Hacker News Reader

Comment by dagw | Hacker News Reader

If I played competitive video games seriously I would absolutely want them to take cheating seriously and take any necessary measure to stop it. At the same time I think they should also make the most serious kernel/driver level anti-cheat software optional. If you don't to want to install it you can still play the game, just not on any of the 'pro' servers.

Think of it a bit like doping tests in sports. As long as you're just competing in amateur races for fun then take all the steroids you want. Once you want to start competing in serious races be prepared for people demanding blood and urine samples.

View on HN

The serious cheats in competitive games these days don't rely on hacking the game anymore. They use a separate computer connected taking your HDMI out as an input and then controls your mouse and keyboard inputs. They use ML models and OpenCV. You watch the second computer's monitor and it draws huge boxes around things you shoot and it assists your aim. It doesn't even have to be a PC game, you can do the same with consoles and a computer.

Kernel level anti-cheat is meaningless.

Do you have some links for this? It sounds super interesting.

https://arstechnica.com/gaming/2021/07/cheat-maker-brags-of-...

There are also videos on Youtube demonstrating it in actions

https://news.ycombinator.com/item?id=27790723

But there is plenty when you google the key words.

The setup makes it impractical. You would need a very powerful computer to run it in real time. And it won't tell you about what u can't see. You would need to do the same for audio and game mechanics.

Check what a top player can do: https://youtu.be/8Qf5Xlty1Cg I doubt an AI will be anywhere close in the next 10 years.

The AI doesn't have to play the game, just recognise enemy players and calculate how many pixels to move the mouse so that you are aiming at their head. Even the top player you showed could probably have improved his run quite a bit if instead of having to aim he got a guaranteed headshot every time he fired his gun in the vicinity of an enemy.

Cheats like this are actually ruining high risk/reward stake games like Escape From Tarkov & Hunt: Showdown.

You have to tune cheat tools down or your stats will get you banned. An abnormally high headshot rate is a sure way to get banned.

That's where anticheat is going imo: algorithmiclly defining what the fastest more precise human should be capable of. Anything beyond that is "cheating".

Too much ego, friend. You should've used the past tense.

Skill tunable AI, developed by a private poorly funded group, that plays the game for you https://www.youtube.com/watch?v=revk5r5vqxA

AlphaStar, developed by a large corp, mopping the floor with Battle.Net's Starcraft 2 players https://www.youtube.com/watch?v=o1zhmGPmAAM

OpenAI, developed by a large corp, beats top DotA2 team https://www.youtube.com/watch?v=tfb6aEUMC04

On one hand, it's exciting tech [old] news. On the other, people who play competitive games must hope for a solution against abuse. The rest of us must hope it isn't an invasive solution such as, what effectively is, kernel malware.

I like to think this crowd is above typical streamer hysteria, but I guess not?

AlphaStar and OpenAI have nothing to do with the conversation here, right off the bat.

This video is so cringey too with random burps and "skynet" references. The external hosting is not related at all to the "soft" cheating, and that in turn has nothing to do with AI. Packet sniffing vs OpenCV vs Tensorflow for finding a head is not a meaningful difference.

Years ago, and I mean years ago cheat developers were building this type of "soft aim" cheats that don't snap.

And the separate input via controller _is_ detectable if this becomes enough of an issue... your "AI" is generating input and it often will generate input that controllers wouldn't. Because the "AI" is ML and only being applied to finding on screen targets.

Making a second model to generate realistic input would be a whole other achievement no one has even bothered with because... they don't need to.

On-host hacking is rampant and these games can't seem to stop it, so no one has bothered.

I'd be for kernel-level anti-cheat if it actually worked, but script kiddies are beating this "kernel level anti-cheat" handily.

To me the answer is boring simple statistical analysis, increasing the cost of recovering from bans, etc...

But that doesn't sound as sexy so I guess we can keep this farce of a cat and mouse game going.

This is why anti-cheat should be performance based. Build a model based on the way that verified elite players play the game, and their performance. Compare “normal” players against the model and look for outliers. Manually review accounts that look suspicious.

Cheat behavior is obvious. Ridiculous long distance accuracy. Weird movement. Weird map awareness. Weird awareness of other players. Etc.

Incorrect, there’s lots of cheats that are completely seamless and just improve your performance without being a completely binary “perfect headshot” vs “always miss” or something.

