mlthoughts2018

> “The only good use cases that come to mind is if there is some particular reason for you to evade the conventional and easier systems of communication and storage.”

This is the number one model for valuation of Bitcoin-like cryptocurrencies. Say you live in Venezuela and dealing with custodial intermediaries in your local fiat currency is completely untenable. You can’t store money in gold or other precious metals. You can’t keep it in a bank account. You can’t transfer it (with fees) in a foreign exchange market.

All these things require dependency on the local fiat currency (plus more for custodial fees if dealing with anything like gold or stocks or bonds), not even factoring in issues of the fiat currency’s own extreme volatility and transaction fees. Banks being trusted central parties or Visa being able to process 40k transactions per second sure won’t mean shit to you at that point.

But with Bitcoin you can drive across the border and actually buy milk or medicine, no fiat currency involved. You’ll happily see the volatility and slow transaction process / high fees as a worthwhile overhead cost for independent purchasing power outside of your local fiat currency.

Now, of course this is a rare and small use case by volume and by market cap in comparison to normal functioning fiat currencies. At any given time only a small amount of the world demand for financial transactions would fall in some destabilized government zone where independence from local fiat currency is materially important and having a preemptive stock of cryptocurrency holdings really matters. And somewhat lesser there will be some small market for people with extreme risk aversion preferences or “prepper” outlook (eg, “What if the US dollar collapses?” - most of us don’t care to hedge that extreme tail risk, but some people do care).

The total demand for currency holdings like this will obviously be much lower than any total market cap of a major fiat currency.

But it may still be higher than the total market cap of cryptocurrency today - that completely depends on your personal take on forecasting and speculating.

In other words, there are several straightforward and uncontroversial reasons why buying and holding cryptocurrency rationally makes sense. You could happen to be wrong about the speculative future demand for this type of transaction capability. You could be wrong in the magnitude of the forecast. But being wrong doesn’t make it irrational hype or conceptually broken from first principles, as you foolishly assert. There are many very uncontroversial, directly obvious and rational reasons to speculatively buy and hold crypto, depending on your personal appraisal of that market and future demand for that transaction capability.

I am saddened to see how much effort you went to in order to copy / paste this example everywhere in this thread. It isn’t relevant and has no connection to Scott’s efforts to stay anonymous nor the NYT’s malicious behavior by not respecting his request.

You’ve pasted this everywhere for nothing. It doesn’t contribute to the discussion or count as evidence for anything. It’s just a completely disconnected and irrelevant piece of trivia that you are trying to astroturf onto the discussion as though it had any credible reason to be considered relevant (it does not).

This feels almost like a false flag sort of extreme comment, knowingly violating the rules, just to make it seem like Slate Star Codex supporters are this kind of discourteous, bro-y, insulting aggressive group.

Maybe it’s not actually a false flag, but let’s be clear that it utterly doesn’t represent the kind of values most SSC readers subscribe to. While I don’t like the original comment’s criticism of why people like SSC, that in no way justifies this type of extremely inappropriate response, full stop.

In Eastern Europe everyone interjects assertions that don’t follow from any part of the discussion? Because that is the article’s approach to connecting topics of late capitalism and gender/queer issues and structural inequality. They are disjointedly introduced in an unmotivated, ungrounded way that violates basic, universal conventions of successful writing and communication in all nationalities.

I really doubt what you are saying here. Eastern Europeans are just as capable of logical trains of thought supported by compelling writing and evidence as anyone else in the world - which is very different from this article.

You can take a bus to Colombia and actually buy food or medicine. You call that nothing, so fine, don’t but Bitcoin if that works for you. It’s surprising you don’t think others might possibly disagree.

This is why you would speculatively buy Bitcoin and hold it long before that happened, at a time when people are happy to accept your fiat currency for Bitcoin. People learning from experience in Venezuela, Zimbabwe, North Korea, would probably feel glad to put some fraction of savings into crypto just in case, knowing that the interim volatility risk of it sitting in crypto is a worthwhile price to pay to have a reserve money supply that does not need to pass back through their original fiat currency on the way to being spent (as precious metals, stocks, bonds, securities, options, etc. all would).

Gold would specifically not be a good choice in that situation, because you have to pay a custodian to deal with the gold and store it for you. That means either transacting in the original risky fiat currency any time you convert between the currency and gold (such as to buy things), or else transacting in the original risky fiat currency to get a foreign exchange to another currency and then transacting with a custodian - which defeats the whole purpose of converting the asset away from the original fiat currency that exposes untenable risk.

