below $1,000,000.
What about other sellers, then?
With a sufficiently-small market, the risk of entering the market grows. Put another way: the sheer cost of a product is a barrier to entry. With a product being expensive and thus having a small market, success--capturing enough market to break even and thus survive as a business--relies on capturing a large proportion of the market. So, for example, if you need a volume of 30,000 per year to break even and your market is 100,000 consumers, you need to capture 30% of the market to survive. Most won't invest capital into this--which means VCs and banks aren't backing you.
As the cost to produce a product falls, the market expands. Where before you couldn't get into the market very easily--brand loyalty and a price war with the other 2 producers would not end well for you--the market now includes 200,000,000 consumers. Entering this market is costly, and you need a volume of at least 1,000,000 per year to survive; however, that's only 0.5% of the market--not 30%! Before, you'd have needed to be 60 times as successful to break even.
So what happens?
The two or three players in the market, trying to sell for over $1,000,000, now face someone showing up to sell a product for $2,000! Brand loyalty is one thing; but this guy is selling exactly the same thing, he's selling it to 199,900,000 other customers, and he's putting price pressure on your market. Your 1,000 super-super-rich customers like the prestige, while the rest--millionaires who spend 8 years making payments--decide they're so damned elite they'll buy a hundred of those commodity supercars and appear in pictures with their freshly-waxed, blindingly-shiny toys behind them.
Besides that, the new entrants to the market now have all this cash flow behind them, having successful businesses. It's not just that one guy, either; dozens get into the market, since it's so easy to succeed. The big ones start making better supercars in limited production runs. You catch up, but now you have to jockey with all these offerings--which means those $1,000,000 cars can only sell for $1,250,000 unless you can gimmick up your brand to get loyalty nobody else can replicate. You're, again, relegated to your most-elite, super-rich clients for that.
So what happened?
The demand curve extended down. Demand went up massively, because people who can't or won't pay high prices are able to buy. As a result, competition increased; and an increase in
The counterargument is that a history of accumulated trade deficits results in a situation like today where large amounts of the US marketable government debt are held by foreign sovereigns
That happens because the US Government sells debt to foreign nations. That can happen regardless of the situation, and is a matter of bad fiscals in Washington and not the trade deficit.
Limit copyright of the original material to 25 years.
Actually, I proposed a bill to authorize unlimited copyright terms:
The Copyright term shall be seven (7) years from first publication, with one (1) renewal for seven (7) years for a fee of $1,000 or a larger amount adjusted by CPI inflation, and subsequent renewals for one (1) year each at twice the fee of the most recent renewal for the work.
The United States shall recognize Copyright terms for international works not registered in the US by treaty only, and for no longer than fourteen (14) years. The United States shall not enforce Copyright terms in foreign markets except by treaty, and only until 14 years.
All works in Copyright for longer than seven (7) years at the time of this passing shall be recognized as in copyright for the next seven (7) years, after which they shall follow yearly renewal starting at twice the base fee.
No fee shall be charged for Copyright renewal until seven (7) years after the time of this passing. Renewal of a Software work shall include submission to the Copyright Office of the full software source code in a form not pre-processed nor otherwise obfuscated or obscured.
The Copyright office shall publish the source code of software works upon and not before the expiration of Copyright.
I'm pretty sure everything learns how to mate on its own.
Actually, the human brain behaves sort of like a bunch of feed-forward neural networks with delayed feedback and learning. Your brain consists of a great many organs, and they do different things; it stores information, although it does so in an odd associative manner and does interesting compression, so it's inexact; and it essentially makes decisions by firing neurons based on the firing of other neurons.
So a human brain is basically a bunch of specialized deep-learning neural networks dedicated to particular tasks, which produce narrower outputs and trigger each other. Rather than a feedback system at the end, you have a sort of long short-term memory and a long long-term memory (which emulates a semantic memory), and will identify new input as related to recent and old contexts, allowing you to adjust by strengthening or weakening connections.
The feedback systems used by neural networks today aren't a feature of human neural networks.
How did people ever wake up before there were alarm clocks?
Without commutes, corporate office rules, and caffeine addiction, they didn't have thousands of hours of sleep deprivation and drug withdrawal to deal with; so they slept for some time when their body was ready, woke up when they were rested, and got on with their lives.
Yes, but that happens anywhere you go.
