HNHacker News
TopNewBestAskShowJobs

smikhanov

1,958 karma · joined November 26, 2008

My name is Sergey

My blog: http://www.mikhanov.com

My Twitter account: http://twitter.com/i_am_neuron

Feel free to get in touch with me about anything (I live in London): http://www.mikhanov.com/contact

submissionscomments
smikhanov··on Key IOCs for Pegasus and Predator Spyware Removed with iOS 26 Update
> But there is no such possible way to use an iDevice without relying on Apple's services.

There is. One can go through the iPhone setup wizard and opt out of everything. You don’t need to have any accounts, neither iCloud nor App Store one, or to be logged on to any Apple services to use your phone.

Someone who knows more about iOS than both you and me could comment further on whether subtle things like aGPS would continue to function as expected, but everything you specifically thought of when you wrote “to use an iDevice” would work.

smikhanov··on A career is a pie-eating contest and the prize for winning is more pie
More like 16 year old, but the rest is spot on.
smikhanov··on X-COM creator Julian Gollop discusses his most important games (2019)
Wow, I didn’t know Laser Squad was done by the same guy (and his brother, full indie-style, it turns out) — I remember playing that game on Spectrum in 1992 or so, and it was absolutely next level, no other game came close at that time.
smikhanov··on Millau Viaduct
I especially like the “Team” section of this page. Great recognition given to everyone who participated in this project, all the way to the humblest architecture school intern!
smikhanov··on Hundred Rabbits – Low-tech living while sailing the world
Yeah, when you take the rent and many of the temptations of the big city lifestyle out of consideration, the cost of living gets surprisingly low.
smikhanov··on Reflections on OpenAI
If we were to believe the embodiment theory of intelligence (it’s by far not the only one out there, but very influential and convincing), this means that building an AGI is an equivalent problem to building an artificial human. Not a puppet, not a mock, not “sorta human”, but real, fully embodied human, down to gut bacterial biome, because according to the embodiment theory, this affects intelligence too.

In this formulation, it’s pretty much as impossible as time travel, really.

smikhanov··on Trans-Taiga Road (2004)

    no settlements or towns aside from Hydro Quebec's settlements for workers (these are private and are not open to the public - they will kick you out)
Will they really kick a passing driver out when it's freezing outside? Heck, wherever the population is this sparse and conditions are this harsh people normally actively invite you to their places. This sounds so weird.
smikhanov··on Addictions Are Being Engineered

    Do the math on how many people are necessary to run a web site with on-call rotation, minimum moderation, and someone to run the business. The number of $2/month subscription required to make that work is prohbitively high.
Is this really so? Let's try doing the math: you're describing a distributed team of maybe 10 people, likely less. Let's assume you need $600K/year to run this business (is this the right number? not sure, feel free to correct me). At $2/month, that requires 25000 paying users.

Difficult, but not impossible. At $5/month (the $3 difference wouldn't trigger any price sensitivity, talking from real experience) that's 10000 users. If your service actually provides value, you can crank it even higher. Again, difficult, but totally within the realm of possible.

    if the unpaid social network competitors can't get any traction, what makes you think adding a $2/month signup hurdle would improve the situation?
Because the "traction" needed to make free and paid social network work is vastly different. You need insane scale (millions or tens of millions of users) to make free social network viable, hell, my stomach hurts from only thinking about it. A paid social network business, run with certain austerity, can be profitable with one thousand paid users.
smikhanov··on Engineered Addictions
Sorry to mumble, I'll add one more thing. Back in 2013, I was running a productivity startup, and we tried courting Evernote into acquiring us (unsuccessfully, though damn we were a good match). I remember those times vividly: Evernote raises a sizeable fraction of a billion in funding in several rounds, employs like 400 people, their CEO goes to places like Le Web or whatever and expatiates from stage about building a "100 year company".

Fast forward to 2022: Evernote itself is acquired by a random app studio, and the whole service is now run by something like a dozen people. The CEO of what used to be a "100 year company" moved on.

I thought about this a lot. They never needed those 400 employees, even in 2013, it was absolutely possible to run the same service with a tiny team, it's just that the people at the company's top would be completely different people with completely different aptitude towards building their businesses. It's only if you really, really want to be on stage at Le Web, then you go to investors over and over again and convince them and yourself that a note-taking app needs 400 employees, and you're building a company as a product, not a product as a product.

Looks like the author of the original article here also didn't actually want to build a social network business but rather wants to be in the hothouse of Silicon Valley. Well, good luck to them.

smikhanov··on Engineered Addictions
Kagi is cool (as a company), really.

