Biden: Anyone Can Code

[ClaytonB]

ETA... the other point to be made is, I'll also be one of the first people to know true AI exists - because finance is one of the first places it'll start to get used.

You sound a bit like my 5-years-ago self. I encourage you to look into the SOTA in ML a little deeper, especially in finance. I happen to know that there are HFT firms based in Chicago that are using dedicated, hardware-accelerated ML algorithms (of unknown design, this is strictly trade-secret stuff, I'm sure the CIA itself is jealous of the INFOSEC this stuff is wrapped under) for the purposes of trend-trading. They are hiring a lot of FPGA developers and this is how I know that this is going on and that they're making a lot of money doing it (FPGA developers aren't cheap and hiring a small army of them to work in a finance company so you can keep the IP 100% in-house trade-secrets tells you that somebody has hit a gold-vein). I'm not disagreeing with anything in your post, just consider this one of those insider tips you've been waiting for...

[Swordsmyth]

You sound a bit like my 5-years-ago self. I encourage you to look into the SOTA in ML a little deeper, especially in finance. I happen to know that there are HFT firms based in Chicago that are using dedicated, hardware-accelerated ML algorithms (of unknown design, this is strictly trade-secret stuff, I'm sure the CIA itself is jealous of the INFOSEC this stuff is wrapped under) for the purposes of trend-trading. They are hiring a lot of FPGA developers and this is how I know that this is going on and that they're making a lot of money doing it (FPGA developers aren't cheap and hiring a small army of them to work in a finance company so you can keep the IP 100% in-house trade-secrets tells you that somebody has hit a gold-vein). I'm not disagreeing with anything in your post, just consider this one of those insider tips you've been waiting for...

From what I've heard the computers don't do any better than the humans.

[ThePaleoLibertarian]

Oof. That is just terrible optics. Biden is so out of touch; the "learn to code" meme was such a huge deal and he has no idea.

[Aratus]

Anyone who is an idiot can run for president.

2016 proved that, especially in terms of WHO emerged that summer as the party nominees on both sides.

[fisharmor]

You sound a bit like my 5-years-ago self. I encourage you to look into the SOTA in ML a little deeper, especially in finance. I happen to know that there are HFT firms based in Chicago that are using dedicated, hardware-accelerated ML algorithms (of unknown design, this is strictly trade-secret stuff, I'm sure the CIA itself is jealous of the INFOSEC this stuff is wrapped under) for the purposes of trend-trading. They are hiring a lot of FPGA developers and this is how I know that this is going on and that they're making a lot of money doing it (FPGA developers aren't cheap and hiring a small army of them to work in a finance company so you can keep the IP 100% in-house trade-secrets tells you that somebody has hit a gold-vein). I'm not disagreeing with anything in your post, just consider this one of those insider tips you've been waiting for...

Oh I'm fully aware that it's been decades since institutional traders were tunneling through mountains so they could run point-to-point cable as straight as possible so the speed of the signal traveling on it could be ruled out as a potential cause for missed trades. I'm aware of how it is companies like Robin Hood can now offer you for free a service that used to cost $25. There is serious chicanery going on under the hood there.

None of it is Artificial Intelligence as envisioned by the average non-coder.

I understand what you are getting at. I understand that AI within computer science circles doesn't mean what everyone thinks it means. What I repeatedly try to address is the fact that everyone thinks AI is Skynet, HAL-9000, Commander Data. That is what everyone thinks is coming to tik der jerbs. And that idea is pure fantasy.

We're in almost total agreement on the topic. It's nothing to fear. I just go on to specify the reason there's nothing to fear is because there's nothing resembling what people fear, and frankly, there never will be. Not as long as you keep hearing stories about hiring "a lot of developers" (i.e., people) to work on a "gold-vein". Veins are not motherlodes.

[ClaytonB]

From what I've heard the computers don't do any better than the humans.

Not in the usual sense of stock-trading, no. But mathematically, the stock market graph looks a lot like a chaotic physical system, for example, grains of seed bouncing through a threshing-machine. If you "squint", it basically looks random. But it isn't completely random. If you look at the graphs very, very closely, they are not completely unpredictable, just as the path of a grain of seed bouncing through a threshing-machine is not completely unpredictable if you have a very high resolution, high frame-rate film of its motion. What these ML-based finance houses are doing is applying machine learning (Big Data) methods to the trajectories of the stock prices so that any non-randomness in their motions can be extracted and exploited. While humans have been trying to do this for decades, the fact is, humans are just too slow and inefficient. This is machine work. Once you have two or more HFT firms doing this competitively, what starts to happen is that they are collectively transforming the stock price graphs into actually random trajectories. Each firm is competing to have a slightly better ML algorithm that can notice patterns missed by the others. And so on. This is why they're so secretive. Their in-house, trade-secret formula for extracting non-randomness from the stock price movements is literally a money-printing machine and it only works if nobody (outside the firm) knows its formula. If another firm comes to know the formula, that other firm can just incorporate it into their own ML trading algorithm and moot the other firm's moves in the market, forcing them to back to the drawing-board.

[ClaytonB]

None of it is Artificial Intelligence as envisioned by the average non-coder.