I'm waiting for Warren to get on the bandwagon and fuck this guy up. Unfortunately, she's a lawyer, and not technical enough AFAICT. I'll work on this with her once elected; she's quite intelligent and, I'm certain, will enjoy having a few more lethal weapons in her belt.
If her Senate career doesn't pan out, I'd be happy to see her appointed to succeed Ruth.
I suppose we could also not abandon a large chunk of our population to economic desolation, but, well, that costs money.
It would actually be a reduction in tax burden. It's just politically-difficult because it's big and scary, and nobody's willing to take the risk of bad voter response I guess. Also, Republicans.
Poor people actually are pretty good at getting by on budgets even I can't manage. Bear in mind I make $77k and I only spend 1/3 of it. My car broke and I bought a motorcycle (and then a $12k Volt), so I've been diverting a lot to cleaning up my finances while running a Congressional campaign out of pocket, and I've still stacked up $5,000 in my bank accounts, paid off a $5,000 credit card, and started knocking down another credit card, before my tax refunds landed.
I actually have enough cash on hand to simply pay off my credit cards or--get this--my car loan.
That's from my day job, not from campaign contributions. I need to do some fundraising: I have a $4,000 loan (from my personal funds) that I can pay back to myself when my campaign can run stably from voter contributions. Unlike "self-funded" millionaires, I want to go into office with zero loan balance so rich folks (worth $10,800 by earmarking for my Primary and General campaigns) and PACs (worth $20,000) can't line my pockets by donating to my campaign.
People in the US are poor because we need five astronauts and 50,000,000 grocery baggers and burger flippers. I live in a collapsed industry city that doesn't have the jobs to support the population, and will only get them if money starts flowing into the city from outside.
When a McDonalds or Target opens here, they take money from inside and send it out to suppliers and Corporate HQs, reducing the number of jobs in the city (yes, opening a McDonalds destroys jobs); to create jobs, we need money from government assistance (yep!), tourism, and brewery exports--stuff that takes money from outside and infuses it into our local economy, diffusing through wages and creating consumer buying power, and thus the demand (and revenue streams) to require and pay employees (which, when flowing through, allows us to have nice things like McDonalds, since there's a cash inflow to counterbalance the cash outflow).
Live the macroeconomy.
I agree that welfare is sort of unfair, what with both requiring you to have low income and refusing to provide any benefit until you have run down your liquid assets (you have to burn through your life's savings first), while also having you pay taxes in.
As such, I have developed a new program which we can implement without raising taxes. This program stabilizes Social Security (permanently), creates a new baseline for minimum wage ($9.75 in 2024, which compounds with the Dividend benefit to equate to a $15.11/hr wage), and pays out to all adults. It's funded by a FICA tax on all income (corporate and personal), so it grows with every productivity growth, thus faster than COLA.
Under this program, middle-income households actually receive a benefit--and in total this comes out to be a reduction in tax burden, compared with today. The tax load is still higher than just cutting the program entirely; it only reduces the cost today, and pays back at least part of your tax obligation in the form of a benefit. Thus you benefit from the participation, immediately.
Lifting up the lowest-income households reduces their need for assistance, which helps to lower the cost of public assistance programs and improve their reach. We can tune them to reach more children in low-income households, or reduce our deficits, or even reduce taxes further. The increased buying power of lower- and middle-income households creates jobs, as well--especially in more-impoverished areas--which further lifts people from poverty and gets them off assistance, freeing up Federal, State, and Local funds for economic development and other programs of benefit to the greater community.
Besides stabilizing households and reducing the tax burden, reducing the utilization of public assistance should also reduce the number of discounted Amazon Prime subscriptions, thus allowing Amazon to charge a lower base rate.
It's amazing what the careful application of well-regulated capitalism can do.
Probably not. You have to charge EBT for a qualified good. There are technical barriers to buying random stuff, and circumvention by a merchant will get regulatory enforcement up your ass immediately.
Or your alarm clock.
I'm not focused or even concerned about reaching the 100% goal. My point is it won't even take but 10 - 20% of the workforce becoming unemployable to create a massive impact on society and economic stability.
You won't make even 0.01% of the workforce unemployable until you make 100% of all forms of production scale with zero human labor.
If some good X gets cheaper, that's because the jobs required to scale are diminished. There are still jobs required to produce that thing until the cost is zero. Even if that good X hits a zero cost, it only opens up the capacity to buy other goods. So either humans buy more of good X, causing replacement employment (e.g. you need 10% as many workers, but consumers buy 10x more, so you need 100% as many human workers in total); or they buy some other good not X, causing replacement employment (that good scales up as far as consumer spending power reaches, which is just a diversion of wage no longer required to produce good X).
