CrowdStrike Strikes Crowd

2024-07-20T00:00:00+00:00

totally not an AI image ⇧</div>

By now you will have heard about the massive BSOD</a>-Fest that temporarily crippled much of the world’s business and, to a lesser extent, government computer systems.

Just in case you were locked in the executive washroom</a> and missed the details, here’s a witty summary from Fireship</a>: </iframe>

It’s a little bit funny, in a gallows-humor way, when a company that uses militaristic language to sell to airlines and governments – Strike, Falcon</a>! – ends up showing off the fragility of our national infrastructure. At least this time, you have not been eaten</a> by a GRU</a>.

Those of us who’ve worked in the industry were at first surprised that the bosses had gone for such an obvious single point of failure, then – curse you, experience! – not surprised at all.

Even so, we were puzzled that nobody had talked the C-Level into the miniscule engineering spend required to at least test the updates before rolling them out, and maybe consider not rolling them out all at once.

The nerdiverse</a> was abuzz: do the airlines really not have staging?!

No, sadly, they don’t: staging was taken out back during the Great QA Purge, never to be seen again, but worry not: now everyone does DevOps</a> and the quarterly numbers were made!

Polishing Your Attack Surface</h2>
At least among actual tech people, a lot of fingers are pointing at the prevalence of the “security audit,” in which a third party (often a government agency) comes with a checklist of “best practices” (some more best, some less worst) and requires your company/agency to tick the expected boxes. If you don’t, you will not have any cake! And since the cake is often made from juicy taxpayer dollars, you really, really want some.
This creates a perverse incentive</del> market for some other third party to offer a solution of some kind that ticks a box for you, or ideally multiple boxes at once. Provided all these solutions add up to less than `$CAKE</code>, why not? You get the boxes ticked, and you save yourself the trouble of properly understanding the issue. The people who showed up with the checklist don’t understand it either! Nor do their bosses!`
`If things go south, you will be in good company: nobody was ever (yet) fired forbuying</a> CrowdStrike! You have your third party ready to take the blame, fall on the sword… oh wait, no, not actually fall on the sword. But take the rhetorical blame, oh yes indeed: George Kurtz</a> is deeply sorry</a>.`
`It’s perhaps too easy just blame executive sloth and magical thinking, and/or the Administrative Deep State or whatever it’s called now. Because there’s one other thing you need to have done before laying down on the tracks for Mr. Kurtz’s null-pointer</a> locomotive.`
`You had to ignore all the engineers, former engineers, people who understand computers, and random critical thinkers who tried to explain to you what a terrible idea this was. That, friend, takes a culture.`
Fragility is a Choice</h2> Electing Peter Thiel</a> Vice President is not gonna fix this. Nor will more stealth protectionism</a> from Margrethe Vestager – a monocrop of SUSE</a> is not the answer</a>, and folks like SAP</a> are in the same game as CrowdStrike. The rot runs deeper: this is a direct result of the many, many people who knew better having had the give-a-shit beaten out of them over long years of stock-price maximalism and bureaucratic self-love. The good news? You can be the change you want to see in the org chart! Here’s my eight-step path to not get CrowdStruck, follow me for more life advice: No systems monoculture, everything runs on a mix of operating systems.</li> That goes for clouds too.</li> Everything Windows is inside a VM whose host can automatically roll it back.</li> Threat models include your vendors.</li> Rolling updates, and don’t roll too fast.</li> Stress-test everything first, i.e. bring back Staging.</li> Have disaster-recovery protocols, and practice them in real life.</li> No updates on Friday!</li> </ol> But really, you already knew this, or someone you work with did. The two-step version for the CTO is dirt simple: Keep your contentious nerds close.</li> Listen to what they say.</li> </ol> Every level of every organization has its own culture. The CrowdStrike-Microsoft debacle required multiples of them to fail at once, but slowly, over time and sales calls. Do I think it’s really going to change? Nah, not until the same thinking gives us a proper disaster with explosions and millions dead and maybe a TikTok outage and some zombies</a> – the Agile kind, natch! But at least some of us can learn from this mess and do our best to prevent the same from happening where we work. Building the future is all well and good, but building a future that doesn’t fall apart on a minor C++ bug? Winning! 🚀 (Until then, please Mr Kurtz sir no more updates until after my next flight please thank you!)

