aes
Sounds reasonable, but a lot of recent advances come from being able to let the machine train against itself, or a twin / opponent without human involvement.
As an example of just running the thing itself, consider a neural network given the objective of re-creating its input with a narrow layer in the middle. This forces a narrower description (eg age/sex/race/facing left or right/whatever) of the feature space.
Another is GAN, where you run fake vs spot-the-fake until it gets good.
Well, Rust has a lot of string flavors, and I like utf-8 being the norm, but there are a bunch of cases where enforcing utf-8 is a nuisance, so getting string features without the aggro enforcement is nice.
There’s probably some fruity way to make this a security issue, but I care about ascii printables and not caring about anything else. This is a nice trade off: the technical parts are en-US utf-8, the rest is very liberal.
Reminds me of Scarfolk
Rust implementation of the Meta-II meta compiler. I used bstr, which was interesting.
Backstage has become quite misaligned to what we were originally trying to do. Originally, we were trying to inventory and map the service eco-system, to deal with a few concrete problems. For example, when developing new things, you had to go through the village elders and the grape vine to find out what everyone else was doing. Another serious problem was not knowing / forgetting that we had some tool that would’ve been very useful when the on-call pager went off at fuck you dark thirty.
A reason we could build that map in System-Z (the predecessor of Backstage) is that our (sort of) HTTP/2 had a feature to tell us who had called methods on a service. (you could get the same from munging access logs, if you have them)
Anyway, the key features were that you could see what services your service was calling, who was calling you, and how those other systems were doing, and that you could see all the tools (e.g. build, logs, monitoring) your service was connected to. (for the ops / on-call use case)
A lot of those tool integrations were just links to “blahchat/#team”, “themonitoring/theservice?alerts=all” or whatever, to hotlink directly into the right place.
It was built on an opt-in philosophy, where “blahchat/#team” was the default, but if (you’re John-John and) you insist that the channel for ALF has to be #melmac, you can have that, but you have to add it yourself.
More recently, I’ve seen swagger/openapi used to great effect. I still want the map of who’s calling who and I strongly recommend mechanicanizing how that’s made. (extract it from logs or something, don’t rely on hand-drawn maps) I want to like C4, but I haven’t managed to get any use out of it. Just throw it in graphviz dot-file.
Oh, one trick that’s useful there: local maps. For each service S, get the list of everything that connects to it. Make a subset graph of those services, but make sure to include the other connections between those, the ones that don’t involve S. (“oh, so that’s why…”)
Could ‘push’, yes, as in, “we mentioned it in passing when rock and roll grandpa wasn’t paying attention, so he wouldn’t throw a hissy fit and withdraw from the service”. Oh, you meant to the labels? Ha ha ha, NO. The labels have basically nuclear option veto powers.
As for changes, well, updates get delivered all the time, for various reasons. (The scratched Turbonegro album being one frequent flyer.) I think a lot of those are bullshit SEO-like reasons, but it is what it is.
Which artist appears in most frequent releases? I forget, but I think it’s Elvis. Possibly Johnny Cash. Why? Because some material has gone out of copyright in some jurisdictions, and so people have the idea to upload them again in ‘new’ compilations. (The content team don’t even beat these down personally – that’s machine work)
