Notices by Open Risk (openrisk@mastodon.social)

@petergleick

The towers of falsehood and abuse are built one layer at a time.

Before "Artificial Intelligence" there was Machine "Learning" of which Large Language Models are just an instance.

Nobody ever questioned if fitting a non-linear model to data actually "learns" anything.

Its exactly what linear regression does. Nobody worries about straight lines running through data and interpolating or extrapolating new points becoming sentient.

We truly live in an age of artificial stupidity.

@dansup @pixelfed @loops Its a challenging task. Normal people using social media have been addicted and the adtech platforms have found all our weaknesses by running large scale social experiments. How easy is it to get people off junk food and into healthy food?

The only advantage of the fediverse is decentralization. Maybe this could be used to better support localized communities. People like their own places, villages, cities. But you'd need the nerds running servers 😉

@HauntedOwlbear @nilesh

To me its all still quite confusing. Activitypub seems quite a bit less prescriptive, a lot of important details are "left to implementation". The Mastodon server network is one such fully developed architecture but quite different designs are compliant, e.g., single account "instances" like gotosocial. Somebody even suggested that the entire activitypub pattern could be implemented over the atproto protocol etc.

"When a measure becomes a target, it ceases to be a good measure" (Goodhart’s Law, alternatively: "Any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes")

Inevitably #xkcd asks whether there is a metric to identify vulnerable metrics 🙂

#openriskmanual

https://www.openriskmanual.org/wiki/Goodhart%E2%80%99s_Law

https://xkcd.com/2899/

@Gargron There are similarities in people's lemming behavior but also important substance differences that are good for the #opensource community to keep in mind.

Blockchain architectures typically target problems that could be solved with slightly less decentralized / automated designs if people really wanted to.

In contrast, "AI" is a class of useful algorithms (LLM) cleverly packaged to be seen as something magical. But they already solve something fairly practical, its not 100% hype.