The whole idea of running your company while thinking about TAM and whatever is totally from a VC playbook. If you don’t take VC funding, you stop caring about TAM. Instead, you care about whether having the profit you have makes you — yes, you, personally — comfortable about your life. This is a much, much healthier line of thinking than trying to capture every bloody dollar in this world you can reach.

smikhanov··on Addictions Are Being Engineered
But for a paid social network you can do marketing too. All you need is for CLV to be higher than CAC.
smikhanov··on Addictions Are Being Engineered
But for a paid social network you can do that too. All you need is for CLV to be higher than CAC.
smikhanov··on Addictions Are Being Engineered
You description is probably spot on, but boy I’d love to have a version of Instagram where I could just pay $5/month and get a time-sorted stream of photos of my friends’ babies, Piña coladas with a beach background, and sweaty mirror selfies in a gym without any stupid ads in the middle.
smikhanov··on Engineered Addictions
How in the world is it easier to attract people to some new VC-funded nonsense?
smikhanov··on Addictions Are Being Engineered
How in the world is it easier to attract people to some new VC-funded nonsense?
smikhanov··on Engineered Addictions
Good people of HN, could anyone tell me why buying an MVP of a social network for $10K from a Belarusian contractor on Upwork (it couldn’t cost much more, it’s like five SQL tables and a web CRUD) and then charging users $2/month to use it wouldn’t work?

Why does the author need to moan about how morally destructive it was to raise VC? Just run your social network from your bedroom, while asking ChatGPT how to rewrite the landing page in React.

smikhanov··on Low-background Steel: content without AI contamination
Alright, let's run this thought experiment further then.

You suggest a connector to connect to a set of robot-compatible tools, fine. That set is again limited by what the robot manufacturer thought of in advance, so you're out of luck if you need to weld things, for example, but your robot doesn't come with a compatible welder. Attaching and detaching those tools now becomes a weak point: you either need a real human replacing the tools (ruining the autonomy), or you need to devise a procedure for your robot to switch tools somehow by detaching one from itself, putting it on a workbench for further use, and attaching a new one from a workbench.

The more universal and autonomous that switching procedure becomes, the more you're in the business of actually reinventing a human hand.

But let's assume that you've succeeded in that, against all odds. You now have a powerful robotic arm, connected to a base, that can work with a set of tools it can itself attach and detach. Now imagine for a second that this arm can't reach a certain point in the car it repairs and needs to move itself across the workshop.

Suddenly you're in the business of reinventing the legs.

smikhanov··on Low-background Steel: content without AI contamination

    Really, a robot which could literally have an impact wrench built into it would HOLD a SPANNER and use FINGERS to remove bolts?
Sure, why not? A built-in impact wrench is built in forever, but a palm and fingers can hold a wrench, a spanner, a screwdriver, a welding torch, a drill, an angle grinder and trillion other tools of every possible size and configuration, that any workshop already has. You suggest to build all those tools into a robot? The multifunctional device you imagine is now incredibly expensive and bulky, likely are not reaching into narrow gaps between car's parts, still not having as many degrees of freedom as human hand, and is limited by the set of tools the manufacturer thought of, unlike the hand, which can grab any previously unexpected tool with ease.

Still want to repair the car with just the built-in wrench?

smikhanov··on Low-background Steel: content without AI contamination
Legs? To jump into the workshop pit, among other things. Palms are needed to hold a wrench or a spanner, fingers are needed to unscrew nuts.

Cars are not built to accommodate whatever universal repair machine there could be, cars are built with an expectation that a mechanic with arms and legs will be repairing it, and will be for a while.

A non-humanoid robot in a human-designed world populated by humans looks and behaves like this, at best: https://youtu.be/Hxdqp3N_ymU

smikhanov··on Low-background Steel: content without AI contamination
It’s about keeping different corpuses of written material that was created by humans, for research purposes. You wouldn’t want to contaminate your human language word frequency databases with AI slop, the linguists of this world won’t like it.
smikhanov··on Low-background Steel: content without AI contamination
Prediction: there won’t be any AI systems repairing cars before there will be general intelligence-capable humanoid robots (Ex Machina-style).

There also won’t be any AI maids in five-star hotels until those robots appear.

