iiinova — where your digital self keeps evolving

i joke but it genuinely really hurts to watch the game that in many ways saved my life a decade ago turn into a case study for everything that's bad about corporate art and intellectual property

1519524

Sarah Shearing @sshearing.bsky.social · 06/10/2025

top 6 non-eevees: mudkip mudkip mudkip mudkip mudkip mudkip

010

Reposted by Sarah Shearing

Merriam-Webster @merriam-webster.com · 01/10/2025

We are thrilled to announce that our NEW Large Language Model will be released on 11.18.25.

647249378241

Reposted by Sarah Shearing

Lee McLeod @leemcleod.bsky.social · 29/09/2025

My whole group chat is mourning competitive magic and trying to quit but I'm so glad magic is making more money than ever.

1133

Sarah Shearing @sshearing.bsky.social · 28/09/2025

I also started during lorwyn and im also planning on lorwyn being my off ramp! Ive kinda already off ramped but ill stop doing cube updates after lorwyn too

020

Reposted by Sarah Shearing

Katie (the other one) @kayfabegamergirl.bsky.social · 27/09/2025

Card Titan I take back everything bad I've ever said about you

69515

Reposted by Sarah Shearing

Anna Phylaxis @quatoria.bsky.social · 10/09/2025

Charlie Kirk was a man of violence. He cheered for violence, he helped create violence. He used violent rhetoric to inspire others to violent actions. He spent his time on this earth making it a more hateful, violent place. I do not consider it especially shocking that he met a violent end.

752035420

Reposted by Sarah Shearing

Sean T. Collins @seantcollins.com · 23/08/2025

I need Democratic online types to understand that because of this, Newsom is dead to me. Dead. I'd sooner vote for an actual corpse. There is no hope, none, that he'll win me over, *because he has established he is a bad person*. There's no coming back from this. Take my advice and move the fuck on.

13161652153

Sarah Shearing @sshearing.bsky.social · 19/08/2025

Oh we are nowhere close to AGI and it doesn't have to do with costs, it has to do with the models themselves. The models don't think. They don't 'learn' in the sense that we do. They *can't* grow. They're just the worlds most sophisticated matrix multiplication systems.

120

Sarah Shearing @sshearing.bsky.social · 19/08/2025

Right, and I'm saying I think you're wrong, with the smart money. The problem is, whether or not it can be done, I think its close to 0% for anyone to actually try to do it. There's simply no money in it. Magic doesn't have the same allure to it like league or history to it like chess.

110

Sarah Shearing @sshearing.bsky.social · 19/08/2025

yeah basically. The question isn't can it be done. It can. the question is, can you get the right people on the problem and can you secure the funding to do it.

230

Sarah Shearing @sshearing.bsky.social · 19/08/2025

granted, the question we're trying to solve isn't optimal play, its 'better than a human'.

010

Sarah Shearing @sshearing.bsky.social · 19/08/2025

Actually I think this would be the most difficult part about building an AI model for magic; you'd need to build a simulation engine to perform the reinforcement learning, and uhhhh. It can be done I guess. but man I wouldn't wish to be the one in charge of that

120

Sarah Shearing @sshearing.bsky.social · 19/08/2025

my point was that its explicitly not solvable, and even though there is mathematical structure to language, no one actually uses any of it when building these models.

110

Sarah Shearing @sshearing.bsky.social · 19/08/2025

I would agree that LLMs don't beat humans at language in general, but when broken into smaller tasks, models often outperform humans now (e.g. a model can outperform humans on translations consistently, or summarization). I just brought up language as its my area of expertise.

110

Sarah Shearing @sshearing.bsky.social · 19/08/2025

The outcome is often just a new position that is favorable by some reward metric. It might be computing 5-10 positions deep, but that's an infinitesimally small slice to all the future positions possible.

100

Sarah Shearing @sshearing.bsky.social · 19/08/2025

I think there's a misunderstanding about what the outcomes here are. The machine isn't taking some input position and computing every possible future position from that and determining which wins and loses. Because that's not computable.

100

Sarah Shearing @sshearing.bsky.social · 19/08/2025

like one layer activates mostly on numbers, or quotes. Or another mostly on prepositional phrases. But that's an oddity. In general, none of it is human readable in any sense, and whatever the model 'learned' in its weights is a complete black box

000

Sarah Shearing @sshearing.bsky.social · 19/08/2025

There's no module for syntax, or word order, or structure of any kind. We just build absolutely enormous models and then throw all the data we can find at it and let it learn or discover any structures on its own. You'll often find layers that seem to do something specific to a human

200

Sarah Shearing @sshearing.bsky.social · 19/08/2025

but like, for example, we build neural networks for language. They translate, summarize, etc. But, as someone who builds language models for a living, the current state of the art doesn't actually utilize anything about the structure of language

110

Sarah Shearing @sshearing.bsky.social · 19/08/2025

I would need to know how you understand neural networks to work I guess. But for the most part, neural networks as a whole aren't solving problems; they approximate them. That's kind of the whole shtick with optimization algorithms (and neural networks are optimization algorithms).