I guess your last encounter with aimbots was in the early 00s…

If other players can't tell someone is cheating then it's not actually a problem since everyone will feel the game was fair.

In competitive gaming, having an actually fair game matters more than having a fair feeling game.

In proper competitive gaming (ie. with a prize pool) you can apply much more rigorous standards like kyc, a live video feed; even standardized hardware and drug tests.

In magic the gathering, the standards are divided in three: casual (called regular), competitive and professional. Basically the competitive mode is for people who play the game to win, sometimes with small prizes, but still want a very clean playing field. I feel that the standardized hardware and drug tests are feasible at professional, but not competitive, and anti cheat rootkits is interesting in competitive.

Ah, the old Lance Armstrong argument.

The area between thinking others are "absolutely not cheating" and "definitely cheating" is huge, and where most online gamers end up spending their time. Absolutely cheating is not a problem, because most games allow you to report people, and they'll get enough reports and be banned shortly. Not being able to tell is the problem, because you feel something was unfair (which you might also feel when it actually is fair), but you feel it more often and it leads to poor experiences and people quitting the game.

Cheaters quit after a while because the game loses its challenge, and people being cheated leave because the game.feela unfair, and eventually there's not much of a community left.

The great part is many games have the aimbot built in. Players use a hardware dongle to make a keyboard/mouse look like a controller input, and the game applies aim-assist to the keyboard/mouse input.

You don't get mouse axis movement though -- it emulates it through a Controller stick so it's not exactly an ergonomic and mouse-like experience, from what I have heard.

Have you ever played Warzone? People sniping 700+ meters and they can't even see you on their screen. Obvious map hacks, aim bots, and even some hilarious shit like super speed or the ability to drive a car through walls to run you over.

Besides, it's not just a matter of improving accuracy. Elite players are elite in multiple facets of the game. Movement, general awareness, map awareness, accuracy, etc. Many of these metrics are easy to measure.

Cheaters will show up as outliers and it is at that point that their accounts should be manually reviewed.

With casual gaming I really don't even care if someone is "cheating" to improve so long as I feel the game was fair. I can put a red dot on my screen to improve my aim, as an example.

Warzone is weird. I can't understand why speedhacks, 'silent aim' (you can shoot in a direction different from your view angle) or flying cars are even a thing.

This was impossible in Q3A (1998) without introducing an input lag thanks to client side prediction. Server also didn't send you updates on other clients that are too far away, behind a hill - you'd pretty much only receive positions of those you could see, hear or that could see you. In Warzone, you can see everyone on the map with a wallhack.

I really wonder why the game is designed this way. I guess it might be to limit required computation power of the servers? They still do stuff like hit detection, so server surely has some idea of the physics.

On top of all that, this stuff is possible even today, months or even years after such exploits being widely known. They do have a kernel-level AC driver, but they don't even seem to detect trivial hacks from the server-side.

They did make one. It was called FairFight[1] and was mainly used by EA in the Battlefield franchise. It also was a dumpster fire: bans were applied to legit players all the time, moderators were overwhelmed and manual review took too long and many times resulted in nothing.

Behavioural anti-cheats are not the answer. Implementations are crap and under-developed, and I don’t blame them: you would need tremendous (server-sided) power to correctly process the models considering how complex multiplayer games are, and it just isn’t worth the investment and complexity to them.

It’s just cheaper to pay rootkit developers (like EAC or BattlEye) and knowingly infect client devices. Hell, their drivers are even WHQL signed by Microsoft. And when false positives hit? Just blame the anti-cheat vendor. Potential HR problem dodged.

[1] https://www.i3d.net/products/hosting/anti-cheat-software/

Valve do this, here's a good talk:

Robocalypse Now: Using Deep Learning to Combat Cheating in Counter-Strike: Global Offensive https://www.youtube.com/watch?v=kTiP0zKF9bc

This is kinda weird, CSGO has spinbots and xp-farming bots: one moves faster than is humanly possible, the other moves on the spot or in a circle with exact constant rotation ... the only conclusion I can make is CSGO want these bots, because auto-blocking them seems trivial.

Why use deep learning when so, so many bots (and their host accounts) could be blocked with trivial heuristics? My only guess is allowing spinbots/farming somehow provides a financial gain in CSGO?

