Notices by Matthew Skala (mattskala@mstdn.io)

@lain AGI is not like how I imagined.

@lain necropants(u) https://en.wikipedia.org/wiki/Necropants

@lain It's good that they explained what "telekinetic" means.

@lain Yeah, and I don't think the jury stipend will cover feeding it. Maybe if I start immediately I can train the whale to boop buttons labelled "guilty" and "not guilty" with its nose.

Today I learned that owning a whale without a license, although illegal, does not render you ineligible for jury service in Ontario.

@lain That's hot enough it may even *melt* the pipes.

@lain ...but yes, ChatGPT and only ChatGPT killed Stack Overflow.

@lain Well, they're vaguely rabbit-like, aren't they? I think rabbits are known for having that whole business figured out.

@lain That was basically the punchline, underneath all the philosophy, of Ghost in the Shell.

@lain Sounds like they're already well on their way.

@lain Not if the Dems revive their taco truck on every corner promise...

@lain Does it taste like file folders?

@lain After a bad experience with Air Canada on my first trip to Japan, I flew on Korean Air for subsequent trips and I liked it a lot. That was pre-scandal.

@lain Needs more toy dinosaur.

This week's Matthew Explains video (viewable without login, as a treat) covers a very recent psychology preprint rather than the usual computer science.

Experimental support for the idea that opposition to AI is, for many people, *moral* rather than consequence-based. Many really see AI as bad *just because it is AI,* not because of purported consequences like "It uses lots of water!"; even if people will use consequences as rationalizations for their moral positions.

https://matthewexplains.com/11190144/

@lain And apparently a detachment of tiny space aliens flying their saucer in through the window in the background, presumably bearing gifts.

@SuperDicq @phnt You might be interested in what I've written on the subject of *really* free models: https://elevenfreedoms.org/ .

I think that "open" training data in the sense of "download it and reproduce the training process" is the wrong way to think about openness of models, not least because *the training data is not the source code* in the sense of preferred form for modification. But we do definitely need more openness and freedom, and I spell out the freedoms we need.

@lain Well, that depends - which one am I?

@lain "Imouto *was going to* start it!" never worked on Mom, so if he can get it to work, much respect.

Offline most of today. Failed Internet. And the cable company tries to route all technical support to a chatbot which, guess what, only works when you have a working Internet connection. So you get through in one of the rare moments of connectivity, and the chatbot says, oh, we tested your connection and it's working now, can I offer you a pre-approved Mastercard? And then the connectivity dies again.