Will Garner @considersomething.bsky.social · 15/10/2026I think as Ctrl-R goes on it does a lot of interesting work to ensure that while the quest marker gets you there, you will have to look at the environment to understand where you are. Resonant's "establish normal so you can do weird" phase is maybe out of balance and doesn't register as what it is. 000
Will Garner @considersomething.bsky.social · 09/10/2026the whole point of expertise is people figuring out the last few percent of the details, which the current AI paradigm is never going to be able to do. It's doing weighted combinatorics. Locked in to whatever the rated dataset says is most likely, plus a random variance to appear specific. 000
Will Garner @considersomething.bsky.social · 09/10/2026do I want to "understand" my car? nah. Do I want the people who made my car to understand it? yeah. do i want them to gain in experience and improve their understanding? also yes. What if they just shuffle things together that mostly kinda should work? Tesla. glue the panels on, they'll look okay. 230
Will Garner @considersomething.bsky.social · 28/09/2026always thought it would be fun to set up an arcade machine that was an emulation library that was like play random game for ten minutes: free continue same game for another fifteen minutes: one dollar pick a game for ten minutes: five dollars 010
Will Garner @considersomething.bsky.social · 27/09/2026probably! My denotation vs. connotation sense is usually pretty clear so I'm trying to figure out why it is insisting these are both useful things and not just a typo. I would love to see if there's a a different usage history in games, because it does feel like that's where it coming from. 000
Will Garner @considersomething.bsky.social · 27/09/2026at the end of a hone in process, you know exactly where the target is but are not there. At the end of a home in process, you are where the target is. 000
Will Garner @considersomething.bsky.social · 27/09/2026As interpreted on a radar screen (or more accurately, video game HUD interpretations of them), honing works as a visual metaphor - the better a fix you have on a target signal, the more the slice of the circle pointing towards it narrows. Hone in on a signal, while missiles home in as they approach 200
Will Garner @considersomething.bsky.social · 26/09/2026fruit flies are useful for study because they are ALMOST simple enough to understand, so new understandings can be tested against it. The almost is critical. A fruit fly model that only does the things we understand (i.e. what we can currently recognize) is useless without a real one to return to. 000
Will Garner @considersomething.bsky.social · 13/09/2026Pamela and Jason were also escaping The Shop just off-screen in Firestarter. 010
Will Garner @considersomething.bsky.social · 10/09/2026accelerated growth, and then it sees a way to progress towards that distant original goal and betrays everything people thought it was doing in an instant. Believers have a lot of experience with exponential growth in simple artificial systems, not much with what always stops it in the real world. 000
Will Garner @considersomething.bsky.social · 10/09/2026(this is both the weakest link and the most credible given the world we live in. Why wouldn't we just turn off the corporation that we all know is actively harming the world? well.) 100
Will Garner @considersomething.bsky.social · 10/09/2026AI does a specific thing, does a good job of it, gets more resources, people lose track of the original thing it is doing and are impressed. Transition period where it has so much money on tap it can pay people to do stupid things in the real world 110
Will Garner @considersomething.bsky.social · 10/09/2026if you want to know (and this is not an endorsement, I don't believe this, i just read the story), this is the straightforward novelization of it: www.royalroad.com/fiction/4801...royalroad.comThe NumberThere were two things I knew instantly when I began to exist. The Number was 1922916.12. And I had to make the Number go up. [participant in the Royal Road Writathon challenge] Note: If you enjoyed t... 100
Will Garner @considersomething.bsky.social · 30/08/2026(for anyone else who has to look it up the plants right out of the ground do look delightfully like a bunch of the chickens that got the name next and then) 100
Will Garner @considersomething.bsky.social · 14/08/2026feeling crazy that no one is mentioning this in the replies. he's 23? meaning when he was 17-18 the covid pandemic hit and i assume there were a lot fewer house parties. Remember when they didn't go to high school in person for a year? 190
Will Garner @considersomething.bsky.social · 13/08/2026redirecting collecting impulse into drawing old toys. 000
Will Garner @considersomething.bsky.social · 04/08/2026"is this going to be about more ant-mimicking spiders?! oh... no. sigh. ai sucks." #realfakeantsonlyplease 031