> My only guess is allowing spinbots/farming somehow provides a financial gain in CSGO?

Valve devs have a lot more leeway than your average developer. They honestly might have just thought it would be cooler.

Beyond that they now have the infrastructure to detect more subtle forms of cheating through deep learning using Overwatch (the reporting system, not the game) as ground truth data to train their AI.

As a former CSGO player (not a very competitive one) this trust factor really sucks, and Valve oversells it.

If you play with lower ranked friends and do really good in matches you will get reported a lot which will tank your trust factor and almost guarantees it that will only see hackers from them on.

They say they use other factors like account metadata, but I have a 10 year old account, with purchased games and verified phone and still got "Low Trust Factor" after playing with friends and started to have an obvious hacker every other match.

That is why I am a former CSGO player

If you want to detect obvious cheating behavior shouldn't you train the recognition software on cheating behavior not elite players?

The average cheater is most certainly not investing in thousands of dollars of hardware to run intense AI models on a second high power machine. These are mostly theoretical hacks.

The average cheater is paying $10-50 for cheat software. You may not like the anti-cheat software, but it works against the most common cheats.

I think they could have an opt-out league that doesn’t require anti-cheat but remains separate. You might be one of the few non-cheaters in there, but at least it would be an option.

No, not even. Your average cheater these days are C and D tier streamers who feel a pressure to perform and have a budget. And it's far easier to do this than you think. They keep getting caught out doing this type of cheat specifically.

They're buying trained models from people who have the fancy hardware.

You don't need thousands of dollars of hardware. You need one (1) Raspberry Pi 4, plus an HDMI capture card and a Raspberry Pi Zero to act as fake mouse/keyboard. You can totally do that for $50. (Probably less if you use an Arduino clone instead of a Raspberry Pi Zero.)

Heuristic software works against the most common cheats. Standard anti-piracy tricks work against the most common cheats. Kernel-level anticheat is only useful against a very specific kind of high-end cheat, and if you're going that far, you might as well use an undefeatable hardware cheat.

You're still talking about a bunch of work and effort. The people aimbotting in CSGO aren't doing it because they love hard work and a challenge.

It boils down to a risk assessment. The fact is people interested in cheating continue to escalate the norm. That's the arms race.

Meanwhile we are letting shady companies illegally spy on users [0] and mine crypto [1], for what? What is the gain? What is the value proposition of kernel-level drivers in the anti-cheat application? What's more, the average gamer doesn't know anything about computers. How can they be expected to make an informed decision about whether or not they should buy a game with kernel-level anticheat if they don't know what that means? Even worse, there is no legal requirement that a game's marketing material say anything about its anti-cheat implementation. These aren't tin-foil hat concerns, these are the enablers of real scandals that have happened and are happening.

0. https://www.wired.com/story/kernel-anti-cheat-online-gaming-...

1. https://play.esea.net/forums/492152

Let's not forget that a large need for anti-cheat comes from focus on centralized matchmaking overr community-hosted and community-moderated servers that were more common before the game industry's focus on monetizing their games to the last bit.

The people they do buy cheats from do the work and effort. They don't seem especially bothered by kernel-level anti cheat stuff. It keeps them in business, since they get less quick home-brewed competition.

And just consider how much money flows for "micro" transactions. There is plenty of "players" willing to pay money for hardware cheats.

My observation is that competitive online gaming is over. It's just a cesspool of cheaters now.

So you're arguing that all these attempts greatly raise the skill ceiling / investment cost to cheat.

Sounds like it works, then.

“Did you do it?”

“Yes.”

“What did it cost?”

Heuristic anti-cheat works about as well as heuristic anti-spam. It might actually catch more legit players than cheaters, due to the false positive paradox[1].

[1]: https://en.wikipedia.org/wiki/Base_rate_fallacy

The big issue is not the existence of cheating, but the scale of the problem. A software hack is infinitely scalable, a hardware setup is not.

Some cheat dev only needs to develop that "high-end software cheat" once and can sell it millions of times around the world, with super easy distribution over the internet. A setup with hardware video capture costs $50-$200 for every single user for hardware alone, and needs at least a bit of computer knowledge to set up.

> it works against the most common cheats.

Emphatically, it does not. Writing exploits and mitigating them is an arms race, and it is always easier to exploit than it is to defend. The entire reason for such software being sold is after all that it works, even when it is the bad/easy fundamentally detectable kind that runs inside the same OS as the game.

