Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026If that’s your definition, I assume of “understanding,” not “thinking,” and believe it encompasses everything that could be reasonably described as “understanding,” without including anything that wouldn’t, then your argument is that under that definition, AIs have meaningful general understanding? 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026Without defined premises, all conditional statements are false. You wanted to rely on implicit definitions, I refused, you’ve been trying to argue against straw men since. 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026I wasn’t making an argument, I was saying that I must have missed it. Which, like, how many times have you replied to me now? Anyway, walking you through this before you whine again, the steps are: 1) Define terms & make positive claim 2) Decide whether the claim is both true & meaningful 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026I didn’t dismiss those as facile. Arguing against a straw man is your tell. 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/20261) I’m not making the positive case; the burden of proof isn’t on me 2) I’ve said multiple times that there are multiple definitions of “thinking” and “understanding,” including a lot of unhelpful ones, which is why it’s important for people arguing that LLMs are eg “thinking” to define their terms 200
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026“understands” its substrate, and individual ants can be argued to “think,” but saying that really doesn’t get you very much past the ants colony or very far toward claims that they think or understand like humans do, or would recognize us as thinking. 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026surroundings through the iterated behavior of individual agents in stochastic behavioral algorithms, but don’t understand, say, the offsides rule, or human intentions or words, even if you could train them to form the shape of words when cued. To the extent a colony is a single entity, it 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026The argument that they generalized their internal models, rather than building a new internal model for each problem would seem to require that their internal model be legible & shared, but sure, let’s say that’s “understanding” for the sake of this argument. Then ant colonies “understand” their 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026Literally where has either Will or Michael given a definition of “thinking” or “understanding”? Neither of those are simple terms, & like I’ve said, have plenty of fatuous & facile definitions that don’t necessarily mean anything interesting that isn’t already inherent in descriptions of AI w/o them 200
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026Got caught? You wildly overestimate how important you are. Lol just say yes, you believe representation is reality and AI Seinfeld is the same mind as real Seinfeld That will save us the time of having to argue whether math is real or abstract. 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026No, it goes into why computational functionalism is wrong Once again you’re not beating the perception that your approach is superficial spam and can’t actually articulate the arguments you’d like to make. 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026Must have missed it in your mania—where, exactly have you defined what you mean by “thinking” and “understanding”? 101
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026Nah. I’m not “rejecting an empirical benchmark,” I’m saying it’s a non sequitor without having defined your terms. Maybe spend less time trying to be snarky and reposting your own replies and more time understanding the very basics of how to make a clear argument . 200
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026Oh, sorry—is the map the territory? You’re replying like you’re on a meth jag, and I guess I just don’t care about you that much. 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026& a way to verify that, through checking it against others who can understand the problem. It’s also implicit that “understanding” has multiple nested levels & provides a framework for arguing whether or not an algorithm that produces novel math solutions is epistemically “understanding” them 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026He specifically says that it would take reverse engineering the proof to increase understanding of the problem. Claims that AI “understand” (separate from thinking) then have a reasonable definition: whether the model or entity can abstract and generalize from this approach to other problems, 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026I already asked you if the map was the territory and don’t feel like scrolling through your compulsive replies—did you answer that? 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026Nah. As per prior, the method that was used to solve the “open math” was iterating billions of times on previous strategies, effectively brute forcing the Navier Stokes without enlarging the understanding of it. They’re all iterated probability vectors based on a training corpus. 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026It’s not a functionalist vs essentialist argument at its core, and you can’t actually conceive of anyone else’s arguments to be worthwhile 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026He’s off on a clownish computational functionalism. www.noemamag.com/the-mytholog...noemamag.comThe Mythology Of Conscious AI | NOEMAWhy consciousness is more likely a property of life than of computation and why creating conscious, or even conscious-seeming AI, is a bad idea. 110
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026Those aren’t the only two positions; neither represents mine; your assertion is bullshit that wastes both our time 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026lol Yes, the burden of proof & definition of terms for evaluating an argument being on the positive case is literally foundational logic. Or “reasons” Either one Probably just gonna mute you if you can’t recognize that 110
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026So, when people, or dogs, or whatever, think, their statistical model is updated by a prompt? And LLMs can’t return the same information from the same seed because they’ve updated their statistical model through thinking? 010
