Scalable storage, add an NVMe disk when you run out. Do that at least once more for redundancy. You'll save a crap ton of cash in the long run. (10/10)

If you're actually serious about saving money in the LONG run (for some up-front investment), consider actually building out a private cloud. Go rent a cabinet, shove some high density storage in there, and boom. (9/10)

a single-threaded non-parallelizable agent on an HPC cluster because you think it'll "go faster"--it's a waste of your time.)

In the SHORT term, your best bet is to go with a bare metal solution, OVH or something. But keep in mind that they don't scale either! They're just dirt cheap. (8/10)

Containerize your binaries, fine--library and upgrade management is an entirely different discussion--but don't pretend your k8s cluster is going to magically make your nodes scalable or your data easily transferrable. That's like trying to shove (7/10)

"oh, but AWS is a BIG NAME and that makes them TRUSTWORTHY" -- my tip to the blockchain CEOs out there: those folks, immediately fire them. They know nothing. (These are the same folks that insist that shoving blockchain nodes in k8s clusters are worthwhile... NO. (6/10)

I've known companies burn tens of thousands of dollars a month on simple blockchain setups. Really dumb. Those companies, they either don't exist anymore or they've gone through some really big reboots. Most of time time when you advise against it they go (5/10)

to sell you their overly-expensive EBS (or otherwise) to host it, and don't get me started on your ingress/egress costs. Validation is, in most cases, not a viable long-term financial solution in isolation... so be smarter with your cash. (4/10)

in which case, yeah, you've possibly undermined the point of the blockchain in general, philosophically speaking...) you're going to be loading up tons of immutable data with nowhere to put it. And the Big 3 are all-too-happy (3/10)

horizontal scaling at the blockchain level is NOT what I'm annoyed with here. I'm talking about the actual nodes physically running on the machines.) Unless you're cheating at validation (e.g. running fully-pruned nodes without regard to the block history... (2/10)

when leaving college and having the tendency to use ChatGPT just like they do when copypasta'ing from StackOverflow, but I'll maybe punt that for now.) (7/7)

ought to avoid using it for entire workflows at this point in time, just trusting that our prompts are going to yield a fully-functional codebase without any amount of understanding on our side. (There's also a rant here somewhere regarding students generally being unprepared for the workforce (6/7)

automatic brakes, things like that? Fantastic! And I think that's the line to draw here... should AI be used to replace the creative work that folks do? No. But should it be used to assist, e.g. code review, inspiration? Absolutely. AI can be great for small, discrete work. But ethically we (5/7)

That cannot be piloted manually, say with some of those taxi services that are available in some cities, do we really want to pit large companies against the little man in terms of claims when wrecks occur? No way! But, using vehicle automation for driver assistance e.g. automated parking or (4/7)

I kind of think of it like I do autonomous vehicles. And that is, I'm not very comfortable with fully-autonomous driving. The operator is expected to be attentive but that's not very effective for keeping folks' attention at the wheel. And in the event that we gain fully-autonomous vehicles... (3/7)

the nature of what I do but I'd be hard pressed to fly by the seat of ChatGPT's pants. But maybe I'm old. I won't deny that having a script automatically written for inspiration, or for debugging purposes, has been very helpful at times. So what's the limit as we continue the march toward GAI? (2/7)

...instead of relying only on those factors that rely on the end user (your MOST vulnerable asset in any company, practically always, even with training) to keep your infra safe. Always assume your folks are compromised. Hence, my love for Swiss cheese. (4/4)

...and vendor diversity is critical. I mean, even in this case we're talking MFA-- which has diversity built into the concept. So MS failed to account for ratelimiting in their TOTP slippage? Cool, maybe combat that as the end org by implementing "where you are" and "how you behave" factors... (3/4)

Okay. So that's a technical control that failed or something. But seems like risk introduced by inclusion of Big Vendor is increasing by the day. Botched MS updates, CrowdStrike snafu, Cloudflare misconfigurations, blah blah blah point is, paying serious attention to defense in depth... (2/4)