Any links confirming the existence of such systems?

While there are convincing looking proof of concepts for computer vision based cheats, I haven't seen anything to suggest that this approach is used meaningfully anywhere. And considering the millions of cheat bans that get processed in mainstream games to this day, I think its safe to assume most cheaters aren't using anything sophisticated.

Citation needed, seriously. These things need to be crazy fast to process and respond fast enough.

no this is wrong, these cheats are completely unreliable, most people use DMA devices and NeTV2 overlay for league ACs

> Kernel level anti-cheat is meaningless

Anti-Cheat on the Account-Level is meaningless. You ban a cheater? Who cares. He will back with a new Account some Minutes later. Hardware Bans are useless aswell, as they can easily be spoofed. The only effective strategy for Anti-Cheat is a Universal Identity Provider, which can proof you are a real Human with no Cheat Record.

I never got how this worked on Steam. You have a consistent Steam ID, which the game is tied to, and all your achievements/ranks/etc.

Isn't that what you are talking about here?

You can create a new Steam account, sure, but that would start you again from scratch as a total newbie (and losing access to anything you'd bought)

I'm amazed at how much money some people have to burn on new accounts. Looking at minecraft servers, there are so ridiculously many people banned.

I've wondered if there's some sort of economy here, i.e. someone with a ton of stolen accounts monetizing them by offering to cheat all day on your competitor's server.

Oh yes, absolutely. There are a number of "alt shops" selling Minecraft accounts accounts for a few cents each.

Most of these accounts have security questions set, so you can't change the email and truly "take ownership" of the account, but it's good enough for cheating on servers.

Some shops even have an API for generating accounts or tokens, so your hack client can have a feature where you get a new account right from your game. Get banned, reconnect in a few seconds.

The last point sounds like a security vulnerability of some sort? It shouldn't be hard for Mojang to issue unforgeable tokens.

I'm talking about steam accounts. There are many cheaters our there, who just dont care if they lose an account. They will cheat on an alternative Account and when they get banned, they will buy a new one. They dont even care spending some bucks on the Game again. Edit: Take Rust as an example. A year ago, the Developers made some stats public: 12 Million copies sold, 600.000 Banned Accounts. Thats 5% of all sold Accounts. There is even a Twitter Feed, showing all the Bans:https://twitter.com/rusthackreport Most of the Accounts will show a Playtime of under 30 hours in the Game, mostly no Friends or other games, indicating they were only made for cheating.

Griefers gonna grief. Some people aren't happy unless they're making someone else miserable :(

Side question but what's up with all the capitalization?

> Kernel level anti-cheat is meaningless.

If kernel level anti-cheat lead to "serious cheats" having to resort to the above, then kernel level anti-cheat is actually pretty fucking effective.

> The serious cheats in competitive games these days don't rely on hacking the game anymore.

This may be true for a class of cheats, but it isn't broadly true. That means these category of cheats can only enhance what is already visible.

In FPS games, being able to see players position through walls is a super valuable cheat _and_ requires access to the underlying game engine (to extract player location).

Every time I hear about the seeing through walls hacks, I wonder why the developer just doesn't send the other users locations unless it should be visible.

Maybe because doing the math to figure out if it should be visible is handled completely on the frontend to keep latency low?

This is the equivalent of a bank API giving you everyone's information, and the frontend SPA filtering it to only the logged in user.

Here's a write-up about the subject, as implemented in Riot's Valorant, if you're interested:

https://technology.riotgames.com/news/demolishing-wallhacks-...

And aren't the existing anti-cheat technologies the reason that cheats take that form?

You see no cheap aimbots and conclude the anti-cheat is not relevant anymore. May be the right conclusion is that they're working.

Not saying this justifies the kernel drivers. I don't trust my gaming computer precisely for this reason. I wonder if some form of containerization/virtual machine could make it possible to both allow for this kind of kernel-level anti-cheat without exposing my whole system to untrusted code.

> I wonder if some form of containerization/virtual machine could make it possible to both allow for this kind of kernel-level anti-cheat without exposing my whole system to untrusted code.

No. Because then you could write an aim-bot that runs in the host machine…

As someone who does play competitive video games seriously, rather than having anti-cheat, I would prefer to play on servers/ladders where the players are vetted and where getting caught cheating has meaningful social implications. Open-for-all servers with accounts that are trivial to make is the problem to me.

