Mullvad has a switch that lets you access local IP addresses, but locally resolved hostnames still do not work

people who know about network stuff: is there any way of making local dns resolution (e.g.: my NAS) work when connected to a VPN (specifically Mullvad)

probably bad idea: given that the public key based vault sharing feature in most password managers is a weak point and potential back door, replace this with vaults encrypted to single user passwords plus a replication protocol on top of signal messages

have just published 5 security advisories relating to 3 distinct issues in the node.js and ruby versions of websocket-driver:

node:

- https://github.com/faye/websocket-driver-node/security/advisories/GHSA-xv26-6w52-cph6
- https://github.com/faye/websocket-driver-node/security/advisories/GHSA-mp7j-qc5w-4988

ruby:

- https://github.com/faye/websocket-driver-ruby/security/advisories/GHSA-ghhp-3qvg-889p
- https://github.com/faye/websocket-driver-ruby/security/advisories/GHSA-33ph-fccm-39pj
- https://github.com/faye/websocket-driver-ruby/security/advisories/GHSA-8j3g-f24p-4mpw

have just published 5 security advisories relating to 3 distinct issues in the node.js and ruby versions of websocket-driver:

node:

- https://github.com/faye/websocket-driver-node/security/advisories/GHSA-xv26-6w52-cph6
- https://github.com/faye/websocket-driver-node/security/advisories/GHSA-mp7j-qc5w-4988

ruby:

- https://github.com/faye/websocket-driver-ruby/security/advisories/GHSA-ghhp-3qvg-889p
- https://github.com/faye/websocket-driver-ruby/security/advisories/GHSA-33ph-fccm-39pj
- https://github.com/faye/websocket-driver-ruby/security/advisories/GHSA-8j3g-f24p-4mpw

is this how I accidentally reinvent either observability or functional programming or both? let's find out! (later when I don't have heat madness)

ok to make this actually useful I am going to need to make it so I can reproduce failures that occur at a rate of about once an hour, i.e. make it so I can record all the randomness of the cryptography and latency and replay it deterministically

oh no

me writing escodb: wow it's cool how I've constructed it so I can run the concurrency tests and easily wrap the storage layer in a proxy that adds random latency, neat

me running the tests: I've made a huge mistake

RE: https://mastodon.social/@jcoglan/116676362605359342

btw this and my previous "they want the hawking radiation without the black hole" bit aren't metaphors. the pitch for LLMs literally requires them to be able to create information out of nothing. it is physically impossible that they work as advertised

I'm hearing the perpetual motion machine costs a lot to run? huh

you know what else can turn poorly defined specs into poorly written programs for the cost of a software developer salary

if only there was a way of telling a computer what to do that didn't require paying data centre landlords four figures a month

also writing good tests is incredibly hard, and tests exist in a feedback loop with the implementation. the set of tests you need to constrain a program to the behaviour you want depends on how the program is written and vice versa. a lot of people have been taught that tests are empty boilerplate and thus really undervalue test writing as an analytical skill

RE: https://infosec.exchange/@david_chisnall/116668252539246029

this is a very good expression of why LLMs cannot write good tests, and why software implementations cannot be derived from tests. tests are the result of an author distilling the categories of behaviour of a program into a small number of examples that demonstrate key differences between cases. they are not the categories themselves, they are the "for example" part of your explanation of a concept

this tech has people devaluing their own work to the point they're willing to actively destroy it. big tech has created a machine that makes everyone else tear down everything they themselves built

I've been saying that LLMs are machines that destroy information and I wasn't quite ready for the implication that this means destroying software that was working fine that you have relied on for decades

copying and synchronizing data is one of those things that sounds simple but really isn't, and rsync is historically very good at it. it seems baffling that someone with the necessary knowledge to create rsync would willingly chuck it through the slop grinder

the prospect of having to replace rsync with something else, considering how much trust I currently put in rsync to not lose my data, is not a fun one

RE: https://mastodon.social/@ZachWeinersmith/116584276117791530

would actually place my understanding of "AI does not work" even further to the left of column 1 because part of how it does not work is that it has been hyped as being applicable to absolutely any information processing task, which is more harmful than if it were merely bad at a clearly delineated problem

it does also namecheck inscryption and I *think* void stranger but I might be wrong about that (these references are by inclusion of visual items and the one I think refers to void stranger is quite ambiguous)