Will Garner @considersomething.bsky.social · 31/07/2026Maelstrom by Peter Watts (2001) got "ai will fill the open internet with fabricated content based on whatever people search for, rendering it useless" 141
Will Garner @considersomething.bsky.social · 26/07/2026there are two statues guarding a door. one says "i am ontologically unable to remember what 'ontological' means." the other uses a version of the ontological proof of the existence of God to state that he can beat up fictional characters. What one statement can you make to reveal the solution? 000
Will Garner @considersomething.bsky.social · 05/07/2026the purple on that one was always closer to blue. the more interesting question is, does it still have the (only-in-this-gimmick-version-maybe?) bright pink gums in the corner of the mouth? 010
Will Garner @considersomething.bsky.social · 29/06/2026the best pitch (enough to bookmark, but I haven't switched yet) I've heard is for Proton, who has a mail/vpn/cloud storage. they also have an AI thing (that they say is open source and never trained on user data), but it's a Swiss company so maybe the EU privacy standards will hold up a bit better? 010
Will Garner @considersomething.bsky.social · 13/05/2026the no-true-Jedi or (more frequently) the no-true-Batman process. 000
Will Garner @considersomething.bsky.social · 13/05/2026i think there's some practical benefit in applying the no-true-scotsman process to fictional heroes with a lot of material behind them, if you do the extra step of realizing at the end you have not discovered the true form of the character, but instead a working model of what you value. 100
Will Garner @considersomething.bsky.social · 25/02/2026It's in the basic design of the thing. Create a statistical model, and add randomness so it's different every time. Not that "always pick the most likely" is that great, but "pick within five degrees of the most likely answer" is guaranteed to be wrong regularly and never get better. 000
Will Garner @considersomething.bsky.social · 22/02/2026(bad answer: fancy evaluation requires complete sentences, expanding the phrase to a sentence restores the fancy) "they talked to me" "What subject did they talk to you about?" "About what subject did they talk to you?" (real answer) "about what?" sets you up to get answered "chicken butt" 030
Will Garner @considersomething.bsky.social · 12/02/2026Art co-ops in which artists manage the storage and disposal of their own various solvents and materials with no oversight burn down with truly unfortunate regularity. and artists will absolutely experiment with historically-abandoned-because-danger materials and processes, given a chance. 120
Will Garner @considersomething.bsky.social · 26/01/2026if you're dreaming about power to stipulate, how about making each member take a test about the bill before they're allowed to vote on it. (from SF novel Vigilant, by James Alan Gardner) 010
Will Garner @considersomething.bsky.social · 20/01/2026weirdly, Cameron's plot summary of the second book of the Lord of the Rings trilogy will focus on the titular towers, but describes only Sam and Frodo approaching two holes in the ground, briefly descending each before loudly asking each other, "which of these two towers do you think Gollum is in?" 020
Will Garner @considersomething.bsky.social · 16/01/2026there once was a lad of my creation who developed quite an aspiration he said "i'm not your ward" and then drew his sword declaring, "i am a sovereign nation" 010
Will Garner @considersomething.bsky.social · 10/01/2026horse eating grass and having bits left dangling out the corners of the mouth: very common. 000
Will Garner @considersomething.bsky.social · 10/01/2026being from Kentucky and having seen untold numbers of horse logos, the neck is horse enough to carry the drawing, but those lines are hard to explain. Let me suggest: it's got some sort of long plant leaves in its mouth. It's been eating crops and it's on the run now. This is a wanted poster. 110
Will Garner @considersomething.bsky.social · 01/01/2026it's cyberpunk, Charlie Brown (Charlie Brown bouncing around the Killing Floor out of control, tumbling and flipping, while a vat grown assassin just misses him, forever. His HUD flashing SNOOPY in the corner of his vision) 000
Will Garner @considersomething.bsky.social · 27/12/2025oh! this is a good one. The finite set of DVD movies contains a set of movies watching over your shoulder as you watch two hours of an infinite number of different movies - but enough of those recordings are identical enough that, discarding copies, there is a finite set of them. 000
Will Garner @considersomething.bsky.social · 27/12/2025that makes it a hash function - useful for reducing sentences of infinite length down to a set length, but guaranteed to produce identical duplicates which cannot be re-expanded. 100
