kjuulh

I know, but if something is split up and stored in a vault somewhere the likely hood of a vulnerability napping the wallet falls. Defence in depth and risk analysis.

The name of the hacker is a bit on the nose lol.

I can't understand how you could let this happen. If they stored everything in fat hot wallets, that just seems like a recipe for disaster. Especially with the "features" of crypto. Math for the most part, doesn't care about your feelings, or money for that matter.

It might actually have been Libra. It was part of the whole get away from MS thing.

My mom still uses OpenOffice at her job (municipality), and honestly they haven't really felt a huge difference, they still get their work done like before, but heck there were some training to do as they were switching. I suspect at this point they would need the same training just to get back to word etc.

I think it is an interesting question of what makes someone effective at using AI, which changes in what context you're in.

If you're at a small start-up and effective ai user looks different than one at an enterprise company, or hyper-scaler with infinite resources.

For me being effective at using AI, is as much the harness and workflows I built up alongside the services to support it, good ci, good tests, deployment pipelines and more. This allows us to streamline development and let the AI agents work much more autonomously, and in some cases even do automatic development and triaging.

Does that put me in 0.01% of spend, nope, but it makes me effective at my company and for personal development.

That looked like a really fun project, and quite the persistence you've got to actually stick with it for such a long time. I might have to give it a try when i get home.

Docker and containers are amazing. They definitely have their caveats, whether that is security, bloat, you name it. But the fact that I can produce an artifact and have it run pretty much anywhere without much overhead is amazing.

For me firecracker has mostly taken over most of my hosting needs, as i can actually take a docker image and unpack it into a firecracker image, and still have a slim process (vm), with great performance, though with a bit extra bloat, but now also have a secure workload that I've got better control over.

Docker and docker compose however, is still the core of my development workflow, and has been for many years, in my eyes, docker compose is the best we've got for simple composition of various services.

If you didn't know of an embargo, and you posted about it, I can't see how you were wrong.

This getting out before the fixes were published sucked, but they would've sucked anyways.

This getting out at least allowed me to inform some of my previous colleagues that they had services that were affected and they were able to shut them down until we had a proper fix available.

I feel like in the beginning it was a way to highlight the fidelity of the models as they struggled to produce good latex and svgs figures. But especially now with the iterative loops that is no longer such a good benchmark. And honestly these pelicans are pretty much the same, so yeah it feels totally benchmaxxed.

We ran actions runners ourselves in kubernetes at previous jobs, and at least back then, a lot of the errors came from the github services simply not telling the runners to handle jobs. So in general it didn't matter how much compute you had, they never got scheduled jobs.

It feels more like volume than speed. AI has actually improved our speed quite a bit, as it is fairly easy to just get AI to make it fast. Will vibe coders do this, probably not. That said our volume is probably 10-30x in commits and ci jobs as we had pre-march.

Great read. I always find these turn a feature of a computer into a database an interesting read/idea. It certainly has upside, and downsides. For example having an entire filesystem as a sqlite database etc, or in this case an executable. It would be nice to have a structured way of talking to our utilities, rather than interpreting various commands as json through a few layers of tools to then get the output in a common format. That said I'd probably rather have a split up tool-set, than everything having to be in a single database for it to interoperate.

Reading this port-mortem / plan shocks me, this doesn't look like a service that has been serving high-throughput services for more than a decade. In fact it is almost like they've barely started. It seems the solution has been capacity, capacity rather than architectural or data changes.

> Our next milestone is an architecture that scales read capacity linearly with the number of readers, enabling unlimited read operations

How do you not have read-replicas / read caches at this scale yet? Which is what I am reading from this statement. You can of course get really far with sharding and whatnot. But at some point it might become worth it to engineer your data into a model that scales better.

Great article. I find that I learn something every time I read a post about linux kernel work.

I guess an LRU with priority would handle VRAM for games pretty decently without going getting too application specific.

What about VRAM to Disk specifically NVME, would direct to disk be feasible for large workloads, I know it is used for streaming in assets directly via. PCIE, but i wonder how the performance would be on compute workloads running with NVME as a swap for GPU VRAM.