Never trust the client. Trying to install a rootkit on your users' computers won't change the reality of client-server security.

I agree, and I think that like DRM anti-cheat is generally a customer-hostile losing battle. The big problem is that companies are trying to solve a social problem (cheating) with a technological solution and, predictably, failing. Partly this is motivated by profit: if you required online gamers to be real verifiable people, with all suspected cheating was judged by a human being, and bans were permanent and landed you on a watch list for other games, cheating would probably drop dramatically but so would sign-ups and your ability to resell the game to the new account of a cheater who's old account was banned.

Personally, I'm completely fed up with online competitive multiplayer for other reasons so it doesn't really affect me. I just don't by the games that would use kernel-level anti-cheat to begin with.

> I agree, and I think that like DRM anti-cheat is generally a customer-hostile losing battle. The big problem is that companies are trying to solve a social problem (cheating) with a technological solution and, predictably, failing.

I don't disagree with your general take, but you can't accuse Valve Anti Cheat of being a failure.[0]

[0] https://en.wikipedia.org/wiki/Valve_Anti-Cheat

I think a lot of the prevalence of cheating has to have some economic roots. A certain kind of cheater, the kind who buy new accounts (maybe with new cosmetics) every time they are banned, isn't necessarily someone a company wants to get rid of permanently.

The benefits of this sort of income may be high compared to the benefit of being a "serious competitive game".

I think Minecraft servers are instructive, because they're often run just as cynically as free to play games, but with less competence, so it's easier to see what's going on. You do see a lot of pay-to-unban there, for instance.

We have to trust the client to an extent. How can a server tell if a player clicked their mouse or if a triggerbot did?

The officially approved mouse timestamps and signs the click using a hardware embedded cryptographic key (only slightly joking).

That's when you just rig up a solenoid to click the real mouse :P

fingerprint/palmprint sensor on the mouse... plus of course timestamped continuous video stream of your face and both hands at all times.

Pssh, that's when you just rig up a hyper-realistic android.

It can't. There's not much more to say, even with the current anticheat it's no way to be certain a real human being clicked a real mouse. As the other comment shows, you can start adding more and more invasive monitoring solutions but this likely won't be tolerated by many and will eventually be bypassed anyway.

Never trust the client

I wonder if we'll eventually end up at a point where all competitive gaming is run entirely sever side and IO is streamed to cryptographically signed and verified hardware.

The problem with this logic is that most multiplayer games have post-sale economies with cosmetics and other add-ons, and I would guess that most of those sales comes from casual players.

So while anti-cheat helps protect the pro/serious players (who in some ways are community/marketing leaders in this world of Twitch and YouTube), it also creates a fair environment for all the casual and aspiring players who keep spending money on the game in droves.

Your argument falls short.

Pro cheating uses hardware to cheat. It streams the video output and keyboard/mouse input to an AI/ML. Making these kernel anti-cheat useless there.

Casual players looking to spend money on cosmetics aren't affected by cheaters because the system influencing the players to buy isn't affected by the cheaters. The way most of these "influence systems" work is by using marketing strategies. It'll augment your match, or otherwise the players you meet, to see cosmetics you don't have. Coupled with gambling mechanics to induce addiction and sheer marketing material ("best offer now!") makes for this whole pro-kernel-anti-cheat industry move highly dubious.

Source: i worked at a company that writes such systems. The day I decided to quit was when my naive ass heard the team lead, on a recruitment call, explain to a candidate what we do: "we refine a company's ability to get kids addicted".

It's a completely different game (no pun intended) than what news outlets want us to think.

> Casual players looking to spend money on cosmetics aren't affected by cheaters because the system influencing the players to buy isn't affected by the cheaters.

This is a rather confused thing to say. Are there players out there that just grind cosmetics and don't care about win rate, game quality, or wasted time? Probably a handful.

But the overwhelming majority of players in a game are there to have fun and cheaters utterly destroy that. Cheaters, smurfs, etc, are a plague on online multiplayer gaming where one troll ruins everything for everyone else match.

At the end of the day it's just a game, and if the game disappears or whatever humanity will go on. So let's not lose perspective. But saying casual players aren't affected by cheaters is flat out wrong.

