esc_colon_q

Literally declaring yourself the winner of an election that you solidly lost and then repeating that claim over and over again as your supporters rile themselves up and prepare for violence abso-fucking-lutely qualifies. If the election really was stolen (it wasn't), then violence on a grand scale would be more than justified (it's not), so pushing a false claim to that effect when you have as much power to make it stick is one of the most directly violent acts that can be committed.

"This election was stolen" is way more dangerous than "Fire!" in a crowded theater because if the bullshit sticks then you actually end up with war.

Have you ever heard Kara Swisher go off on this? She's vicious on the topic.

Don't forget GA - Dems won full control, and will now have at least some ability to legislate. So Twitter et al need to be thinking about how to get ahead of the complaints that the left has against section 230 (the left hates it because they are not forced to aggressively moderate, the right hates it because they are allowed to moderate at all). This was not looking likely to be something they had to worry about until the surprise sweep in Georgia, so I'd argue that may have played as big a role as the craziness in DC.

Yeah, so much of Google's culture is already about slowing down work so that it takes 10 (5 eng ICs + lead, analyst, manager, PM and PgM) to do the work that 2 engineers would do with identical quality in any other company, and the old-timers are the absolute worst in terms of keeping that status quo in place. I can't recall seeing a single team there that was properly sized and wouldn't function better with half the team and an eighth of the process.

There's a good argument to be made that a big chunk of the value of an engineer to Google is strategic, simply that they are locked up and aren't working at FB, Amazon, Apple or Microsoft. I was never at a high enough level to have a view into the data that would confirm that, but it certainly felt like even if you weren't particularly productive in the environment everyone up the ladder was perfectly happy to let you malinger on the payroll forever, as long as you weren't so bad that you did damage to someone's pristine art project of a codebase. So maybe inability to fire isn't really such an issue - even now, seeing anyone fired at Google, let alone an old timer, is extraordinarily rare.

If you make these things Big Process, each of which requires spreadsheets and certification by high level manager and weeks of delay, etc, then yes, people will think of them as meaningless hoops. If you can avoid dumb red tape then things can go quite well.

In games, when QA reports into the same person managing the product, things run smoothly and everyone remains aligned towards shipping a good product. When it's a separate org that loans people out and has OKRs that don't connect to individual products, they become horrible trolls who are only out to fuck the product up by submitting so many bug reports that it can never be released. Sim with whole-company legal teams ("never ship anything, we might get sued") and shared security ("never run services or we might get hacked") and ops ("never use servers or they might crash"). Alignment is everything.

I personally just think everything works better when the best people are in charge and rewarded for being the best. Yes, there's tons of luck, and it's not fair in any "everyone deserves a shot" sense. Still a better strategy, because most people are bad at most things. We should take care of them, but we should not pretend they're competent if they're not.

And people from neuro come into AI for short rotations pretty regularly to do astoundingly naive work there which amounts to nothing at all.

Maybe both fields could stop flinging shit and realize that they are both missing a key piece of the puzzle...

There can be and often are, because it's important to account for future expected flows - credit that's been extended and is due to be repaid, for instance, has to be accounted for. The thing is, these flows should be clearly and cleanly separated from actual cash on hand.

> IA will also remain quite essential, because for the foreseeable future, computers will not be able to match humans in their ability to reason abstractly about real-world situations.

I broadly agree with what this article says, but depending how you define "foreseeable future" I find this to be a dangerously naive viewpoint that just assumes nothing will change quickly.

I'm not stupid enough to say abstract reasoning about the real world is a simple problem or right around the corner, but there's no evidence so far to indicate it's much further off than, say, object recognition was when Minsky (or more likely Papert, apparently?) assigned it as an undergrad project. We pour exponentially more money into research each year, and have more and better hardware to run it on. We're going to hit the ceiling soon re: power consumption, sure, but some libraries are starting to take spiking hardware seriously which will open things up a few orders of magnitude. There are dozens of proposed neural architectures which could do the trick theoretically, they're just way too small right now (similar to how useless backprop was when it was invented).

Are we a Manhattan Project or three away from it? Sure. That's not nothing, but we're also pouring so much money into the boring and immediately commercializable parts of the field (all the narrow perception-level and GAN can that NeurIPS is gunked up with) that if any meaningful part of that shifted to the bigger problems, we'd see much faster progress. That will happen in a massive way once someone does for reasoning what transformers did for text prediction: just show that it's tractable.

Two monitors doesn't work great for me because I always feel like one is "primary" and one "secondary", where all the crap like Slack and email that distracts me goes, which just hurts productivity.

I switched to an ultrawide, and though it's about the same amount of screen space I tend to organize it differently, keeping mostly on-task stuff visible. It's real nice to have 3 tabs of related code and an iOS simulator running as well as a console without having to tab switch.

Twitter's CEO is basically an absentee leader, time-splitting with his other company which actually makes money and spending his free hours sucking up to rappers because he is desperate not to appear as an uncool white guy. Is it any shock that their strategy is garbage?

Until a rollback due to a failure itself fails on a complex stack and prod is hosed for hours while you try to figure out how to unfuck.

Maybe CF has gotten better, but I'm not sure since I totally jumped ship for GCP.

Evil enough that new mocks should be banned by fiat and enforced by code that requires rarely granted permission to override against projects that are 5 years old and already have hundreds of them in place and do not have architectures that support better ways of testing?

It might be a nice thing to avoid them, but you're always working with legacy code, and I've literally had "I need to add a button to accept new permissions" blow up into "I need to refactor our entire class structure across 200 files because I'm not allowed to add a new test that follows our old patterns nor commit this code without coverage, and oh yeah, nobody wants to review that CL in one go so I have to figure out how to break it into 20 bite sized changes. There goes my quarter...". That's just dumb, and that type of dumb is very fashionable at Google.