This doesn’t make your statement invalid, it’s just that the gap between today and the moment you’re describing is so unimaginably vast that saying “don’t worry about AI slop contaminating your language word frequency databases, it’ll sort itself out eventually” is slightly off-mark.

smikhanov··on Plwm – An X11 window manager written in Prolog
Looked at the source, it’s so compact, wow.
smikhanov··on AI system resorts to blackmail if told it will be removed
This should be read as:

If you take a huge amount of human-written text soup, train a neural network on it, add a system prompt “You are a helpful assistant”, and then feed it a context consisting of (a) a mailbox with information about someone’s affair, and (b) the statement that this assistant is going to be switched off, then that neural network may produce a text with blackmail threats solely because similar patterns exist in the original text soup.

…and not as:

Warning! The model may develop its own questionable ethics.

smikhanov··on The world could run on older hardware if software optimization was a priority

    I don't know how to solve that problem or even if it will ever be "solved".
It will not be “solved” because it’s a non-problem.

You can run a thought experiment imagining an alternative universe where human resource were directed towards optimization, and that alternative universe would look nothing like ours. One extra engineer working on optimization means one less engineer working on features. For what exactly? To save some CPU cycles? Don’t make me laugh.

smikhanov··on The world could run on older hardware if software optimization was a priority

    for software, the top end of the market is now non-existent
The issue here isn't absence, but misrecognition: the top end absolutely does exist -- it just doesn't always look like what people wish it looked like.

If by "top end" you mean "built to spec, hardened, and close to bug-free", it's alive and well in heavy manufacturing, telecommunication, automotive, aerospace, military, and medical industries. The technologies used there are not sexy (ask anyone working at Siemens or Nokia), the code wouldn't delight you, the processes are likely glacial, but there you will find software that works because it absolutely has to.

If by "top end" you mean "serves the implied user need in the best way imaginable", then modern LLMs systems are a good example. Despite the absolute mess and slop that those systems are built of, very few people come to ChatGPT and leave unsatisfied with its results.

If by "top end" you mean "beautifully engineered and maintained", think SQLite, LLVM and some OS kernels, like seL4. Those are well-written, foundational pieces of software that are not end-products in themselves, but they're built to last, studied by developers, and trusted everywhere. This is the current forefront in our knowledge of how to write software.

If by "top end" you mean "maximising profit through code", then the software in the top trading firms match this description. All those "hacker-friendly" and "tech-driven" firms run on the same sloppy code as everyone else, but they are ruthlessly optimised to make money. That's performance too.

You can carry on. For each definition of "top end", there is a real-life example of software matching it.

One can moan about the market rewarding mediocrity, but we, as technologists, all have better things to do instead of endless hand-wringing, really.

smikhanov··on LegoGPT: Generating Physically Stable and Buildable Lego
Absolutely, that’s without any doubt.

There’s a difference though. The “cool” Terminator Skynet pursues its own goals, and wasn’t programmed by humans to kill. The “boring” insurance-rejecting Skynet is explicitly programmed to reject insurance claims by other humans, unfortunately.

So still, no need to worry about our AI overlords, worry about people running the AI systems.

smikhanov··on LegoGPT: Generating Physically Stable and Buildable Lego
Also why it’s OK to stop worrying about our future robotic (or AI) overlords.
smikhanov··on Notation as a Tool of Thought (1979)

    APL (or Clojure) way is about making your base types truly useful, 100 functions on 1 data structure instead of 10 on 10
If this is indeed this simple and this obvious, why didn't other languages followed this way?
smikhanov··on Stay on schedule with Raspberry Pi Pico W and an e-ink dashboard

    If you’ve got a busy schedule, then it’s all too easy to become overwhelmed.
This is why they want to interest you in something that is not assembled into a finished product, but presented as a bunch of PCBs, and relies on a custom self-hosted Python server for which one can "always choose to add different sources by editing the Python code on the server".
smikhanov··on California Attorney General issues consumer alert for 23andMe customers
It’s not about “the HN crowd”, it’s just that the time has changed so much. Do you remember 2008? Facebook was a swanky way back then to reconnect with your classmates and leave some “Like” things under their wedding photos. Google was seen as a way of organizing the world’s information to make it universally accessible and useful, with their niche ad service being still in limited beta. Twitter was so unknown that it still hasn’t even won a Webby Award.

And 23andMe was already offering a $100 DNA sampling in 2008.

It’s easy to be cynical about this in 2025. Those who didn’t live through the early 2000s, can’t even imagine the amount of optimism surrounding the tech industry at that time. Giving my DNA to a cool new Silicon Valley firm in 2008? Sure, why not, it was like buying a ticket to some utopian future.

← PreviousPage 2 of 12Next →