If you live in a situation where the only local fiat currency is even worse, then the “volatility fee” of cryptocurrency may be well worth the money to decouple yourself from that bad fiat currency risk. Maybe you can purchase things directly with the cryptocurrency, or maybe you just wait until you can drive across the border to a place with more stable fiat currency, sell cryptocurrency for that currency, and buy milk. Simple currency exchange or traditional custodial stores of value like precious metals wouldn’t work, since you’re still beholden to the original fiat currency problem.

At any given time, some small slice of world currency demand will be focused on this “protection from unstable fiat currency” and some other small slice would be like a prepper mindset hedging their perceived risk of facing that (eg “but what if the US dollar collapses?” from people with very extreme risk aversion preferences).

That total “market cap” demand for cryptocurrency will surely be way, way lower than traditional fiat currencies, but it may still be much higher than the current total market cap for crypto, which means there is room for completely rational speculative buying and holding of crypto, depending on your outlook and risk preferences, for perfectly sound reasons.

I highly doubt Bitcoin is ever going to become a major direct medium of exchange. But also it does not need to become that. It can offer big value in a lesser role that would still justify speculative buying of crypto today.

I disagree. I read the full article and enjoyed it a lot. But the attempts to pull in “late capitalism” and gender / queer issues were just poor writing. It’s like mixing all the paint colors and just getting an ugly gray. There was no convincing connection to “structural inequality” in this at all, nor any sense of significant radical opposition to entrenched powers that be. Piling that stuff on just made the essay bloated. It felt more like having an unsupported agenda of radical left issues forced upon me with no serious intellectual basis or connection, with lots of language thrown in to make sure you know that the author feels those radical aspects are 100% correct (without justification) and must be taken as truth and as an assumed shared starting point for the reader. It significantly detracts from the article, lowering both the basic writing efficacy and the intellectual merit, and watering the whole thing down as an assertive (but unsupported) criticism of capitalism instead of the rather interesting discussion of negativity that it used as a bait and switch.

It being by Stephen only lowers the credibility. This is the same guy who wrote up a big slide deck insulting all the cretin, moron programmers in his eyes who aren’t using functional programming, by saying we are in the “bloodletting and leeches” stage of software at this point of history. (And then he took down the slide deck after getting backlash.)

Stephen is bombastic in all the worst ways, while insulting and looking down on other engineers. I’m willing to look past Stephen being the author and give the general story of Bitcoin power consumption a fair shake despite it (and a clear-minded assessment suggests it’s a total non-story), but Stephen being the author absolutely does not help with credibility.

You’re not factoring in the military industrial complex cost of military enforcement of fiat currencies. You have to be factoring in the cost of militaries, fossil fuel exploits, and more.

The cost of operating US dollars as a transactional concept is many, many, many orders of magnitude greater energy cost than Bitcoin.

I find this completely tone deaf, factually unconvincing, ignorant of the true required comparisons to the sum total power consumption of the whole transactional, fossil fuel and military industrial complex systems and infrastructure required to support the existence and usage of first world fiat currency.

“Bitcoin uses a lot of energy” utterly doesn’t mean anything on its own like this.

It’s infuriating to see such an intellectually dishonest and disgraceful, disingenuous line of argument taken with even the smallest modicum of seriousness.

This just needs to be ignored out of existence.

Even if it were an easily changeable characteristic, we should consider it wrong for any society to threaten harm or punishment or ostracization for this. That the characteristic is or isn’t easily changed is not relevant.

It’s not a question of moral relativism. You’d have to argue that any behavior is a crime in order for it to be legitimate for perpetrators to suffer consequences. Otherwise, it is wrong that someone should suffer consequences, all else equal.

I like nachos. Tomorrow I might not like nachos. It is wrong in any society if my liking or disliking of nachos results in me being killed, bullied, shamed, etc. Any society that claims a right of legitimacy to impose social norms in this way on the liking or disliking of nachos is a morally wrong society. There’s no debate about this. You can either argue convincingly that nacho preference is a crime, or else any attempt to treat it like a crime is wrong.

Core banking systems are some of the most horrendous examples of legacy tech debt and duct tape barely keeping a system from falling apart in the whole industry. Pick any large bank, BoA, PNB Paribas, Deutsche, Goldman, JPMorgan, etc. and they all have a widely known reputation for paying poorly, requiring significant extra work hours, offering zero career growth, offering zero opportunity to work on interesting projects, and have some of the most bureaucratic environments with no freedom, no ability to innovate or introduce new ideas, of any industry.