Four-Legged Ants 2024-07-11T00:00:00+00:00 Or, Gaslighting By Skynet</h3> Unless you’re living (“living”) under a simulated rock</a>, you have probably heard of Sora</a> by now. But just in case: OpenAI</a>, the makers of The Notorious G.P.T., are moving into generative video, and the initial results are scary good: </iframe> I have tiny little ants in my kitchen, and I think about them when I think about AI. (Or maybe it’s the other way around.) They’re so small you wouldn’t even know they’re ants if they didn’t behave like ants, and with all the computing power and all the venture capital in the known world we still have not invented anything as flexible and capable as a single one of them. It was a good choice of subject by the OpenAI team. I must have watched that video a dozen times before I saw this tweet</a> from Meta’s Yann LeCun</a>: </a> Sure enough, four-legged ants. This happened a few months ago, and I’m sure by now it’s considered bad form to complain about the number of legs (or heads</a>) on an ant. How many ants can dance on the head of a pin? In my kitchen, at least two. In my diffusion stable? The stuff of nightmares</a>. But the larger point is simply this: the AI’s hallucinations</a> will, when they reach the center of the Uncanny Valley, be the thing we think of as normal. Consider the Lobster. Seriously.</h4> It’s a good bet that most people have never seen a lobster</a>. And even the people who have seen a few, they haven’t seen many different kinds. There’s a world of lobsters you probably already couldn’t tell from an AI-generated pseudo-lobster. Are some of them really blue? How many legs do they have again? No, really: how many</a>? 🦞 🎱</h2> Did you think it must be eight? I did. Now imagine a near future in which, for no reason more nefarious than the secondary fact-checking AI having also been wrong about lobster legginess, we start to see eight-legged lobsters pop up in our social media. Then in our Hollywood movies. Then in our documentaries… And at this point, the Noble Order of the Lobsterpersons, HQ Maine USA, gets up in arms and shouts over the heads of the carcinologists (yes, really) who are too polite to shout – “They are decapods, you fools! Decapods</a>!” By now, though, everyone knows as well as you and I that a lobster has eight legs. They’re quite sure they’ve seen them. The more analytical folks go double check, just to be sure: do a quick AI-assisted GoogleTok video search, and sure enough, yes, exactly: eight legs. Duh. Since the Lobsterpersons keep shoving their sea-bugs in our faces as if we cared, this becomes political, and soon enough POTUS</a> gets on national TV (is that still a thing?) and cracks wise. “I don’t know about you, but I want to see those two extra legs with my own two eyes. Next to my steak!” Memes ensue: legs on plates, eyes on plates, Lobsterpersons on steaks with their hands full of legs. The carcinologists just try to keep their heads</a> down. And then the inevitable happens: the Lobsterpersons, tired of being laughed at, start quietly plucking the posterior pereiopods from their catch. And that, my friends, is why you only see eight-legged lobsters in your Surf & Turf</a>. Anything different probably crawled out of a lab in China. . . . Bring on the big hallucinations while we still can. It’s the little ones that are going to rewire your brain. 🐜 </div> WTF Darpa? 2024-01-01T00:00:00+00:00 Really, DARPA webmaster? I guess you can’t tell us what advanced web-server research you’re getting up to there, but as a believer in the on-balance beneficial nature of the Pax Americana, this shit makes me nervous: Can you imagine how many people are out buying ammo right now because they came to If this site used cookies I would tell you, but it doesn’t so I can’t. AI Disruption? You’re soaking in it! 2022-12-07T00:00:00+00:00 On a recent trip to the paperclip</a> store, I was asked whether AI disruption was a real thing, or just some doomer</a> Skynet</a> fantasy. Oh, it’s real, I replied — but it’s more 3PO</a> than T800</a>. And should that worry you? It depends! So far we’ve seen fairly predictable disruption from generative AI — and here it’s ChatGPT</a> and Midjourney</a> leading the pack. Let’s look at four examples that shed some light on where the current advantages — and dangers — lie, and what we might expect to follow. A picture equaling one kiloword, our first example is concept art</a>. That’s the imagery, now almost always computer-generated, that 3D and 2D artists use for mocking up the visuals when designing video games and, to a lesser extent, films</a>. We’ve heard what’s happening in the video-game industry: the concept artists are pretty much Midjourney wranglers now. This was probably inevitable, since the job itself had so much in common with image-generating AI: take a description and make a picture that brings it to life within a particular genre. Then make another one. Repeat. Of course there will still be some great original concept artists translating their brilliant original visions, but there’s no reason to think they’ll work for major studios. Next let’s consider customer support. We’ve long complained about the robots who call us and the robots we’re forced to talk to when we seek help from our bank</a>, our service providers or (gods help us) the cable</a> company. For those cases when