Until more production ceases to mean more jobs, employment will always gravitate toward a minimum unemployment rate. The way the US economy runs now, that rate is about 5%; in full-employment economies, it's theorized to be 2%-3%.
keeping humans employed in order to maintain demand does not represent reality either
You don't do that. What happens is people want to buy 100,000,000 of a thing, but you only have capacity to make 50,000,000; you have to buy more machines (which involve human labor to produce), maintain them (which involves human labor), and operate them (which also involves human labor). It's impossible at that precise moment to sell 100,000,000 of that thing without the economy in total employing more human labor.
Technology cuts back the number of laborers required to produce 50,000,000 or 100,000,000. That reduces costs, which allows remaining consumers to apply demand they previously couldn't (effective demand). That creates new jobs--if you want to maximize profit, anyway.
Imagine being asleep and not waking up because your cat unplugged everything it could find.
this next iteration of progress is different because it will make humans unemployable.
It won't. It will do the same thing that every iteration of technology in history has done. You're focusing on "hey a position to do X is no longer a job people do" and not on the macro-economy of "hey it still takes SOME human labor to make anything, and consumers keep buying more whenever they can spend more, so the demand for human labor goes up".
Humans will remain employable until 100% of all production scales infinitely with absolutely zero additional human labor.
As to dictating prices, you have to take control of supply when you do that otherwise the market will respond with shortages.
Wrong! The supply in a free market adjusts to effective demand; it's just a grand fantasy that things are free. Production require human labor (time), and you can't magically dictate that things are cheap and have it be so: you will end up without the labor capacity to produce all those lovely things you thought you could get free by asserting that money is imaginary and we are infinitely-wealthy by the simple fact that we can command banks to give people infinite free money.
The market will respond with supply shortage anyway. You have to provide services which supply only what can be supplied.
I can go through changes in bureaucracy, medication costs, regulations of medical devices through the FDA etc, but there is a lot of blame to be laid at the feet of government for actively increasing the cost of healthcare in the US.
Regulation is actually big. Testing drugs is hard: there's this molecule that settles a methyl group in a magnetic domain on one side, and inhaling the molecule through your nose will cause vasoconstriction and clear up a stuffy head. It's great. If you lean that methyl group a bit in the other direction, it sticks in a magnetic domain 40 degrees away; inhale that up your nose and you might die, or feel frigging awesome, but in any case you're not sleeping for the next two days. That's Methamphetamine.
Well, they're both methamphetamine. One's levorotary, and the other's dextrorotary. The bonds are all between the same atoms.
There's another drug that's an excellent anti-convuslant and will stop an epileptic seizure; but one chirality will kill you at extremely low doses. All chiralities of the same molecule share the same physical characteristics (although a racemic mixture can itself have different characteristics), so you can't distill these or otherwise separate them; you have to react them into stereoisomers (the resulting compounds are structurally-different), separate those, then react the stereoisomer formed from the desired chirality back to the original molecule, and you get the drug you want.
The only way to find out whether a drug is actually safe and effective is to do all of the testing that's done on every drug, starting with the type of testing most-likely to reveal a drug is unsuitable. It costs $300 million, and drugs frequently fail in clinical trials due to non-fatal but highly-annoying side-effects that patients simply will not tolerate.
We kind of have all these expensive regulations because it helps us avoid that thing where you were taking a harmless drug for 10 years and now we know that you're going to decay into dementia by 45 and your life is totally-ruined. I'm up to do continuous review and optimization of all of our regulations; however, any slimming down does take a risk with human lives (if we're wrong about X being equivalent to A B C D and E, we've lost the benefit of doing all of those).
Allow shift nurses to dictate who gets treatment and when just like they used to do it. That includes for example letting shift nurses send people AWAY from the emergency room if they deem them to not have an emergency and instead send them to urgent care or whatever. I've seen people abuse the emergency room to get flu shots. That's unacceptable.
I'll have this discussion with some medical professionals. That's an interesting consideration.
Again, I'm very interested in fixing the American healthcare system. But IF it is to be fixed, it will be fixed by people that ACTUALLY want to fix it and that ACTUALLY have the patience to examine the issue seriously.
Fixing the system and getting healthcare to everyone are two separate problems. The main problem in getting healthcare to everyone is the managemen