I worked in gaming as well, so I'm familiar with the issues you're referencing, but hyperbolic exaggerations don't help us address it.

Then simply have the anti-cheat option be the default option? Most casual players will not bother opting out.

> Just not on any of the 'pro' servers.

No, the other way around. On "pro" servers there is moderation that makes cheating a non-issue. It's in the casual space that cheaters ruin everything, because there isn't any human moderation.

I'd phrase it as "you can still play the game, just not against other humans".

"...other humans that you don't trust"

I often care about online multiplayer, but only with my friends (who I trust not to cheat). I don't want to install a rootkit for that.

I agree but it should only be present for pro competitions (and by pro I really mean playing is their job).

Why not let the players decide if they want to play on the 'serious' server with serious anti-cheat or on 'fun' server with minimal checks.

I disagree. As someone who plays online shooters only casually once or twice a week, nothing sucks more than having a full session in Apex ruined by an obvious cheater. The fact that I don't spend 8 hours a day doing this makes it far worse for me, the time for gaming is very limited when you have a small child and having it ruined because of some script kiddies is incredibly demoralizing.

Let me put it this way - I'd much rather play an online shooter with a hardcore kernel level anticheat than a one without it.

A kernel anti-cheat isn't going to fix that, by far the best servers are community moderated and usually cheats aren't really complicated.

The problem is that community servers are a thing of the past for most games.

Yes, but as many other people have said - no technical solution will elminiate all cheaters, but aggressive anti cheat does make a positive difference to the experience. It can make a difference between running into cheaters in 1 out of 10 games vs running into cheaters every single game.

And yes, community moderation is great, except that it has mostly gone away in most games.

I doubt that there's anything a kernel-level anti cheat will catch that a normal won't (if we're not talking about very sophisticated cheats we see in pro competitions of course). I'd happy to be proven wrong but those companies don't even release statistics on the efficiency.

You used to be able to vote-kick the obvious cheater, not sure why that is not an option anymore, these days.

Abuse usually...

Yep, people just spamming cheater reports on someone who killed them, it's worthless without human verification and that's expensive/difficult.

It's already human verified if you require say >50% players to vote, assuming the cheater is not running a Sybil attack too.

Yea but you:

a) can't allow "vote to kick" from the opposing team in a competitive environment. That has obvious abuse written all over it, either intentionally or unintentionally (opposing team is losing or opposing player is legit good but they think he is cheating) b) when you allow vote kicking of "same team only", like CSGO, there is a lot of grief type of harassing where parties of players (like a 3 queue) can vote kick teammates just for general griefing reasons, intentionally deranking, and the like.

> If you don't to want to install it you can still play the game, just not on any of the 'pro' servers.

The worst thing gaming has ever done was think it was anything like professional sports. I think that I'm in favor of cheating in games just because it rankles people who take gaming way too seriously.

I'd be interested to hear your opinion defending professional sports taking itself as seriously as it does then. From my perspective, the only thing that makes it serious is the amount of money and advertising involved.

Sure the technological barrier is important, but I fail to see how they're categorically different depending on what they require of the participants.

I don't have an issue saying they both should not be taken seriously, or that both should be taken seriously, but to draw a line between them in terms of "seriousness" I think needs elaborating to me.

> I'd be interested to hear your opinion defending professional sports taking itself as seriously as it does then.

Also stupid. Athlete worship is as bad or worse than celebrity worship. The obsession with and tribalism from professional sports has creates some of the most obnoxious bores in the world. All for a bunch of people kicking or throwing a ball around.

> people who take gaming way too seriously

Every activity has its own popularity, and number of fans of various seriousness... Cat and dog shows, caber toss, curling, singing competitions, on and on and on, the list is about infinite, since it encompasses everything within which someone could potentially be more skilled.

Singling out video games among all the other things humans do competitively feels like a personal bias.

I personally don't care about probably 99% of the 'competitions' out there, but I don't give anyone crap for being obsessed about their own thing.

(I don't play games online multiplayer style, so I'm not in the target group anyway.)

I don't think there's such a thing as "gaming", as you use the word... everyone involved does their own thing, in infinite variety as well.

22 adults competing over a ball isn't inherently any more serious than 2 people playing a game of Starcraft.

I don't think highly of regular sports either.

Reply on news.ycombinator.com