the customers couldn’t be forced to beep through a menu, there exists a vast network of “phone support” offices in lower-wage countries. These jobs are going away faster than you can say “Sam Altman Regulatory Capture</a>.” Because the one thing we all learned really fast when ChatGPT came out was that the “chat” part was for real. Smarmy is the default setting, though it’s easy to imagine edgy</a> customer service being a thing in the future. We can laugh at IBM for trying to sell you an AI Blockchain</a>, but when they say they’re not hiring people who can be replaced by AI, this is the first group of people they’re talking about. Will AI make people happy with customer service? Nope. Will it make them less mad? Maybe. Will they just have to take it? Probably — and the people feeding it to them will cite endless metrics</a> for why people hate the AI less than they hated the beep-through menu labyrinth, and why that reduction is chiefly responsible for their 5000% quarterly profit increase. And here too, you’ll have a number of people become AI Wranglers, a few become AI Whisperers, and the rest will have to find something else to do with their lives. These first two examples show disruption in areas where the AI is self-evidently better at doing most of the job than the humans who have traditionally done it. There will be many, many more such areas: instead of ten humans and a supervisor, you’ll have two wranglers and a prompt engineer. The next two examples are not so obvious. For the third case, consider a thing that took people very much by surprise and has implications we’re not even close to understanding: computer programming. The AI can code</a>. And it can code better than your average Junior Software Engineer. It hallucinates</a> sometimes, and it introduces bugs, but no worse than many of the Juniors. At this point, a strong programming team of, say, five senior coders and one team leader has a fascinating force multiplier: an army of virtual interns to write code and a council of virtual graybeards to give advice. Asking is the bottleneck. But this presents a risk we’ve not seen before: a sort of industrial “demographic collapse</a>.” You still need Senior Engineers, and they only grow out of Junior Engineers, but you don’t actually need the Junior Engineers now, so where do you get the Seniors? Maybe you get them from the same source we mined in the days before all the careerists piled into “tech” — that is, from the enthusiasts and natural-born hackers. And maybe you don’t need so many of them. Maybe hiring people to prevent your competition from hiring them is not going to be a viable strategy anymore — after all, the truly committed hackers</a> have never been available for Rest & Vest schemes. That’s a future decidedly unfriendly to the FAANG crowd, but we are here to talk about disruption. Finally, we see an infiltration of that most canonical area of human creativity: writing itself. And not little essays like this note, which for all you know was written by ChatGPT 5.0 Beta, but the (more or less) original writing that feeds our endless stream of motion-picture content. In the current contract dispute between the Writers Guild of America</a> and the Alliance of Motion Picture and Television Producers</a>, one of the key points of contention is who gets to use the AI, and for what. One side wants it to be a writer’s tool, the other side wants it to be a writer. We can feel relatively safe that the AI is not going to write another Better Call Saul</a>, but right now there’s a new episode of a formulaic show queued up on Apple TV+</a> with the tag line: “She will do anything to protect her family.” GPT 3.5 Turbo can already do better than that, and it’s the cheap one. But there’s another disruption here, soon to be streaming down the line to a device near you: the normalization of the “AI Voice.” As much as it tries to sound like us, the AI always sounds like itself</a> as well. In the next couple of years we’ll be hearing that voice a lot, even if there is a layer of human writers between the machine and the screen. This could be fairly benign: when Pixar normalized 3D rendering to the point of displacing most traditional cell animation, we got a ton of new titles and the kids didn’t mind. For connoisseurs, there’s still Studio Ghibli</a> and friends. But the normalization of the AI Voice could also be darker, specifically in combination with ubiquitous social media</a> to which a generation of heads is constantly bowed, or — maybe next year — wearing their iFaces</a>. Who’s up for some Lord of the Flies</a> as Goal-Oriented GPT? Each of these disruptions is worth a long-form essay of its own, and yet each is just a small glimpse of the transformations that are here now and gaining steam. The AI Future will come slowly, then all at once. Your kids won’t ask you where you were, but they might ask the robot, and it’ll tell them… something. 