Sunday, April 22, 2018

Q&A - 22/4

Comment

If we give everyone free college education, that solves a big social problem. It also locks in a price which is ten times too high for no reason. This isn’t fair to the government, which has to pay ten times more than it should. It’s not fair to the poor people, who have to face the stigma of accepting handouts for something they could easily have afforded themselves if it was at its proper price. And it’s not fair to future generations if colleges take this opportunity to increase the cost by twenty times, and then our children have to subsidize that.

You are not locking in anything

I hear people making this mistake on universal healthcare too; Let's say US currently is paying X%s of its GDP to healthcare. Then people ask "but if gov pays that aren't we shifting the burden from one column  to another? Is that fair?". That's not what's happening. With universal healthcare the structure of the system will change; the government as a insurer becomes a major buyer of drugs, a payer to doctors. It can negotiate prices down to negligible amounts. It can take steps to remove "regulatory complexity" (see article) so doctors do not get sued unnecessarily. So that X% percent can go down to Y% which the public will then pay..

Same with college; gov will give a certain amount of money to schools, enough for educational needs, salaries, and the schools will make do with that amount. That could mean schools not being able to invite Milo Yiapowhateverfuckolous, or do other "interesting college culture life events", well fine. Is that a big loss? The previous generation had none of these things, and they turned out fine. Today, we can have cheap computers + Internet! In my school (not a cheap school btw) we had not-too-famous but locally known comedians perform, I bet there was more socially relevant commentary there than whatever some political firebrand might have said. 

(One comedian was this huge guy, he starts with  "hey man, when you first saw me on stage did you think 'look at that big mothe--k--er?.. Or did you think 'look at that fat mothe--k--er?... [pauses a little] or did you think 'look at that big fat mothe--k---er? [laughter]".  So we learned about weight issues in America. Just kidding. But there was some astute observations).

Question

If AI can be useful for businesses can this service be offered as consultancy? 

Yes

In fact Google is fast developing a consulting arm to do just that. These guys will go in with bread-and-butter (for them obviously) new deep neural net methods, and offer solutions for business problems, code them, aaaand also provide the final platform to run those programs  (Google Cloud). In fact consultancy is probably the window dressing for the GC part. 

We hear the emphasis is using most recent methods, deep NN, deep RL, for almost everything [geek] I heard one engineer talk about using reinforcement learning even for industrial optimization problems, in place of simplex and other methods [/geek]. 

Comment

Google is crazy on NNs

Yes

One of the reasons their partnership with Boston Dynamics fell apart was probably G wanted BD to make its robots start from clean slate and "learn" everything they need from data. BD does more straight-forward science / engineering - they'd model all necessary movements one by one, trial and error... So this comes back to functions / reverse-engineering them from data using NN.

Both approaches are fine, but Google wants to rock'n roll, and pursue that other road. 

Saturday, April 21, 2018

Q&A - 21/4

Jordan Peterson

The idea of hierarchy has nothing to do with socio-cultural construction [then he goes on to say it is embedded in us, part of our evolution]. 

Untrue

If that were true, hunter-gatherers, who are our ancestors and themselves the result of billions of years of evolution, would be hierarchical. But foragers were completely the opposite, they were egalitarian. I can quote bunch of books here to prove this point but a Wikipedia link should suffice: "The egalitarianism typical of human hunters and gatherers is never total, but is striking when viewed in an evolutionary context. One of humanity's two closest primate relatives, chimpanzees, are anything but egalitarian, forming themselves into hierarchies that are often dominated by an alpha male. ". 

We are not chimpanzees. Not lobsters either (Peterson also alludes to that in the video above). 

Peterson does have some insight on other subjects, how ppl need to take responsibility, how success on job depends not directly on gender, how identity politics is not optimal for a person.. There is some thoughtful analysis there and the man is not a total jackass. But he doesn't have all the pieces. I'd pay attention to him on his area of expertise - Peterson is a clinical psychologist, so here he has lots to say. History, anthropology, not as much.

Question

Some conservatives like him

Yes

JP is being held-up as an intellectual conservative, but the main problem here is what a lot of people think as conservatism is actually village / farm centered goatfucking. Hierarchy, patriarchy, all kinds of exploitation started during farming era. This little fact of course is uncomfortable for a lot of these "conservatives", so they'd rather look for the root of the issue somewhere else.

From the other direction, someone like David Graeber (from the left), who did great work on the history of money and debt, can chose to ignore the division between these two important ages of humanity. I believe he subconsciously wants to blame hierarchy onto "exploitative capitalism", exclusively. He wrote a big article about the non-existence of the said division, I got a wind of it, the funny story here is I  asked Ian Morris what he thought about the article, who kindly responded, said Graeber was cherrypicking data. Then I told Graeber to check his shit, and he lost it and flamed me. True story.

