No shit, if you train a stochastic parrot on biased data, you get biased output?

[aud@fire.asta.lgbt]

@[email protected] RIGHT?!

"tech bros give a shit about anyone except themselves" challenge level: impossible

[xgranade@wandering.shop]

@aud Like, account systems have no business assigning anything other than a fucking UUID to individual accounts, then at *most* using stuff like OAuth claims as evidence of names.

[xgranade@wandering.shop]

@aud That is, everything should be more or less of the form "this technical system believes that the person who wrote commit abcdef123 is the same person who claimed the name X at time Y because of evidence Z."

[aud@fire.asta.lgbt]

@[email protected] THANK

YOU

do you bet your ass that's how I'm assigning IDs? yes. yes that is. because everything else can change ffs.

[aud@fire.asta.lgbt]

@[email protected] (obviously some extra challenges with federated stuff but nothing impossible to work around while maintaining human readability)

[xgranade@wandering.shop]

@aud Now I'm wondering if there's something like RDF / Notation3 where every triple links to evidence of that claim.

E.g.: <person with UUID X> <is> <url to external account> because <proof>, where because is a built-in part of the grammar and not a separate URN-ed verb.

[aud@fire.asta.lgbt]

@[email protected] well shit, I hadn't heard of this stuff before. This is cool!

[xgranade@wandering.shop]

@aud Notation3 is honestly very, very cool imho. It never really took off because of more general backlash against Semantic Web concepts, but it's a fun way of building an ontology for some knowledge base.

The trick, though, is how to extend that to an epistemology over the same ontology. By default, N3 recognizes the verb "to be" as primary, but that isn't great for epistemology...

[xgranade@wandering.shop]

@aud The Wikipedia article on blank nodes has some great examples of how you can use N3 ideas to represent "there exists a person such that I know the following list of facts about them" without needing any concrete ID for that person.

```
_:entity
ex:fullName "Alice Carol" ;
ex:homePage <http://example.net/alice-carol> ;
ex:hasAddress _:address .
```

[xgranade@wandering.shop]

@aud That carries the same issues with names, but if you had something like `_:entity is <https://wandering.shop/@xgranade>` and used WebFinger to look up the external URL to build up more tuples like `<https://wandering.shop/@xgranade> displayName "Xandra Granade"`, then by transitivity of `is`, you could imply the properties you want.

That's basically a very rudimentary theorem solver with external lookups.

[aud@fire.asta.lgbt]

@[email protected] ooooooooooooooh, that's... very interesting.

There's a part of me that would love to enable public-private key auth to optionally enable people to be like "yes, this is me" for reasons, except then they'd have to store the private key and that can be tough across multiple devices. And passkeys... I am not going to look into passkeys. Not unless someone literally pays me I get the feeling that's a corporate mess.

[aud@fire.asta.lgbt]

@[email protected] shiiiiiiit that is... alright, I have to play with these ideas.