If you want to be underpaid, constantly worrying about sword of Damocles layoffs, pushed to work unpaid overtime all the time, all while your job causes skill atrophy and you become unemployable anywhere else because you only do maintenance on a mind-melting abyss of legacy tech debt ... then work in banking.

No they don’t lead to burnout. Those are just basic attributes of working hard and taking healthy engagement. More like the advice of, “hire people who give a shit.”

Burnout comes from other things added onto a situation. Believe me I’ve seen lots of severe burnout at places that are “maintenance only” no-ambition shops. Burnout is much more about disenfranchisement and a big gap between motivation and requirements. Rarely is burnout a function of raw hours worked (and when it is, it’s easy to solve).

Yeah, agreed. Maintenance, reliability and tech debt should be first class concerns. Product strategy should include them, allocate time for them and give clear guidance of the trade offs that drive prioritization of these lines of work vs prototyping, feature delivery, etc.

Neglecting maintenance is a very bad idea, but setting it up as an enshrined cultural north star of an engineering department is actually even worse.

I disagree very strongly with this even though I strongly support ideas like “choose boring technology” and “reliability is your number one feature.”

The issue is that none of this can be decoupled from politics, and “maintenance first” tells you that the force with political power is some version of a Dilberty group of rent-seeking IT leaders who just want to be paid without having to do much work.

That has pros and cons compared to politics driven by product leaders with unrealistic ideas of shipping new features, but at least they have hustle, ambition, animation and lifeforce. I find the trade offs are worse with anti-ambition “maintenance first” IT leaders.

Setting IT maintenance as a core strategy, rather than something that serves a strategy, is complete death. Employees better hope they like the dead end career they have at such a company, because it’s going to cut them off from any lifeline to relevancy.

The company also better hope they are right in their estimate that a big mass of customers will be satisfied with just maintenance and reliability. If you are in any kind of competitive industry, advanced features or new tech that allows price cuts for your competitor are going to cripple you, and you will be way too slow and dinosaurish to recover or react.

Finally “maintenance mode” places tend to pay poorly. Not because innovation is unfairly paid better, but because they want to offer claims (but usually not realities) of work life balance and other fringe benefits, while just pay less, giving crappy 5% bog standard raises, no bonuses.

When an employee has hustle and ambition and wants to achieve correspondingly more and produce correspondingly more output in exchange for significant growth in compensation, they’ll just be given a line that they can’t be, in principle, because the company’s fundamental way of working is incremental maintenance. A career goal of ambitious projects is conceptually disallowed, along with any goals of related comp growth. So as a result, you can bet you’ll be working with a bunch of extremely mediocre colleagues, a lot of them only taking a 9-5 job that fits other obligations like family stuff, and figuring they’ve given up on ambition or wealth building and you should too.

It reminds me of a famous quote by TS Eliot when he grew sick of university life at Oxford (paraphrasing)

> Oxford is very pretty, but I don’t like to be dead.

Presumably that would be true of any possible intervention and if we overcome that risk in every endeavor reliant on human productivity then it’s not particularly important to call it out as a risk here.

I’m sure it’s more complex than this, but why isn’t the federal government jumping at this as a massive employment opportunity. You could train a wide variety of people for these jobs, and the way the jobs could be structured, support roles might not require college degrees in a lot of cases. Award grants or conditional extra city funding to cities that grow the program and make it work. People get to feel good about giving back, the country can feel good about reducing needless arrests, we can make bigger strides to solving homelessness, we can make good on various social justice reform items not only without necessarily massively defunding police, but also by freeing police up to actually do the much harder job they are trained for and stop treating police like they need to also be an arbitrary mental health issue handler.

There’s a lot of complexity for sure. This idea will hit a lot of roadblocks. But it seems like the potential upside across a dozen or so dimensions is super high.

Of course this is the same country that shuts out widespread nuclear power, so maybe this is the last we’ll hear about it when some police equipment manufacturer starts lobbying against it the minute one of these workers is involved in a violent encounter or significant arrest or something.

“If you don’t want to live there, don’t live there” - the siren song of inhumane exploitation.

This comment appears to take an unwarranted parochial approach to how to taxonomically call something “reinforcement learning.” This is just not relevant to the content of the interview and the discussion of the large-scale recommendation problem as RL.

Agreed, and we software engineers just keep letting them do it to us by agreeing that leetcode algorithm hazing should be the absolute measuring stick by which all software hiring is judged.