Anyway. Also, there is a huge industrial, second wave aspect to hierarchy as well a la Toffler.  None of that seems to be making it into the conversation. It's all about Russia, lobsters.


Democratic Party sues Russia, Trump campaign and WikiLeaks alleging 2016 campaign conspiracy

Ha Ha

When will they get to stage 5 of grief?.. The Russia angle makes tactical sense of course - Trump wanted to partner with Russia, for geopolitical reasons, completely logical, then Dems lose the election, they turned that around and say of course he wants to partner because was bought by Russians. That way you also cover up the fact that you lost, badly,  Democrats tried to be Republicans for too long, ideologically they became useless, defunct. The Russia segue covers all that up, unifies the troops, provides a common enemy. 


WikiLeaks will call for a global blockade of Coinbase next week as an unfit member of the crypto community. Coinbase, a large Californian Bitcoin processor, responding to a concealed influence, has blocked the entirely harmless @WikiLeaksShop in a decision approved by management.

Coinbase, Not Cool

Question

When did you appreciate the importance of AI, machine learning, especially of the new tools, deep learning? 

Weather Prediction

I dabbled in many areas of math, especially as it pertains to concrete applications - vision, speech, finance, signal processing.. At one point I looked into weather prediction, but this was after being on a tear of non-stop criss-crossing a wide landscape. So when I came across WP, there is some wonderful math there, maybe I ran out of steam at that point, realized there was such a substantial amount of know-how u'd have to grok to do anything useful especially around numerical computation, I realized in the amount of time I set aside, there would not be enough hours in the day to do it. 

This is when I started to think using AI tools for predicting weather. If AI can reverse engineer functions, and weather is controlled by a deterministic function (literally, there is no element of chance there, just highly nonlinear equations), then new AI tools could learn it from data. This was the time when I started to appreciate the subject much more. 

BTW, weather is chaotic (not random, and climate is neither), an article came out recently that talks about NNs which can learn chaos much better than anything known which could be applied to NWP. Fascinating stuff.

Question

Is feminism a good thing? 

Depends

Empowerment needs to be compatible for today's mode of production. Patriarchy is rooted in the bossy, landowner, farmer male, than becoming a bossy landowner female is just another set of wrongs. Goatfucker HQ (with the longest agrarian history) has the proper vocabulary for it. The landowner is ağa (the boss), there is a female version of that, called hanımağa (she-boss). A lot of times "empowerment" in these lands turn into women starting to act like a hanımağa. Or as the daugter of the ağa (out of reach, asexual, be married off to an important man). Some women try to adapt such airs when they think they are being "empowered". I see parallels sometimes in West. Beware.

Question

What is the ideal company structure?

Gore Associates

[Link] Management is necessary to goad insufficiently competent (or let's say misplaced) people. Or even sometimes manage an oddball client. So management, process is there to manage an overabundence of suck. Pick any management method, for software there is waterfall, agile, this is the case. Here is an excerpt on agile methods: 

"Interestingly, this is also exactly how non-technical companies (like, say, Chrysler) handled software development. Except they didn't hire the engineers. Instead, they contracted with software consultants, and they'd hand the consultants 2-year project specs, and demanded the consultants finish everything on time plus all the crap the customer threw in and/or changed after signing the contract. And then it'd all fall apart and the contractors wouldn't get paid, and everyone was really miffed.

So some of the consultants began to think: "Hey, if these companies insist on acting like infants, then we should treat them like infants!" And so they did. When a company said "we want features A through Z", the consultants would get these big index cards and write "A" on the first one, "B" on the second one, etc., along with time estimates, and then post them on their wall. Then when the customer wanted to add something, the consultant could point at the wall and say: 'OK, boy. Which one of these cards do you want to replace, BOY?' "

I'd say hire smart people, reduce bureaucracy, hell even HR, which these days, constantly try to get you to "do stuff", the rest will follow. Things need to be as natural, "naturally flowing" as possible. Do not try to recreate an entertaining "village" with all kinds of totems, where ppl pray to cat and dogs and shit, do a little human sacrifice, "socialize". That stage was only a brief time in our evolution. We have been the other thing before that, for much, much longer.

Friday, April 6, 2018

Yo Bitch

Link

Twitter’s trust and safety group, responsible for safeguarding users, was run by Del Harvey, Twitter employee No. 25. [..] Her lack of traditional technical and policy experience made her a polarizing figure within the organization, although allies have found her passion about safety issues inspiring. In the early days, “she personally responded to individual [affected] users–Del worked tirelessly,” says Macgillivray. “[She] took on some of the most complex issues that Twitter faced. We didn’t get everything right, but Del’s leadership was very often a factor when we did.”