I understand what you are getting at. I understand that AI within computer science circles doesn't mean what everyone thinks it means. What I repeatedly try to address is the fact that everyone thinks AI is Skynet, HAL-9000, Commander Data. That is what everyone thinks is coming to tik der jerbs. And that idea is pure fantasy.

Agreed. The issue I have is that, in addition to false understanding of the generality of AI, most people lack an understanding of just how powerful narrow AI really is. This understanding matters in areas such as digital privacy. Part of the reason nobody cares about their digital privacy very much is that they imagine that they have some intuitive understanding of how they sort of just "blend in to the crowd" and so it really doesn't matter if the big companies collect all this data, there are just too many people. This is absolutely false. Not only are the ML algorithms being run by the big advertisers (Google, doubleclick, etc.) able to track large data-vectors containing bioinformatics regarding your present state (location, activity, etc.) in real-time, they have moved past that to applying predictive analytics to these data-vectors. In short, they know where you're going to be and what you're going to be doing before you do!! None of that has anything to do with Skynet or HAL-9000, it has to do with GPU's performing countless petaflops of matrix multiplications on data-points siphoned (freely) from your real-time tracking device and from the same devices owned and operated by billions of people around the globe. They don't need HAL-9000 and I don't even think they want it... what they want is more like Kitt from Knight Rider, but evil (Karr but not fully autonomous, I guess?) They = the globalists, in case it isn't obvious.

We're in almost total agreement on the topic. It's nothing to fear. I just go on to specify the reason there's nothing to fear is because there's nothing resembling what people fear, and frankly, there never will be. Not as long as you keep hearing stories about hiring "a lot of developers" (i.e., people) to work on a "gold-vein". Veins are not motherlodes.

True. But I think caution is required. We are entering a new phase of "force multiplication" where the old Moore's law is less important than how it is speeding itself up. I mean this metaphorically... the physical limits are becoming less important as the algorithmic designs become more and more efficient. At some point, we will not even need a "physics roadmap for computation" (such as Moore's law) because we will be able to just throw brute-force ML simulation of material properties at the problem until the machines spit out the right material/process needed to move down to the next smaller, more power-efficient process. I see this as a virtuous circle, not a vicious circle, but I still worry that many people in the wider public are not comprehending the true ramifications of this to their day-to-day lives. We shouldn't be so comfortable with all this data just being siphoned away without consent for unknown and unspecified uses. Digital data lasts basically forever. I don't think society has actually adjusted to this new reality and that has me worried.

</soapbox>

[shakey1]

Anyone who is an idiot can run for president.

Been proven time & again. +rep

[Brian4Liberty]

Not in the usual sense of stock-trading, no. But mathematically, the stock market graph looks a lot like a chaotic physical system, for example, grains of seed bouncing through a threshing-machine. If you "squint", it basically looks random. But it isn't completely random. If you look at the graphs very, very closely, they are not completely unpredictable, just as the path of a grain of seed bouncing through a threshing-machine is not completely unpredictable if you have a very high resolution, high frame-rate film of its motion. What these ML-based finance houses are doing is applying machine learning (Big Data) methods to the trajectories of the stock prices so that any non-randomness in their motions can be extracted and exploited. While humans have been trying to do this for decades, the fact is, humans are just too slow and inefficient. This is machine work. Once you have two or more HFT firms doing this competitively, what starts to happen is that they are collectively transforming the stock price graphs into actually random trajectories. Each firm is competing to have a slightly better ML algorithm that can notice patterns missed by the others. And so on. This is why they're so secretive. Their in-house, trade-secret formula for extracting non-randomness from the stock price movements is literally a money-printing machine and it only works if nobody (outside the firm) knows its formula. If another firm comes to know the formula, that other firm can just incorporate it into their own ML trading algorithm and moot the other firm's moves in the market, forcing them to back to the drawing-board.

I use psychohistory...

[Swordsmyth]

Not in the usual sense of stock-trading, no. But mathematically, the stock market graph looks a lot like a chaotic physical system, for example, grains of seed bouncing through a threshing-machine. If you "squint", it basically looks random. But it isn't completely random. If you look at the graphs very, very closely, they are not completely unpredictable, just as the path of a grain of seed bouncing through a threshing-machine is not completely unpredictable if you have a very high resolution, high frame-rate film of its motion. What these ML-based finance houses are doing is applying machine learning (Big Data) methods to the trajectories of the stock prices so that any non-randomness in their motions can be extracted and exploited. While humans have been trying to do this for decades, the fact is, humans are just too slow and inefficient. This is machine work. Once you have two or more HFT firms doing this competitively, what starts to happen is that they are collectively transforming the stock price graphs into actually random trajectories. Each firm is competing to have a slightly better ML algorithm that can notice patterns missed by the others. And so on. This is why they're so secretive. Their in-house, trade-secret formula for extracting non-randomness from the stock price movements is literally a money-printing machine and it only works if nobody (outside the firm) knows its formula. If another firm comes to know the formula, that other firm can just incorporate it into their own ML trading algorithm and moot the other firm's moves in the market, forcing them to back to the drawing-board.

