buc.ci is a Fediverse instance that uses the ActivityPub protocol. In other words, users at this host can communicate with people that use software like Mastodon, Pleroma, Friendica, etc. all around the world.
This server runs the snac software and there is no automatic sign-up process.
This is the sort of thing I'm referring to with https://buc.ci/abucci/p/1794739419.625208
The "How complex systems fail" tretise was written in 1998 by someone with experience in system failures, but whose views are by no means the last word on the subject and were not even at the time it was written. Also, I suspect the austerity economics we've all been suffering under since that time has led to significant degradations compared to the state of things in 1998.
There is a certain irreducible complexity in socio-technical systems, but I think in reality we are far away from having systems that are so well engineered that most failures are attributable to it. Human accountability for deploying potentially dangerous systems that are too complex to analyze matters regardless.
#dev #tech #software #engineering #complexity #ComplexSystems
Nathan should share the core ideas in his prompts as aphorisms.
While that's often true of complex systems, I think this misses the point, dangerously. We can and frequently do engineer out this kind of complexity. That's what engineering is and is for. The failure in these instances is that this appears not to have been done, or not done adequately.
Socio-technological systems have existed for millenia. Yes, things are more complicated now. But that does not justify throwing up our hands and saying "welp, complex systems, no way to know". There are ways to know: you engineer the system such that when something goes wrong there is an analysis to be done that reveals causes. If that analysis is not possible, you're not finished engineering. Deploying a poorly engineered system is a fault traceable to individuals, and they ought to be accountable.
We are today awash in systems that have not been engineered properly, software systems especially. Don't let that blind you or lead you to fatalism.
Columnists who presented Huang as the AI hype industry’s sanity backstop have reverse-ferreted into pontificating about systemic governance failures.I suspect we'll be seeing a lot of reverse ferrets.
#Microsoft #cloud #infosec #internet #Office365 #MicrosoftOutage
#Microsoft #cloud #infosec #internet #Office365 #MicrosoftOutage
The real pisser is that in computer science/computer engineering we've developed lots of techniques for avoiding the pitfalls that arise when a system evolves "organically". "Evolved organically" was frequently used pejoratively or accusingly. I guess when your company is valued over a trillion dollars those ideas seem quaint and passé.
In Latin, invidia is the sense of envy, a "looking upon" associated with the evil eye, from invidere, "to look against, to look in a hostile manner." Invidia ("Envy") is one of the Seven Deadly Sins in Christian belief.https://en.wikipedia.org/wiki/Invidia
Open source does not protect you from slopification. We are experiencing the folly of that belief with countless open source projects right now. It is infeasible to fork a complex piece of software and take up developing it on your own if the corporate masters go in a direction you don't like. Therefore by committing to Typst one takes on a risk, and personally my tolerance for this shape of risk is very low. YMMV
Make sure that you prioritize tools that are resource-friendly, locally hosted, open source (or at least open weight), were created under ethically sound conditions, for the public interest, preserve data privacy and ensure transparency and reproducibility.In my estimation, when it comes to generative AI there is no such tool that would be known to non-specialists. I suspect only a bespoke tool could satisfy all of these criteria in a meaningful way. For this reason I think you need to list examples of the tools you have in mind.
The "open weight" "local" models are largely coming out of the AI labs that are making hackable hardware inaccessible to many peopleis what I said originally and wish I'd stuck to. China's industrial policy, which has also given rise to open weight AI models, is at best unconcerned with widely-available hackable hardware, but nevertheless results in a restriction of hardware availability. I don't know enough about this aspect of the industry to comment, but I can say with some confidence that the US hyperscalers in particular are actively militating against such access (and I suspect China's large companies will follow suit if they have not already because this is how the economic model works).