If you feel like you barely squeaked your way in the door as someone was whipping you and yelling “dance monkey dance” then you’re helping them identify you as a gullible goober they can pay peanuts (relative to your value add) while overworking you, and get you to thank them for the sweet ping pong and happy hours.

Until young candidates just say no to this shit, then it’s an abyss of open-office, grab-ass culture we have no one to blame for but ourselves.

There are some fundamental limits to commitment-focused culture though. People have life obligations. Even if they wanted to work 60 hours a week, their sick mom or their childcare or their kitchen renovation or an especially busy graduation season for nieces and nephews just requires their time, and their knowledge that their job is secure and they won’t be penalized on raises, rewards, etc., for simply having a life.

Because of this, the commitment-focused path just doesn’t work, flat out. At best you can use it briefly to exploit unwitting young people who don’t know their worth and don’t have enough self-confidence to say no, and ride that into a pump and dump IPO situation, after which your company will promptly become process-focused and all the young people will quit and be replaced by mid-career people who don’t mind inheriting the mind-melting tech debt as long as you pay well and only expect 40 hours per week.

But how do you get anyone up the food chain to care? I’ve never witnessed that anywhere I’ve worked. If you took things like team surveys, code review load, meeting load, etc., and tried to make a case that changes are needed, you’ll get laughed out of the room.

The only metrics that will matter are “did you get the stuff done that we wanted?” (where “stuff we wanted” is defined vaguely and the meaning shifts to suit product leaders ever changing political landscape), “did you get it done really fast?” (where “fast” is vaguely defined according to product political circumstances and never puts any weight on engineering estimates, staffing needs, or resource limits) and “did you get it done cheaply?” (where “cheap” is defined vaguely based on various internal politics and budget turf wars as well as larger company financials - and when the money is good and no one looks too hard at this, it allows sweeping other issues under the rug, no one cares if you burnt a quarter working on the wrong problems).

Assessing effectiveness, in principle, is purely a political concept that operates from the top down.

This has been true in every company I’ve worked for or knew a colleague or friend who worked there - from tiny “lean” startups to extreme cultures like Bridgewater to every mid-sized or large tech company.

fastapi is very similar to what you’d get from flask with an API spec tool like flast-restx or flask-marshmallow + incubation time to build your own components & patterns around it + fastapi is faster.

So if you are committed to the component tools you already built with flask add ons, stick to flask unless you need the speedup. Otherwise if the project is new or you’re otherwise uncommitted to your existing component tools, switch to fastapi.

That’s a lot of semantic hairsplitting. They are both “reinforcement learning” in the same way a Honda Civic and an aircraft carrier are both “vehicles.”

Also to be clear - I think it’s great to hear this perspective and putting together the interview is well done. I’m not trying to criticize you nor the merit of talking about this topic - it is well worth it.

I was just asking, since you develop these kinds of interviews, what do you think would be a good way to get the other side of the story and talk to big tech practitioners who do not agree with the leap to reinforcement learning?

That’s zero reason to reach for a bazooka like reinforcement learning though. You could use simple Thompson Sampling or many other multi-armed bandit methods.

Balancing learning vs serving the optimal result is fine - many companies have approached ranking and recommendation that way for decades - but reinforcement learning would still not be justified unless you present some extremely compelling evidence.

The “This is a reinforcement learning problem” section is very unconvincing. It presupposes that approaching the problem like a game or “capture the flag” is good or somehow better than supervised learning based on attributes that are known to correlate quite strongly to user preferences.

Given the very widespread complaints about this type of recommender system - eg modern YouTube and FB Newsfeed rankings are widely panned as reinforcing biases, optimizing for pure engagement which leads to reinforcing outrage, both by ML experts and general users who don’t like the experience & feel it is manipulative, what is your take on how we can steer the conversation in the other direction - that this is NOT a reinforcement learning problem, and that we shouldn’t reward people looking to pad their ML resume with solutions in search of a problem that facilitate their ability to brag about scale & complexity over a solution that very demonstrably serves users poorly?

On your last point, I think it only applies to things like bitcoin where the resource cost for mining and validating is high. Litecoin and BCH are quite different and result in far lower transaction costs, but they aren’t as popular yet.

There’s every reason to think in the longer term there will be some cryptocurrencies with higher costs that behave more like a custodial store of value like gold, some that behave like a cheap medium of exchange, and some that have other purposes like hedging against fiat currency or operating some smart contract based app / game / service.

The valuation model would be different for each.