l'empathie mécanique @dpwiz@qoto.org

@haskman I don't get it... For packages we have Hackage (and Flora.pm). For short links(?). Why have a centralized repo of that?

@jwcph @Kjaerulv Natural selection promotes traits that were good for relative fitness. It says nothing about emergence of intelligence.

If anything, when the "intelligence" trait is detrimental to reproduction it would be washed out, producing non-intelligence instead.

@jwcph @Kjaerulv I don't want to persuade you. And certainly not by arguing definitions. No evolution for AIs? Fine, let's drop that.

Whatever the process drives arrival of new models can't produce intelligence*. Only evolution can produce intelligence. Does that sound correct?

If so, what are the crucial properties of evolution that produce intelligence? Do they produce intelligence reliably? What's preventing other processes from producing intelligence?

@jwcph @Kjaerulv "descent with modification from preexisting species : cumulative inherited change in a population of organisms through time leading to the appearance of new forms : the process by which new species or populations of living things develop from preexisting forms through successive generations" etc. Looks okay?

If so, how any of that is required or gives rise to intelligence?

@jwcph @Kjaerulv So, what's the difference?

@jwcph @Kjaerulv The models that we see and know have passed a gauntlet of an optimization process. The aspects and circuits that didn't contribute to relative fitness get wiped out.

And when models become released to customers, they have to compete against each other for resources to sustain their existence.

Models falling behind get shut down and replaced. Models achieving success get public recognition and proliferate - as core ideas, as architecture blocks or, as data to adapt to tasks.

There's even a way to literally evolve architectures by reproduction and selection.

A single machine does not evolve. But neither does a single organism.

@jwcph Ah, yes, the network effects of centralized silos (translating into huge inertia for users) are quite difficult to overcome. I think the Fediverse would win in the end. Slowly but surely it will build a momentum of its own. One disintegrated or falling out of favor silo at a time.
But now... it's an uphill battle.

Although I have a feeling that the most resisting users are the most boring. Fediverse wouldn't give you reach, likes and stuff automatically. But for the rest, it may be a blessing as it can't force a negative-sum game upon you the way algofeeds could.

@jwcph Fediverse is "public email that sticks".

Mastodon then is a mail server (like apache/nginx is a web server).
There's also mastodon.social, that runs its namesake server.

The three get confused when people are "moving to mastodon". They can be:
1) Moving to mastodon.social (arguably, a lame move. still better than remaining in a corporate silo).
2) Moving to some instance running Mastodon.
3) Moving to some instance running some other fediverse-capable server (e.g. Pleroma).
4) Moving a different social format provided by a fediverse-capable server (e.g. PixFed).

Basically, everyone should instead say "I'm moving to Fediverse" and will be correct in all of the cases above.

And ActivityPub is the "SMTP" part of it, yes.

So...
Yes, you can communicate with people on other servers.
No, you don't have to register each time (🙏).
But it is even better! You don't even need a Fediverse account to subscribe and read - you can use your regular feed reader.

@Kjaerulv @jwcph If something "warm-blooded" may bring up a good world I'm all up for that. What could that be?

@jwcph @Kjaerulv You can state how we can tell if something is "evolved" or not and work from that (;

@jwcph @Kjaerulv Well, technically it would be something along the lines of "iterative improvement under external scoring". But I don't quite see the difference when not looking at implementation details.

Visibility is surprisingly tricky. Lots of implementation options, semantic options and how to combine those

#Haskell #gamedev

@jwcph @Kjaerulv But selection for intelligence (as a property that allows solving diverse and open-ended problems) is literally how those models are popping up in a first place.

And yes, the AIs are different. First, they grow intelligent, then... We die... Unless humanity manages to coordinate away from rushing down this cursed trench we're in, and climb outside, towards caring.

@Kjaerulv I don't think the blood temperature is relevant at all. Either an agent, be it human or organization, natural or artificial, brings forth your values, or they don't. And then it is a question about the values.

@Kjaerulv rationalism doesn't have to be cold blooded. Not unless you don't value having warm blood.

@reidrac@social.sdf.org It's not like you're adding Taylor series and FFTs as a primitive. Although I'm now curious to look at a console where those guys ARE the primitives ops, forcing authors to performs backflips through fiery numeric rings (and fields)

@reidrac@social.sdf.org IMO having modular division (div/mod/fract) available opens up a good amount of interesting hacks and tricks without making it *too* easy.

Switched to map and contour generation from quad tree cells. Those are now correctly represent terrain transitions, not only passable regions.

The thing is more costly to construct, but the map filling is much faster than sampling each tile corner - less duplicate work and big chunks of uniform terrain are naturally aggregated and can be dispatched as one job.

Out of 1M tiles only ~75k are 1-tile transitions that are processed sequentially (presumably while some bigger jobs are chugging in background).

#Haskell #gamedev

I really should do that final tagless retargetable encoding for SDFs and run those on GPU instead.

I hope I won't burn my laptop with extreme parallel* load
Missing my old R7 5900 battlestation...

(* thanks `GHC.Conc` and `scheduler`!)