mlthoughts2018

I think you are wrong. While the cost of authoring and distributing content are cheaper than ever and anyone can broadcast to the planet trivially, the ability to actually bring attention to your content has never been more scarce. It is deeply controlled and manipulated by huge confirmation-bias-as-service platforms in every recommender system, personalized ad system, and content feed across the mainstream internet.

Your fellow citizens deserve more credit. They have brains. They really can and almost always do form adequate understanding of reasonable takes on new information.

But when their eyeballs are subjected to consumerist bidding wars and ranking algorithms to inflame, to stoke fear and insecurity, to render feelings of inadequacy, and they are so thoroughly manipulated, then you get information monoculture.

We need information diversity and information vitamins. Instead we’re allowing ourselves to be spoonfed information junk food.

Why on earth would number of languages have anything to do with it. That doesn’t make sense. Parochial number of languages doesn’t capture the idea of understanding multiple paradigms, low level / high level, type system considerations.

A person might have zero experience except Node.js yet still show high mastery of functional programming, mixed OOP, design patterns, compilers, etc.

Setting any threshold on languages just sets up a meaningless metric.

Hard disagree. Facebook only has legitimacy if you personally agree it does. The minute you walk away, it loses all legitimacy. Same for LinkedIn, Twitter, etc. etc. There is no sense in which society at large using the product creates any degree of legitimacy for important functions like identity management. Precisely zero.

You are tacitly voting that Facebook should be viewed as an arbiter of identity by taking this point of view.

The comp offers aren’t even that great, even compared to midrange tech companies, especially for senior / staff level engineers.

I worked at a midrange ecommerce and content licensing firm, about 20-25 years old, in a major US city, and my total comp as a senior engineer (6 years of experience then) was in the range of $350k between salary, equity and bonus, and this was over a decade ago.

If you find a company that values you and needs your skills, you can get high comp at lots of places, and many have much better work / life balance, less wanton dysfunction and less political drama in the workplace.

Even a 50% or 100% increase over the $350k isn’t worth debasing myself with “dance monkey dance” whiteboard hazing. You might as well ask Facebook to give you a dunce cap on your first day, or a t-shirt that says, “My lack of self-respect is so severe that I agreed to do Facebook’s interview process.”

The mere fact that I’d actually be valued at the $350k place and wouldn’t be rectally probed to ensure I am fungible with every other assembly line programmer is a big indicator that my total earnings over a long time are likely to be much higher away from a place like Facebook. Generous scheduled equity refreshers are no match for actually rising in responsibility and skill growth to critical new positions in a firm.

Facebook is pure self-parody at this point. This post is tragic comedy with no practical value besides entertainment. Anyone pursuing job options following advice like this has a severe problem with lack of self-esteem and capacity for independent thought.

Even if that was purely true, that doesn’t say anything about relative purchasing power to afford mortgage costs though. Home sizes increasing doesn’t mean home prices, relative to purchasing power and technology changes, have stayed consistent with real wages.

And it’s likely not purely true. Home sizes being larger is also a function of REITs and corporate ownership of more consumer real estate, as well as boomers using soaring retirement fund profits to finance remodels, etc.

It’s way more complex than to draw any conclusion that this means purchasing power of homes has stayed close to wages. It hasn’t.

> “ I read the "morally incompetent" as just short hand for "not being honest about current state of mind".”

That’s a great way to put it. I think “incompetent” is the correct term because aside from not being honest with one’s self on these “do good” issues, people can also just be very ignorant about it.

Being ignorant about the actual, measurable, demonstrable humanitarian side-effect of your beliefs and actions is rightfully seen as moral incompetence and should be harshly criticized.

I like a quote from Robin Hanson about this (paraphrased):

> “I wish people felt more of a social obligation to believe accurately and less of a social entitlement to believe whatever they want.”

The final graph on median home mortgage payments is misleading too. That can be propped up by REITs and increasing institutional ownership of consumer homes.

There’s also the issue of more millennials living with parents and relatives, unable to afford their own homes, which can be hidden in that.

Definitely can’t conclude from it that wages and purchasing power have kept up with mortgage costs.

I found the “Price Changes” graph about 2/3 down the page really alarming.

Virtually all of the gains in the form of lowered prices are purely in discretionary consumption and subsistence goods. Cheap entertainment, toys, staple foods.

Meanwhile the insane price increases are in critical needs, medical care, housing, college education.

In my mind that graph is pretty strong evidence that there really is massive inflation and the idea of “asset inflation” does exist, contrary to the author’s lead point that it should be seen as normal growth of value.

For most people, if you show them this data and said, “see, low inflation, don’t worry...” they are rightfully going to be pissed off. “Low inflation” except in everything that really matters for health, home ownership and betterment for my children.

I guess I should just thank the monetary policy overlords for my cheap Netflix while we all die of cancer under the thumb of landlords and all our economic participation from labor dwindles and continues to lose purchasing power for what we most need.

I’ve always found Locust to be incredibly simple for simulating any user behavior. Very well designed task and user behavior APIs and patterns in it.