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026I’m not anti-LLM. I think they have costs and benefits, and sloppy language from both boosters and critics obscures having accurate discussions about them. We understand how LLMs function much better than we understand brains, despite having much more experience with brains. 020
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026dubious metaphysical language that obscures understanding both the benefits & costs of the use of LLMs. 110
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026That’s different from building causal inferential mental models. It’s not because I think humans are special; kind of the opposite: anthropomorphism like “thinking” and “understanding” as they’re commonly, incoherently used are unnecessary to describe what LLMs do & smuggles in a bunch of 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026understanding their subjects, they’re seen as repeatedly iterating in a probability space around superficial resemblances—the way they get better at hands is seeing more images of hands, not building an underlying model of anatomy, because the underlying model is superfluous for the task. 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026many of the arguments around LLMs & math are based on category error—chip fab ML algorithms have been beyond human math for at least a decade. No one was seriously arguing that the algorithms were “understanding” or “thinking.” Likewise, image generation algorithms aren’t described as 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026about thinking claims, but I don’t think that’s what Will is claiming—he’s repeatedly framed the issue around illegibility eg chain of thought, and autonomy, eg rule breaking. I believe one could make an argument that legibility matters for “understanding,” but not for “thinking.” I also believe 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026Notably, a lot of things that are process products of thought would then become thinking—you could make a case for bylaws documents, games that scale difficulty against player skill—including purely mechanical slot machines. That’s fine, & it gives a starting place to talk about what’s interesting 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026Given that action frequently precedes thought, & learning (incorporating new knowledge [justified true belief]) can happen without thought, that makes any dynamic system with a feedback loop a thinking system, & would include a lot of things we generally believe to be unthinking or agglomerative 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026Because I’m not making the positive case and want to know the claims first. There are a lot of trivial definitions of “thinking” or “understanding” that LLMs definitely meet. But those generally aren’t what’s being implied. Feel free to put up or shut up. 110
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026Yeah, no, that’s bullshit, but it does show why you keep arguing against a straw man. 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026My entire fucking point is that minority factions in a coalition have very little power in a general election compared to primaries, & can only threaten a lose-lose strategy by hostage taking, which undercuts building power within the coalition. It burns capital for definitionally less power. 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026For me, this thread still descends from me saying “prove it. Define your terms.” Which, notably, you haven’t. You’re instead trying to argue against a rejection of the positive case without making the positive case. And now, when that’s pointed out repeatedly, complaining of “motivated reasoning.” 210
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026Are you denying that they share any qualities at all? And asserting that the mistaken anthropomorphizing is a rare occurrence with either? 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026facile and fatuous claims about their abilities and functions, and for the vast majority of their work, whether they “understand” is at best irrelevant to their function, & making arguments of risk based on a presumption of understanding leads to a bad mental model of their risks & uses. 110
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026My point is that people are reacting the same to their subjective experience with the chats. More fundamentally, my point is that both of them are built by humans for human users with predictable reactions to text prompts. Arguing they “think” or “understand” requires strong definitions to avoid 220
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026Would Eliza be thinking if it included dice when responding? 110
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026No, Will has repeatedly said “thinking” and “understanding”; others have made sentience claims. Rejecting my mention of sentience here while not addressing the other arguments is weak evidence of any ostensible motivated reasoning on my part. If you believe AIs “think” and “understand,” define it. 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026lol If all you know about politics is general elections, maybe But primaries are where the governing coalitions are formed prior to the general election, and that’s where factions have much more power 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026Yeah, no, I meant this: mathstodon.xyz/@tao/1172078... It’s not about the “stealing,” it’s about not understanding what mathematicians do.mathstodon.xyzTerence Tao (@tao@mathstodon.xyz)A concrete example of how AI advances in solving key open problems could inihibit the future development of a mathematical field can be found in the global regularity problem for the incompressible Na... 200
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026lol yes my motivated reasoning of demanding rigorous coherent definitions before accepting claims ya got me 100
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026No, I’m pointing out that the anthropomorphic claims are substantially the same, especially user subjective claims of thinking and sentience, and math proofs aren’t a good reason to argue that’s removed the illusion. 200
Kierkegaarden Cop @joshsteich.bsky.social · 23/09/2026You can punish them in the next primary, where you have leverage again. Are you not understanding this as cycles? 200