120

Sarah Shearing @sshearing.bsky.social · 19/08/2025

Right, but whether or not its mathematically solvable doesn't actually have any bearing on the models we're using, because for all practical purposes they aren't solvable. So we use approximation / optimization / learning algorithms instead, which are the same algorithms I'd start with for magic.

100

Sarah Shearing @sshearing.bsky.social · 19/08/2025

One counterpoint to this though is that specifically MOBA games, like Dota and League, AIs are not *currently* able to beat the best players consistently. I'd have to look into the literature to see where the state of those AIs left off though, I am not completely aware of all the work done there

010

Sarah Shearing @sshearing.bsky.social · 19/08/2025

For games with hidden information, you can look at for example starcraft (fog of war). I'd have to look up further for other games. I know rocket league was competitive but not necessarily always better.

110

Sarah Shearing @sshearing.bsky.social · 19/08/2025

I mean, chess. Chess isn't actually solvable. There are more possible chess positions than there are atoms in the universe. You cannot build a model that 'solves' chess in the sense that it can know every possible outcome. Go is even worse (10^700 I believe vs 10^20 for chess).

210

Sarah Shearing @sshearing.bsky.social · 19/08/2025

and when I say enough times, I'm talking billions of simulations. More simulations than the total games of magic humans have played.

000

Sarah Shearing @sshearing.bsky.social · 19/08/2025

Reinforcement learning loops don't require you to have perfect knowledge of outcomes. You can simulate outcomes. Simulate *enough times* and the model will learn what actions to take that maximize desirable outcomes

100

Sarah Shearing @sshearing.bsky.social · 19/08/2025

Because we've solved equally difficult problems. Magic isn't actually all that more complex than 'solvable' games. It simply has an additional layer of unknown information. Fortunately, you can still build the same models that take unknown information into account.

210

Sarah Shearing @sshearing.bsky.social · 19/08/2025

Now, alpha go was 2017. So hardware should be cheaper? Unfortunately, not really. With the crypto boom, even old hardware maintains its price. So right off the bat, you have to ask: what motivation does anyone have to actually go and spend that kind of money on this?

000

Sarah Shearing @sshearing.bsky.social · 19/08/2025

Take for instance Alpha Go. The hardware alone to host an Alpha Go system costs around 25 million dollars. To actually run that hardware to train a model is an order of magnitude higher. Add in development costs, and we are easily 9 figures in terms of price.

200

Sarah Shearing @sshearing.bsky.social · 19/08/2025

I've had this discussion before but basically, the only reason there isn't some AI that beats people at magic consistently is because *its incredibly expensive to build*. But we already know how to do it.

210

Sarah Shearing @sshearing.bsky.social · 10/08/2025

Yeah its already set up and ready

170

Reposted by Sarah Shearing

Arya (hogpog_98 elsewhere) @aryawrites.bsky.social · 03/08/2025

Saw this from Tumblr and it really resonated with me. If you engage with/talk a lot about HP stuff, you can and should do better. I don't think people realize how much referencing this stuff continues to give her power and makes the trans people around you uncomfortable.

1368131

Sarah Shearing @sshearing.bsky.social · 25/07/2025

I am pumped for no one. There will never be a UB set i want. I thought FF was one of the worst magic sets ever because it was all so disconnected and the cards look straight up bad. But fuck me i guess, magic just isnt for me anymore

000

Sarah Shearing @sshearing.bsky.social · 25/07/2025

I hate all UB equally. Every UB set except maybe LOTR is a stain on magic. With the exception of high fantasy IP, UB just does not look like magic (which is why LOTR gets a pass. Though it also didnt need to happen).

140

Sarah Shearing @sshearing.bsky.social · 06/07/2025

Some alters i had commissioned from @revelens-light.bsky.social for the vintage cube. Zach did an incredible job!

181

Reposted by Sarah Shearing

big tiddy Rebell @rebell.bsky.social · 30/06/2025

Been thinking about this lately

1925

Reposted by Sarah Shearing

CalebDMTG @calebdmtg.bsky.social · 30/06/2025

If human rights need to "poll well" before you touch them then you have no true beliefs. You are a soulless husk of a human being

5390118

Reposted by Sarah Shearing

Samantha Murphy @samrosemurphy.bsky.social · 30/06/2025

I guess I will continue not to play legacy. I really really wish they were more willing to remove things from the format, I get the feeling they don't want to ban things for players sake in regards to $ but I think the format is worse for it and the player base is shrinking which feels much worse

0203

Reposted by Sarah Shearing

Ryan @r-condon.bsky.social · 30/06/2025

Overall, I like it a lot. If you’re going to have very few windows, use them aggressively in ways that actively values changing play patterns!

151

Reposted by Sarah Shearing

pissduggery 🎆🐍 @fishduggery.bsky.social · 30/06/2025

gonna post a bunch of thoughts on the legacy B&R. I think the most important note to keep in mind is legacy in 2025 is inherently a casual format. There are competitive events, but it is a format defined by casual interest. Without weeklies and paper play, legacy dies. This B&R ignores this fact.

65712