I actually think people are fairly wrong about bitcoin and other cryptocurrencies.

If you approach some notion of valuing bitcoin, the clearest store of value reasoning is that it can act as an intermediary between different fiat currencies without being beholden to any source currency.

For example, gold has the advantage over bitcoin of being a physical tether - you can make stuff out of gold, people like pretty gold things, etc. But to deal with gold you typically are tied to some source fiat currency. I don’t just mean at the time of purchase when you buy gold in dollars or yen or whatever, but the entire process. Unless you yourself take physical possession of the gold and interface to the exchange, then gold as a store of value can only be retrieved by selling in the source currency you bought it, that is used by whatever custodian is holding it for you. Maybe they would agree some exchange rate to give you the proceeds of your sale in a different currency, but it would amount to the same transaction as selling it in the custodian’s original currency then paying an exchange fee to switch to another fiat currency. Most people, even large investors, won’t bother. Gold purchased in dollars is basically just a store of dollars with transaction costs to get to anything else.

Cryptocurrency has significant potential to be different. I could buy bitcoin in dollars, but open a French bank account and realize the proceeds (without transaction costs) in Euro if I want.

For most stable world currencies, it’s not a big deal (though hedging away from fiat currency for some small amount of cash could be valuable for some). But for people truly exposed to hardship due to devalued currency, poor regional governments or fiscal policy, decline of undiversified sources of GDP, this ability to store value without going through the friction of a volatile fiat currency is a huge value.

So then valuing bitcoin overall would be a function of estimating what total asset level across the globe represents the market demand for this decoupling from fiat currency.

I don’t see bitcoin ever growing to the monetary totals of stable world currencies, it will never grow to the size of precious metal markets, but does that mean its current total asset level is right? I doubt it. There’s lots of room for it to grow just based on this minor usage as a shield from fiat currency for store of value.

The biggest risk in my eyes is whether governments will development ways to regulate bitcoin such that it loses any of this benefit of decoupling from fiat currency, and it becomes like modern gold markets where it’s entirely fronted by a particular source fiat currency depending on geography where you want to transact.

A big risk of course, but the point is that it’s not trivial to dismiss actual tangible benefits of crypto as a value store over alternatives.

You could compare it to fine art markets and freeport holdings in that sense, just without requirement of custodians and more accessible to average people.

This isn’t true. It just creates artificial scarcity in the form of the union membership, similar to artificial scarcity of the Bar exam, and college degrees in general, but mere union membership has even less claim to indicate skill, competence, etc.

When unions function as an exclusionary fraternity, they are actually pretty horrible. A good indicator of a well-functioning union is: the union doesn’t have negative effects on people who don’t join, and they are free to do their jobs side by side with union members and nobody cares, nobody judges anyone for their personal choice, no one discriminates on pay or opportunities.

It would seem their backing research and metastudy totally missed the boat and ignored well-known academic experts on the matter.

I definitely trust the academic studies over an entertainment podcast.

This is really, really not true. Advertising lift has been well studied, especially in metastudies spanning hundreds of digital campaigns across Facebook and Google. These are independent academic metastudies, with hundreds of millions of impression data samples.

Positive lift in the range of 0-20% is very common, and many statistical aspects of causal inference on ad impacts are well understood.

Negative lift and flat campaigns are real phenomena too, and it does deserve more widespread publicity that negative lift happens in an appreciable number of campaigns, but that doesn’t take away from the overwhelming evidence that digital advertising works and that the mechanics of positive lift are well studied.

Here are two of the foundational papers in this area:

- https://www.kellogg.northwestern.edu/faculty/gordon_b/files/...

- https://www8.gsb.columbia.edu/media/sites/media/files/Garret...

Particularly Figure 1 (page 26) in the second link. That figure alone utterly refutes any nonsense claim that digital advertising doesn’t have provably positive ROI.

That tells you all you need to know about the credibility of Twitter - (also, what’s with the presumption of nationality, that’s not very useful).

I’m not convinced by this paper. I’ve trained a lot of from-scratch word, sentence and query embeddings in my career, it’s probably the single main thing I’ve done. I’ve never observed rescaling the average context vector to have an impact on application performance. It amounts to rescaling gradient terms, but most of those are being backprop’d from layers with batch normalization, strict activation functions, clipping, etc. There are many, many non-linear effects contributing to how that rescaling constant plays a role, and in anything other a completely shallow word2vec model with no further layers and where you just want to extract the embeddings in some application-agnostic way, that normalizing constant is not going to matter.

Why do work when you can milk it for status?

The paper itself says the only change is normalizing by the context window size C.

Moral Mazes is one of the rare books I return to and read multiple times. Far from being out of date, 40 years later it is more relevant and the research lines up more directly with corporate behavior than really any other research on this topic written since then. It really is a tremendous book.

Twitter lacks both document authoring features and professionalism context for anything like an announcement like this. It comes off like an out of touch person trying to stay “fresh” or “with it” shoe-horning a dry press release comment into a medium people use to laugh at cat photos. Even official Twitter accounts and journalists have the sense to link externally to more important matters. It undermines credibility to put those matters directly on Twitter. A single short tweet of that type is bad enough, but a 12 tweet thread is further forcing long discourse into a platform built to avoid it.