It's fairly normal for a big company these days: extremely slow pace, red tape all over the place, your level matters more than your skill. It's an engineering-led culture so there's a lot of focus on code nitpicking and purity (arguments over whether mocks are evil, etc).

Great place to work if you're high level (5+) and land on a good team with a good manager. Mind numbingly boring if not.

Java was a good syntax. The language itself had way too many limitations back when Sun ran it, for really dumb reasons proffered by too smart for their own good people. I was sadly paying attention to the wretched arguments about how there was literally no way to reasonably make first class functions work, or closures, or value types, or escape analysis, or any of the million things that C# just said yes to from day one which worked out fine. It was a lot of edge case kvetching that really needed a dictator to cut through, but Sun were terrible stewards and let the worrywarts run the show.

The enterprise culture could I guess be blamed, but Sun was a shite enterprise company when it came to financials, you'd think MS would be much worse in terms of command/control but C# ended up as a damn good version of Java, and once Oracle grabbed the ball hairs Java got pretty great. So it's not the whole problem.

Bayes has you covered, if your priors rank something as exceptionally likely or unlikely, then you probably shouldn't pay much attention to new information (especially when it's from some random person and carries a high likelihood of being misleading).

Extraordinary claims require extraordinary evidence, etc.

If other people want to pay more for your place, why should that place's owner be forced to let you pay less than they could make with someone else?

Seriously. What is better about you, other than that it is you?

"Long-term is better for the community" may be an argument that I can get behind re: a tax, but that's not worth very much at all. Maybe a couple hundred a month in a good location? Less in a poor one.

They do own the fuckin apartment. They should be able to do whatever they want with it, within reason. Go pound dirt may be right.

Fixing a bad law is fine, but you can't propagate that back just because you wish you'd thought of the edge cases then. Or at least you shouldn't, if you have any sense of fair play...

It wouldn't be a README if it wasn't out of date and inaccurate.

Yes, that is what happens when direct democracy is allowed, anyone who can influence public opinion (rightly or wrongly) can make whatever laws they want, and it's a big reason that many people are opposed to ballot initiatives as a matter of principle.

Most people strongly disagree with you. You need to offer an argument, not an assertion, since you're callit for a major and unpopular change.

And if like many of my lefty friends by "abolish" you actually mean something that doesn't even come close to meaning "abolish", you're just causing problems with your abuse of language.

If you fail based on any of the "would be nice to know" items, you were never going to succeed.

Find another few founders, ask them what they use. Find another few who use the same, ask what they think. Repeat.

If you can't figure out how to hire, you're not fit for this.

There's really nothing to any of this that should be at all difficult relative to product/market fit, and the answers to all those other things change too often to be worth enshrining. "Ask other cofounders" is the trivial and 99% correct answer, so why complain that someone hasn't done that for you given that the data will be stale every 6 months?

Actually it's quite easy to cast electromagnetism into a purely geometrical form, you just add a few extra curled up dimensions to a flattish spacetime in general relativity and you basically get Maxwell's equations for free. There are other reasons that's not a great theory (IIRC sources and self-action get nasty and you have to make arbitrary choices about scale, etc), but it's straightforward to get force laws out of pure geometry.

Part of the problem is, a lot of the best people in this field don't even realize how fuckin hard they had to work to get to the point they're at, because they didn't think of it as work a lot of the time. It seems easy until you have to show someone who isn't literally obsessed with it how to do it, at which point you start to realize how much experience and reading and studying and experimentation you're drawing from that you can't really skip over without missing important stuff.

There is a huge element of luck or privilege in having the intellectual space to be obsessed with code/math/business/design/etc at all, which factors in heavily as a filter, but yes, you need to do hella work as well. Being naturally smart is super helpful, too, and can cut the amount of work by a large factor, but by itself it's not enough.

> But as the C++ community dreams of better tools, and dependency managers, maybe it would help to define the fundamentals more accurately: Our programs are a set of definitions, some of which are provided and maintained out-of-tree by other people. I think the more closely our tools adhere to that model, the better off we will fare in the long run.

> So maybe we need to ask fundamental questions about the compilation model and examine some beliefs we hold (For example “Compilers and build system need to be kept separated”. Do they? To what extent?).

I think you just put your finger on why Google is a monorepo - Blaze/Bazel kinda suck horribly (they feel super dated and could have been made much friendlier if they'd been designed from the ground up knowing what the pain points are, and esp if they weren't basically an ad-hoc set of new languages/syntaxes), but they're also awesome. The huge benefit is that you can make tools that work across everything and track monstrous dependency graphs.

Calculus was the worst for this: in my first class we spent literally two months dissecting the minutiae of limits and Lipschitz conditions and infinitesimals and blah blah etc, only to get the the punchline, "and you find the slope of a function by doing the obvious thing, which works in the obvious way every time you'll actually be using it in practice". I get it in a college level analysis class or something, but as an intro in high school that's just a great way to make students hate what is at its core a very simple and useful subject.

It is very true that GPT-XYZ is fundamentally limited, like all transformer models, but don't misattribute the root cause to the fact that it's just processing text.

The real limitation is that these are feedforward networks that just do perception without any processing of what they've perceived. You can try to hide that fact for a while by increasing the depth of the perception network, basically hard-coding some processing into the single pass, but you're still not capturing any of the "absolutely requires self-feedback" behavior that we care about that takes a human more than a split second to do (aka almost all actual thought).

A statistical model that made good use of what was coming in could absolutely learn (to take your example) math at every level based on nothing but text, transformers are just nowhere close to having that capability because of their design limitations.