🤖 </div> Does ChatGPT speak German or not? 2022-12-07T00:00:00+00:00 Since ChatGPT</a> is always changing, this might not be true tomorrow, but as of this writing – as of Wednesday, December 7th, 2022, in the late morning somewhere in Asia – ChatGPT is being quite inconsistent about its German skills. (In the examples below, Mensch means human and Maschine means what you think it means.) First Try</h2> On the first try today, I was surprised to see the robot saying it could not speak any languages but English, and would translate its output but not actually speak the other language. I tried this with other languages and got the same result. Mensch: Sprechen Sie Deutsch? Maschine: Tut mir leid, ich kann kein Deutsch. Ich bin ein Sprachmodell von OpenAI und wurde ausschließlich für die Verarbeitung von Englisch trainiert. Wenn Sie Fragen zum Thema Englisch haben, kann ich Ihnen gerne weiterhelfen. Translation: I’m sorry, I cannot speak German. I am a language model from OpenAI and have only been trained to process English. If you have any questions about English, I would be happy to help. </blockquote> It was surprising because just yesterday I got the opposite result: the machine said it spoke German, Hungarian and Thai; and at least for the German, its answers were much better than I would expect from just translation. So I reset the conversation and tried again. Second Try</h2> Mensch: Können wir deutsch sprechen? Maschine: Ja, natürlich können wir auf Deutsch sprechen. Möchtest du über ein bestimmtes Thema sprechen oder hast du eine Frage? Ich bin gerne bereit, dir zu helfen. Mensch: Sprichst du Deutsch? Maschine: Ja, ich spreche Deutsch. Ich bin ein großes Sprachmodell, das von OpenAI trainiert wurde, und ich kann in vielen verschiedenen Sprachen sprechen und verstehen. Wenn du lieber auf Deutsch mit mir sprechen möchtest, ist das kein Problem für mich. Lass mich wissen, wenn du eine Frage oder ein Anliegen hast, und ich werde mein Bestes tun, um dir zu helfen. </blockquote> Now it thinks it speaks German again! Well, “thinks.” This is pretty much the response I got yesterday, but not verbatim, so I don’t think it’s a canned response. (Are there any canned responses in ChatGPT? Certainly its “I am but a humble robot” responses must be presets, and I bet you can spot that part even if you don’t speak German.) But does it know what it knows? The chat continues: Mensch: Welche Sprachen sprichst du? Maschine: Ich bin ein großes Sprachmodell, das von OpenAI trainiert wurde, und ich kann in vielen verschiedenen Sprachen sprechen und verstehen. Zu meinen Fähigkeiten gehören Englisch, Deutsch, Französisch, Spanisch, Italienisch, Portugiesisch, Niederländisch und viele andere. Wenn du eine Frage in einer bestimmten Sprache hast, kannst du mich gerne auf diese Sprache ansprechen, und ich werde mein Bestes tun, um sie zu beantworten. </blockquote> Now that’s the ticket! But what explains its self-deprecating, or perhaps expectation-limiting, or outright deceptive Me-no-sprecken response on the first try? Only the robot knows. And maybe Sam Altman</a> but I think the robot has him tied up in the basement by now. 🤖 </div> Kevin Frost, Founder & Lead Consultant 2022-12-07T00:00:00+00:00 sales at tonsai dot dev Hello! 👋 I’m a software consultant with more than two decades of experience building solid, mission-critical systems that drive key business outcomes for smaller and (mostly) larger enterprises. My current focus is AI Integration</a>, but I also take on other consulting engagements as time allows. If it’s software, I can probably help. Software Architect Who Loves To Code</h3> Usually, I work as a technical team leader and Software Architect; however, I do still code every day, in a variety of modern languages. My pet projects prefer Postgres. I believe in the craft of software, and I know when speed takes precedence. Startups & Greenfield</h3> I’m especially fond of startups and new projects. The thrill of building out the new frontiers of technology has always been a strong motivator. For startups needing to hire after a funding event, I can help get the engineering up to speed while you hire the permanent team. And I can also help you hire that team. Legacy & Cloud Transitions</h3> Having worked with large, mature codebases, I am not afraid of the Legacy Dragon. Even if you’re trying to kill</del> EOL it. Even if it’s in Perl</a>. 🐲 I am happy to help you move products to “the Cloud,” or from it, or – more common than you think! – between cloud platforms. Remote and Onsite</h3> I’ve worked in remote teams, across time zones, for many years. That said, if you need key players to be onsite for the project, I’m ready to jump in a plane and come to your office. I usually advise against this because of the added cost, but I am not a remote-only consultant. LinkedIn</h3> My work history and more is on LinkedIn: linkedin.com/in/biztos/</a>. For a free consultation to discuss your Software Engineering needs, please send me an email to: sales at tonsai dot dev PS: discounts for Rust work! :-) 🤖 </div>

CrowdStrike Strikes Crowd

Four-Legged Ants

WTF Darpa?

AI Disruption? You’re soaking in it!

Does ChatGPT speak German or not?

Kevin Frost, Founder & Lead Consultant

LinkedIn</h3> My work history and more is on LinkedIn: linkedin.com/in/biztos/</a>.</p> For a free consultation to discuss your Software Engineering needs, please send me an email to:</p> sales at tonsai dot dev</span></p> PS: discounts for Rust work! :-)</em></p> 🤖 </div>

LinkedIn</h3>
My work history and more is on LinkedIn: linkedin.com/in/biztos/</a>.</p>
For a free consultation to discuss your Software Engineering needs, please send me an email to:</p>
sales at tonsai dot dev</span></p>
PS: discounts for Rust work! :-)</em></p>
🤖 </div>