What I meant was that from what I hear they make less profit than human traders and cost more.

[Swordsmyth]

Agreed. The issue I have is that, in addition to false understanding of the generality of AI, most people lack an understanding of just how powerful narrow AI really is. This understanding matters in areas such as digital privacy. Part of the reason nobody cares about their digital privacy very much is that they imagine that they have some intuitive understanding of how they sort of just "blend in to the crowd" and so it really doesn't matter if the big companies collect all this data, there are just too many people. This is absolutely false. Not only are the ML algorithms being run by the big advertisers (Google, doubleclick, etc.) able to track large data-vectors containing bioinformatics regarding your present state (location, activity, etc.) in real-time, they have moved past that to applying predictive analytics to these data-vectors. In short, they know where you're going to be and what you're going to be doing before you do!! None of that has anything to do with Skynet or HAL-9000, it has to do with GPU's performing countless petaflops of matrix multiplications on data-points siphoned (freely) from your real-time tracking device and from the same devices owned and operated by billions of people around the globe. They don't need HAL-9000 and I don't even think they want it... what they want is more like Kitt from Knight Rider, but evil (Karr but not fully autonomous, I guess?) They = the globalists, in case it isn't obvious.



True. But I think caution is required. We are entering a new phase of "force multiplication" where the old Moore's law is less important than how it is speeding itself up. I mean this metaphorically... the physical limits are becoming less important as the algorithmic designs become more and more efficient. At some point, we will not even need a "physics roadmap for computation" (such as Moore's law) because we will be able to just throw brute-force ML simulation of material properties at the problem until the machines spit out the right material/process needed to move down to the next smaller, more power-efficient process. I see this as a virtuous circle, not a vicious circle, but I still worry that many people in the wider public are not comprehending the true ramifications of this to their day-to-day lives. We shouldn't be so comfortable with all this data just being siphoned away without consent for unknown and unspecified uses. Digital data lasts basically forever. I don't think society has actually adjusted to this new reality and that has me worried.

</soapbox>

Collecting and selling your data is stalking and needs to be treated as such legally.


What's scary is that there are probably people doing it secretly that would be nearly impossible to detect or prosecute.

[ClaytonB]

What I meant was that from what I hear they make less profit than human traders and cost more.

That's tough to quantify. I think of it more like "micro-profits"... what they're doing is the same concept as HFT, but it's adding a layer of prediction on top. Instead of waiting for you to place a stock buy or sell order and then beating you to the punch (old-school HFT), the idea is to predict when you're about to place an order and, once again, beat you to the punch. But the applicability of this predictive form of HFT is far broader and more flexible, so you can "steer" it to target specific sectors, specific types of activity, and so on. It can do everything that old-school HFT can do, and much more besides. But the profit in each individual transaction may be pennies. The idea is that you're performing such an enormous volume of these transactions every day that the profits stack up through sheer volume. Boring but ingenious in its own way.

[ClaytonB]

Collecting and selling your data is stalking and needs to be treated as such legally.

Agreed. That's precisely what it is.

What's scary is that there are probably people doing it secretly that would be nearly impossible to detect or prosecute.

There is an entire ecosystem of criminal actors who are "free-riding" in the anarchic grey-zone created by the Wild West legal mentality of DC regarding its own spying. When the nation-state is running roughshod over its own legal system, you can be sure there will be an enormous contingent of shadowy mercenaries right in their wake, using exactly the same tools and methods, just on a smaller scale.

[Swordsmyth]

That's tough to quantify. I think of it more like "micro-profits"... what they're doing is the same concept as HFT, but it's adding a layer of prediction on top. Instead of waiting for you to place a stock buy or sell order and then beating you to the punch (old-school HFT), the idea is to predict when you're about to place an order and, once again, beat you to the punch. But the applicability of this predictive form of HFT is far broader and more flexible, so you can "steer" it to target specific sectors, specific types of activity, and so on. It can do everything that old-school HFT can do, and much more besides. But the profit in each individual transaction may be pennies. The idea is that you're performing such an enormous volume of these transactions every day that the profits stack up through sheer volume. Boring but ingenious in its own way.

But if they lose too often all the profits vanish or at least are vastly reduced.

[ClaytonB]

But if they lose too often all the profits vanish or at least are vastly reduced.

That's what drove the boom in HFT back in the day... it's a "no-lose" formula since you are using physically faster electronic communication channels to beat other stock orders to the punch. By moving the stock up a penny (or whatever) beforehand and then selling immediately to the next order in line to be processed (which they already knew was going to be placed), the HFT firm makes "guaranteed profit" on every transaction. Lots of small, but guaranteed, profits can be worth more than larger, riskier profits. Machine-learning based HFT extends this methodology into predictive space. Since the machine-learning algorithm can be tuned for aggression, they can afford to take some losses, just as long as it makes more profits than it does losses. They just train the model to hit the profit-to-loss ratio they want to achieve and it will have that level of aggressiveness "baked-in". Since training ML models is (relatively) cheap, I guess that they just keep training new models on a periodic basis (monthly, weekly, maybe even daily). This allows them to just keep dialing in the level of aggression they want to maintain.