It’s so gauche and embarrassing to see a serious executive announcing a principled career change through Twitter. “Tweet 6/12” ... a red flag should go up somewhere. Can you not buy a domain and link a short paragraph?

This isn’t an issue related to Threader or visual reconstruction of tweets. Rather, how did society get to a place like this? It’s bananas.

Pulled straight out of the Moral Mazes playbook, chapter on “dexterity with symbols.”

Yeah, when reading this I was wondering if Rust has sealed traits like Scala, and also structural typing.

The extreme degree of shared syntax and shared semantics with Scala was really surprising to me.

So Rust is Scala but with some extra complicated additional syntax around borrowing references and generic lifetime constraints.

Like many virtuous science applications, curing disease, fighting child abuse, feeding the hungry, unless we start seeing either profitable business models around it or charitably funded research organizations that actually pay competitive salaries to attract and retain talent, this is all just aggrandizing. It’s usually pet projects of certain gurus or career researchers with extreme personalities and penchant for drumming up funding hype.

When the science is real and we’re talking about professional, legitimate deployment of a medically revolutionary technology, it won’t show up as fractious riff-raff biotech startups built on the back of post-docs and super cheap labor hoodwinked by idealism. Rather it will be companies or non-profits that actually pay high compensation to get results.

All of the criticisms of megacorps listed here are just as applicable to startups (especially with any whiff of VC funding) and mid-sized companies, without competitive compensation.

You very likely won’t be happier at other tech companies, because despite promises of career growth, autonomy, greater responsibility and company mission, it will just be the same ruthless corporate shilling, just with worse hours, worse vacation, worse pay, and worthless lottery-like equity.

I urge a lot of caution. The only reason to work at a startup (all other constraints like visa issues, geolocation preferences, etc., being equal), is because you have absolute faith in the core business model.

Choosing to work at a startup because of the technologies you will supposedly use, the seniority of the role you’ll supposedly be given, the fun-seeming optics and kid-like atmosphere, lack of dress code, etc., is a massive, massive mistake - not because those preferences are wrong, but because startups absolutely don’t fulfill them. They just pay lip service to it.

For 90% of employees, the choice is purely between medium-corp and mega-corp, based on your relative appraisal of work/life balance and compensation.

It would be great if this were different and the charismatic nature of startups really did offer offsetting benefits through learning, autonomy, etc. But that is just across the board a total false promise bill of lies in startup marketing to bait & switch tech workers they otherwise can’t afford on the basis of market compensation.

No, the author is pretty clear. They take issue with verbiage they personally hear from younger people. Here is a quote from the article:

> “ The first thing I tell them is to talk less and listen more. I don’t know if this is a generational thing or they’re just nervous to talk to me or what, but most of these conversations begin with me getting machine gunned to death with someone’s line of shit right out of the gates. Bang bang bang bang bang bang bang bang bang bang bang bang bang bang bang bang…bang…bang bang bang bang… I’m thinking to myself, “Oh my god, take a breath, kid.” Maybe I sounded that way in my twenties too – being excited, feeling like everyone wanted to hear every thought I had on every subject under the sun. I don’t remember. So I give them the best advice I can think of for building a practice in financial advice and developing client relationships. First of all, I says, “You don’t even know what the f* you don’t know.””

So it does not seem the author is talking about sales / client relations, at least that isn’t the primary issue they raise and the context of when they advise younger people to be quiet and listen is very different than any sales / client relations context.

After 15 years as an individual contributor and five as an engineering manager, I have to say, you are flat wrong. Doing the work of implementation is a dime a dozen. Getting quality RFCs and prior alignment on solution architectures is at least an order of magnitude more difficult, more valuable and more rare.

Cocksure senior engineers who think, “RFCs are easy, this’ll be quick” are the single biggest source of money-wasting errors that I have to navigate.

I’ve implemented huge distributed applications serving machine learning predictions in global scale ecommerce products, real-time image processing, eventually consistent and SOX compliant billing applications and dozens of other similar solutions in my time as an individual contributor.

Being a manager responsible for not wasting hundreds of person-years on the wrong architecture is way more challenging. It’s not even the same sport.

I think you have it very backwards. Coding leadership usually involves nearly zero coding itself. Rather it involves getting the other person to write down RFCs, tech specs, requirements, design diagrams, etc., and guiding them through how to think about macro structure of solutions and outcomes for specific product deliverables and business cases.

Training involves teaching specific tactics of coding, like a new technique of multi-processing, or an approach to refactor a legacy class implementation. Helping someone learn tactics is teaching or training, which is a different thing than leading.

A young person could lead an older, more experienced person. For example, they could drive them to outline a solution in architecture documents that better align with product management. That can be true even if there are no tactics of software design that the younger person can teach the more experienced person. So that doesn’t reduce credibility or anything in the leadership sense.

In general it’s important to separate these: mentorship, coaching, teaching / training, and leading. They are all very different things.