Will Garner @considersomething.bsky.social · 27/12/2025A third possible method of infinite sentence generation is fractal - recombining meaning into ever smaller units. But in digital representation, that it limited by the smallest unit of data - you can say a given 1 or 0 means multiple things, but in practice you'd be making identical copies. 100
Will Garner @considersomething.bsky.social · 27/12/2025later he gives an example of infinite sentences, by inserting a loop. but that extends the sentence infinitely. Godel proofs, which state "this is a valid sentence", "'this is a valid sentence' is a valid sentence", etc. both lose their infinite property when capped by length. (same with conjoining) 100
Will Garner @considersomething.bsky.social · 27/12/2025sure, happy to look. Chomsky... simply states that infinite sentences are possible, without discussing their length. From there, the rest is a discussion of how to differentiate grammatical and ungrammatical sentences. 100
Will Garner @considersomething.bsky.social · 27/12/2025i don't know what limits you thought about when stacking boxes. I thought about angle of repose vs. curvature of the earth, tensile problems of an orbital elevator. then I mentally tore another planet into crates and covered the surface of the earth until the diameter met the moon. Finite. 100
Will Garner @considersomething.bsky.social · 27/12/2025If you want to take one 2-hour movie, and interpret it into other languages? There will be plenty of other movies doing exactly that! but not an infinite amount of them. 100
Will Garner @considersomething.bsky.social · 27/12/2025I think you can? because a DVD is a device for storing 3.76e^10 bits, each of which can be 1 or 0, as one long number. There is nothing else there. All possible things which can go on that disc are represented by that finite set of combinations. It is all the 2 hour movies. 100
Will Garner @considersomething.bsky.social · 27/12/2025In particular, AI is pushing the idea of infinite potential when what it's doing is more like a million variations on pieces of a billion works that people have made. Seems "infinite", more than you can ever personally experience. but sad in comparison to true exploration of what's possible. 000
Will Garner @considersomething.bsky.social · 27/12/2025To step back from arguing in circles to why I think this matters: there's a lot misinformation that relies on blurring the difference between things like "millions", "more than you'll ever count", "more than we can use in our lifetime", "uncountable", and "infinite". 100
Will Garner @considersomething.bsky.social · 27/12/2025There are theoretical limits on the number of words that could exist. The logic proofs involve looping or indefinitely extending words - once you cap the length, you limit the language. In two hours, with an average phoneme length of 100ms, you get 72,000 phonemes. Incalculably vast, not infinite. 100
Will Garner @considersomething.bsky.social · 27/12/2025I don't think you're getting me? A DVD stores a finite number of digits. The number of possible combinations of the values of those digits is finite. All those possible combinations, set end to end, are finite. And that makes every possible thing that could be on that DVD. 100
Will Garner @considersomething.bsky.social · 27/12/2025possible 2-hour movies *in a digital format*. Like on a DVD. number of pixels x number of colors x number of frames. same with audio encoding. It's a number. a vast, incomprehensible number, to be sure. but it's not even one of the smaller infinities. 100
Will Garner @considersomething.bsky.social · 26/12/2025to put it another way: The number of ALL possible 2 hour movies (in a digital format) is a finite number, which can be written as a single number. This includes a series in which you give advice to an AI about how they should respond to every possible situation to do the work of consciousness. 100
Will Garner @considersomething.bsky.social · 26/12/2025You probably could, given enough boxes? At what point, when you're standing on a platform of boxes, could you not build another, smaller platform of boxes? If it would require all the computers ever a million years to run a brain for a fraction of a second, then it would still run. 200
Will Garner @considersomething.bsky.social · 26/12/2025To your point, even though it could be represented by a single program, the real complexity multiplier is when it is spun off into many different processes running on specialized pieces of hardware. which is what even our non-intelligent computers do. 110
Will Garner @considersomething.bsky.social · 26/12/2025For a true general intelligence, I would expect that it would have something like the model we have, hooked up with a hundred other very different programs that do specific things, all hammered together in different ways for a long time. Doing one thing to a vast amount of data will never get there. 100