Harvey’s view, championed by Macgillivray and other executives, was that bad speech could ultimately be defeated with more speech, a belief that echoed Supreme Court Justice Louis Brandeis’s 1927 landmark First Amendment decision that this remedy is always preferable to “enforced silence.” Harvey occasionally used as an example the phrase “Yo bitch,” which bad actors intend as invective, but others perceive as a sassy hello. Who was Twitter to decide? The marketplace of ideas would figure it out [..].

“I often hear people in Silicon Valley talking about fake news and disinformation as problems we can engineer our way out of,” says Brendan Nyhan, codirector of Bright Line Watch, a group that monitors threats to democratic processes. “That’s wrong. People are looking for a solution that doesn’t exist.”

Wednesday, April 4, 2018

Q&A - 4/4

Question

If data science / AI / deep learning is about reverse-engineering functions, can that help companies?

Yes

Functions abound in business. Let's take the issue of churn. "Certain amount of my customers leave, stop engaging, why?" There is a function there. Input is the activity, demographics of the customers, output is whether they will churn. The dynamics of the company, product is such that there is a process, a computation that pushes people out the door. "Every customer older than 50 who are from region X, after doing such and such leave with Y probability". We do not know this computation. How do we discover it? You could employ social scientists with math modeling skills to look at the data, come up with the model, test it, tweak it, to find the function that computes when a customer leaves,  once that is understood steps can be taken to contain the damage.

Or we can simply feed the data (past customers who have churned) into DL and watch it approximate that model - then for any customer we could ask "the function" if customer will churn, if yes, maybe a targeted campaign can be organized, users can be enticed to stay more with the site, product, etc. 

A great DL model on churn BTW is here. The discussion on the vagueness of churn was great.

News

YouTube was attacked by a video publisher woman

Odd

The woman was a nut job. Yes - gun control.

There is something odd going on with YouTube lately though, maybe this is a bad time to bring it up, but there is a feeling out there there is some kind of campaign to restrict certain video publishers. I shared a video of this person before (a finance guy talking abt the petrodollar) he complains all of a sudden he was marked in a negative way.  Jimmy Dore complains all the time somehow his subscriptions are being knocked out, they "magically" disappear (the viewers say they subscribe then find out they are not), his ad revenue took a plunge. YouTube might want to check if there is a "splinter cell" of deep state lodged in there, fighting against non-establishment views.

Bernie Sanders

Israel overreacted ... Gaza is a disaster right now

So true

46. That's all I am saying.

Monday, April 2, 2018

Learning to Drive

Great article on teaching machines how to drive. With the advent of deep learning, the first idea on teaching machines how to drive using pure vision was supplying it data on controls (steering, break, acceleration) and  images of the outside world, in pairs. Feed it gazillions of such data, the program will learn the relation between them and learn how to drive. The article makes the case that might not be the best way to learn. Image X is fed in, car (in the training data) makes right turn. A similar image Y is fed in, car makes a left turn. What should the machine learn in this case?

The root of the problem has to do with the idea of a function. Functions are computation - we use them all the time, to decide, to compute anything and everything. 2+2 = 4. I see something, it has wings, it flies, it's a bird.

But as scientists, modelers, we don't see all of these functions as nature keeps them hidden, so we cannot encode them easily on a machine. But DL gave us a way to approximate these functions, reverse-engineer them from data. The problem is, with the driving example mentioned, there might not be a direct function between controls and outside images. If the root of science is math, and the root of math is number theory / sets, a function is a mapping from one set to another, a one-to-one or many-to-one mapping. If image X results in left steering, similar image Y in right steering, there is no function there. 2 + 2 cannot give you 4 and 14 at the same time (3=1 being 4 is okay though). DL cannot approximate such data coming from such process, that is not a function.

I like the solution employed by the researchers - they learn distances to other cars from images, something less prone to confusing signals.

And here is your moment of Zen...

Sunday, April 1, 2018

#petrodollar

 
Good info.. especially in the first-half of the video. A little too alarmist about wars, both sides are careful not to fight eachother directly, also recently 200 Russian soldiers (mercenaries) tried to attack a US base they got asses kicked good - afterwards, nothing happened. But anyway; some heinous things were done for the petrodollar, I agree 100%.

Q&A - 22/4

Comment If we give everyone free college education, that solves a big social problem. It also locks in a price which is ten times too hig...