Notices by Urusan (urusan@fosstodon.org), page 2

This is an interesting video:
https://youtu.be/dDUC-LqVrPU

TL;DW We're starting to see early evidence of diminishing returns with our current AI architectures. If this is true, then eventually they start to improve in a logarithmic manner, making superintelligence (at least using our current architectures) impossible to achieve from a practical standpoint. The issue is that we need too much data on specific things for it to perform well on them all.

There are also more options than just simple allow/deny filtering, especially with the details provided by the AI system.

For instance, it could auto-CW posts (including those from foreign instances without the same CW rules) rather than doing simple allow/deny filtering.

This would be especially useful at the individual level since it would make it easier to set up consistent filtering based on CWs.

This idea really isn't that dissimilar to a spam filter (you even train spam filters on spam examples).

The main difference would be in detail and flexibility. Spam filters are great at detecting immediate, obvious abuse. That's great, and that's still necessary to help deal with large scale attacks.

What about more subtle patterns of abuse though? Or rules specific to an individual/instance/community? If you put this stuff in a spam filter you would filter too much.

Over on my AI-focused account, there was some discussion about AI moderation on Fedi.

The concept is to use the AI system to categorize posts according to some classification system that instance admins can easily tailor, which makes moderation much easier.

This might be actually beneficial for a relatively decentralized network like Fedi. By reducing the cost of moderation, it would mean more small and single user instances, which might be worth the downsides.

@io Here's an example from Car Mechanic 2021. You point at the thing you want to interact with, press the interaction button and it brings up the circular dial menu with your contextual actions. On a controller you can press your thumbstick in the direction of the option you want and confirm to do the action. It works with the mouse as well, you can nudge it in the right direction slightly before clicking.

So here we are in the context of a car on an auction lot and we want to examine it first.

@io One more crucial detail about these dial menus is that they easily allow the options to change, and it's easy to stack them (one radial dial option replaces the current one with a more specific set of options) to provide a hierarchy of contextual actions. They're pretty slick, which is why I think they'll become more prevalent over time.

If you have more or fewer options at any given time, then you just add more subdivisions. I don't know the details but it's not jarring when they change.

I'm pretty sure radial dials are going to be a standard UI element in the next 10-30 years.

They're used heavily in video games to select contextual actions, so my kid's generation will probably be really used to them for this purpose by the time they're adults, and they'll bring them into other contexts as they become designers and engineers.

I'm really annoyed by the recent rhetoric that AI training is the worst intellectual (property) theft in history when the enclosure of the 20th century's public domain under effectively perpetual copyright is in fact the worst such theft in history.

I'm against copyright entirely, but if it must exist then it should last at most 14 years, not a century-ish.

If we had reasonable copyright terms (even 30 years!) there would be no lack of public domain training data for AI.

@tk I prefer the lead free ones for health reasons.

Mmm, tasty.

What system have you used for years and never learned how it works?

I get the impression that a major difficulty faced by open source projects with a major frontend component is that most developers find the work difficult and unfulfilling (at least, after the excitement of a new project wears off), so finding long term maintainers is difficult.

What does Fedi think?

@alex One of these days, this will be an incredible archeological find.

Is there a good, relatively high end ARM or RISC-V laptop available that I could install Linux on?

I'm thinking about something along the lines of the recent ARM-based MacBooks, but not Apple.

@icedquinn Between this and the unexpected effectiveness of much simpler ground to ground missiles (when used properly), it will lead to an overall shift away from large surface vessels. This is a lot like the obsolescence of battleships after WW2.

Sevastopol is useful pretty much entirely for large vessels, so this shift means it's now strategically unimportant, even if I'm sure many countries will continue to act like it is still strategically important.

@icedquinn Well, one of the most interesting developments is the sea drone. It's basically a remote control speedboat bomb, or an extreme range "cruise missile" torpedo.

It's low tech and low cost enough that basically any state worldwide should be able to develop and deploy them without support by the great powers.

They make large ships too vulnerable to get anywhere near an active conflict, which renders most large surface vessels useless.

https://youtu.be/Y6UeNBj9rrU

Russia's invasion of Ukraine has seemingly been almost entirely driven by resources that were useful in the past but are now useless.

The natural gas fields in eastern Ukraine that we need to stop using to avoid climate catastrophe, and now the naval port in Sevastopol which the war itself has proven to be obsolete.

The linked video details how the naval situation has changed, with most large surface ships probably being useless.

@Moon Will you be updating your shibboleth test sound occasionally as time passes and younger people become adults?

Rust libraries currently don't exist:
https://youtu.be/769VqNup21Q

Rust can make C libraries though, they're just unsafe and have other inconvenient issues.

Here's a quick rundown on why our two political parties support Israel:

Democratic representatives: they believe that they will make a lot of money selling arms to Israel

Republican representatives: they believe that Israel committing genocide will initiate the apocalypse which, after years of brutal worldwide suffering, will culminate in every human not in their favored in-group being cast into a lake of fire AND that they will make a lot of money selling arms to Israel

@puniko I'm doing some practice rounds and this happened: