We invented the internet specifically to survive a nuclear war via decentralisation, then spent the last twenty years cramming the whole thing into three cloud providers and calling it "progress."
Ukraine just reminded us in real-time why single points of failure are a bad idea, yet somehow, we’ll still act shocked the next time one outage takes down half the planet.
This is a subtoot about Microsoft.
The main question seems to be about our use of LLMs in our development process. Main concern is with liability and ownership with code.
@aeva They can’t:
1. Copilot uses the same connection infra that has died, and Microsoft has been directing staff to use their internal Copilot…. Sooooo that’s a ripperoni in pepperoni for them
2. The SSO connectors for OpenAI and Anthropic are down, so they can’t even auth those agents for use in the cloud environment.
Basically, the AI has abandoned them exactly when they needed it most. This is slapstick comedy of an epic scale.
Proved my old-school tech prowess today:
Asked an engineer to scrape a list of company email addresses off of a website we use internally. After 10 minutes, they were still refining an LLM prompt to do it.
I used devtools to grab the JSON data, copied it into Word, edited out the surrounding array structure, then copied that into Excel to sort alphabetically, deleted the numeric keys. Ta da! Took me about 3 minutes.
I really, really worry that basic computer literacy is nearly dead.
Weirdly, as we checked out of the hotel this morning all signs of the riot-adjacent activities last night except the graffiti had been tidied away. And it’s interesting to note the presence of one tonne sacks of short planks had been prepositioned near all the fires during a public transit strike.
Conclusion: the fires were preplanned, the police response was preplanned, and they were planned by the same people (hint: not the demonstrators).
cool we got a reverse url -> torrent index going. now we need a little crawler to extract url lists from wacz archives by just skimming the zip index and grabbing just enough pieces pull them out, then populating the index becomes automatic.
so coming soon to a #sciop near you, a p2p wayback machine, where a URL can get resolved into a bittorrent-backed web archive. then we federate the sciops. and we have the federated/p2p archive dot org we deserve.
If you have a Windows 11 machine, the update they just pushed sneaks Copilot onto your machine and sets it to run at startup. I just had to uninstall it (again) when I found it running in my system tray after the mandatory reboot.
Neat, the new node:net/promises API looks convenient! In general being able to for await a stream feels clean to me. Also, we have built in Websockets now?!
https://nodejsdesignpatterns.com/blog/whats-new-in-nodejs-27/
@mauve as someone who started online in the early 2000s: we had (semi)centralised spaces and connections born on-platform. But thinking of something like Livejournal, you didn't have a global timeline exposing you to everything anyone was posting.
In other places (thinking of Deviantart) where you had that exposure, the context was still present because the community was small (to today's standards) and very focused. I guess the same would apply to forums of the time.
First social networks I saw (Friendster, Orkut, MySpace) were weird to me because they didn't seem to have a point. No focus like "this is a community about a topic" or "this is place to blog", but just "this is a place to have a profile and friend other profiles".
A nice thing about the Fediverse is that it brings back, in some places, this "topical community" thing with servers, and a nice middle ground between small (server) and wide (fediverse)
This is kind of how fedi blocking works already (which I like), but that breaks down more with large instances that lose social context between users. IMO >1000 active users is when the context collapse muddies enough that individuals can't expect to trust everyone else.
local llm harness tinkering
I added "summary" fields to my edit/writ/shell tools in my harness and now I have short human readable messages as it's doing stuff which has been a nice middle ground between full auto and manual review. If the command is within the strict allow list, I just see the summary since I can trust that it's just find or sed or something.
Occult cyberpunk. Yap with me about decentralized systems, wearable computing, and biohacking.