Cache-like storage, private user-specific data, blobby or otherwise schemaless data. Stuff like that. But IMO it’s a matter of time until you find a need to operate against this data relationally, and then you regret using document storage. I’ve made this mistake twice now and do not intend to make it again. I now consider document storage architecture to be a performance optimization with significant tradeoffs, and not a choice to be made by default for nearly any scenario.

I really hate these awful “puzzles”. They only work by the asker intentionally withholding what, if any, constraints exist in the problem space leaving it totally vague, but of course there ARE secret constraints revealed if you violate them with your answer.

Me: “I do it without flipping any switches. I just ask the lightswitches which one controls the light, and they tell me.”

Interviewer: “That’s not allowed.”

Me: “Well what exactly is allowed? Can I pull the cables out of the wall and see which connects to the bulb? Oh, I bet that’s not allowed. How about I open my smart home app and just check which of the smart switches is labeled for it? Oh, I bet it’s not a smart switch so I can’t do that either? Oh, then the bulb has a chime that boops when it comes on, so I just listen for the boop. Oh that’s not allowed either? Wait wait wait, the walls are glass, so I just watch to see when the bulb comes on when I flick the switches.”

Even the canonical answer makes a dumb assumption. Ordinary LED bulbs don’t get hot.

Oh you mean like how AI for driving changed cars so nobody drives themselves any more?

Well no shit. I don’t think in sentences, I think using imagination which sometimes involves language but often doesn’t.

“Run ChatGPT - Free AND Faster”

That’s what the post title should be. And it should give users a one-click cross-platform setup to do it including iOS and Android. And the result should actually be comparable.

Like, I am not fucking with you. If OP really believes in what they’re posting, this is the messaging required here and everywhere else to make good on the goal.

End users care about cost, convenience, and quality. Deliver all three, and wrap your messaging around it.

If actually delivering this is not possible, then there’s your answer to why end users pick remote third party paid solutions, and OP should delete this post.

Not exactly aimed at language keywords (although it is aimed at the language designers who decided abbreviations in keywords are acceptable):

I hate abbreviations in source code so fucking much. Reading is more of software engineering than writing. If you cannot be bothered to type a whole word because typing is hard for you, find a different job. Do not force others to engage in mental gymnastics to understand what the fuck a variable or function is supposed to mean.

I had similar frustrations with contemporary science in other fields too recently, for slightly different reasons.

In my case, I was frustrated with all of the studies related to life on Mars where no matter what data scientists crunch, rocks they look at, and gases they sample, the only indisputably meaningful study with a conclusive outcome about whether Mars contains life will be one looking at microbes with a microscope on a rover. Any other study will only amount to weasel words that decisively conclude nothing other than that they collected data points, and the data points look like “this” that “maybe mean something, but more studies are required”.

I could interview college students about their favorite ice cream flavor, come up with fun statistics, and conjecture about why the data looks the way it does. And at the end of the day, I would have no conclusion beyond exactly what my study’s data is. But the point of science is to advance useful knowledge. Knowledge is what allows us to predict the future, to make informed choices, to avoid guesswork. Raw data is not knowledge.

Let me repeat this again: DATA IS NOT KNOWLEDGE.

You think the QA was done by a person? I’ll bet this was Vibe QA’d.

Well, they don’t much any more. The normies moved to apps.

Uh…they host AWS via Route53. Is AWS alone not considered big enough to be “dogfooding”?

The article, at least, doesn’t seem to try to define or measure “productivity”. Well no shit people are going to be happier not being forced to go somewhere for some period of time five days a week.

Am I happier working from home, or having the choice to do so? Sure. Their data strongly backs that. Do I actually get my work done equally well? For me personally yes but anecdotally group decision-making in remote contexts is much slower.

The research here is ultimately pointless, because it drives zero action to the people who would be deciding WFH policy who are making that choice based on business goals, not personal goals. It might inform politicians if they’re driving policy to promote remote work, but without data about productivity tradeoff or lack thereof, there’s no informed decision to make.

The author’s complaints are about abstraction, not scale. Microservices exist to be an abstraction. They have some properties that allow for scaling horizontally in certain ways, but scale is not their primary purpose. The same goes for many cloud services.

Risk mitigation is the primary reason that hard architectural boundaries are used. And by risk, I refer to business risk.

What are the chances a given change will negatively impact the business? How broad might the impact be? Can service be restored quickly without critical data loss?

Microservices allow a business to limit risk by placing very hard architectural boundaries on different requirements. For example, having a billing service that only receives updates when billing logic changes is much lower risk to the business than a billing service that updates because someone needed a newer version of Go to import a package used by their unrelated image compression function elsewhere in the application. What if the update causes the application to crash unexpectedly? What if the update causes a performance problem that starves other work on the machine? What if a supply chain attack was accidentally introduced by the new package version? What if the server configuration changed to support that new runtime version, and it broke other functionality?

The author’s example of managing a Python codebase with custom tooling to enforce certain interface boundaries in the code is not a very broad solution for mitigating business risk. What happens when you need to update Python? The entire application is affected. Perhaps the author’s application is trivial to test, but in my experience, testing complex applications is very costly.

Information

This is the sort of engineer getting wholly replaced by Copilot

No because Neil was short for Neilbert

MCP tooling exists because sometimes your need to reduce heuristics to get the dumb LLM to do something useful. Tools do things programmatically correctly every time based on what they’re programmed to do. The only thing the LLM has to do is pick the right tool, provide inputs, and read outputs. The LLM can fuck up in all sorts of ways, and MCP is a way to narrow the sorts of things it can fuck up on.

Gaming audio still hasn’t caught back up to what is was with EAX 5.

I hate paper money.

Currency was not introduced historically as a store of value. It was and still is a mechanism to represent debt. It is one of limitless replacements for “People are indebted to me and I can transfer some of that to you in exchange for X”. Paper, being anonymous, has no intrinsic capability to represent that underlying debt relationship.

This is why it’s possible to manufacture fake “debt” via currency counterfeiting. But you know what cannot be counterfeited? The actual debt ledger. Sure, you can defraud someone, or hack into the ledger to record fake entries, but there’s always another party listed in the ledger, and the two parties will be able to confirm, deny, or otherwise dispute the authenticity of that ledger entry.

Most importantly, the ledger can be reversed, at least until it’s “resolved” and the numbers are no longer pending. If someone steals your cash, or your crypto, or your gold, or your goats, too bad. If you’re lucky, maybe some of it can be clawed back by you or law enforcement. But ledgers allow disputes to be handled before debt changes hands.

I can have my bank reverse fraudulent charges if I request it within a day or two. I can have Paypal freeze a transfer. But I can’t un-steal my dollar.

People still read constantly for pleasure. It’s just on their pocket rectangles looking at addictive social apps. Just like how long-form video (cinema) has fallen in favor of bite sized pieces, so too has long-form prose.

“Reading” is also not a standalone activity like it was decades ago. People engage in all sorts of activities that incorporate reading but aren’t exclusively reading. You might read short stories in an RPG like Skyrim by picking up a virtual book, but if someone asked you what you were doing, you were “playing a video game”, not “reading”. You might be scrolling through a long article posted on Reddit, but you’re not “reading”. You’re “browsing Reddit”.