site banner
PaperclipPerfector  1mo ago (text post)   25342 thread views

Culture War Roundup for the week of April 27, 2026

This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.

Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.

We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:

  • Shaming.

  • Attempting to 'build consensus' or enforce ideological conformity.

  • Making sweeping generalizations to vilify a group you dislike.

  • Recruiting for a cause.

  • Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.

In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:

  • Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.

  • Be as precise and charitable as you can. Don't paraphrase unflatteringly.

  • Don't imply that someone said something they did not say, even if you think it follows from what they said.

  • Write like everyone is reading and you want them to be included in the discussion.

On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.

3
Top Bottom Old Controversial Comments
Jump in the discussion.

No email address required.

based on all the chess games stored in its database.

It doesn't have a "database," this is a fundamental misunderstanding of what's going on under the hood. With LLMs solving open math problems, I'm puzzled that the discourse remains around "it's just doing what it's seen before" with various levels of unsound understanding.

LLMs can reproduce 96% of the text of Harry Potter verbatim. Even if they do not store all their training data with perfect fidelity, their underlying operations are such that it doesn't matter. It's data compression with variable loss depending on what they were trained on. When 1:1 outputs from their memories of training data can't exist, they reach for similar patterns and smooth over the disjunctions using sophistry. They must be commended for semantic fluency.

What is this supposed to prove? There are people who have memorized the Torah or the Quran. It's still not the case that they are merely doing some kind of database lookup when you ask them about a verse, and that implies a fidelity that simply doesn't exist. And if you concede that there isn't perfect fidelity, one wonders what the purpose of discussing "database lookups" in the context of LLM inference is other than rhetoric.

When 1:1 outputs from their memories of training data can't exist, they reach for similar patterns and smooth over the disjunctions using sophistry.

Dismissing as mere sophistry novel LLM-discovered software exploits and math theorems is absurd.

It goes towards proving the basis for what we observe: that LLMs are very good at recalling large and disparate amounts of knowledge but are poor for functionally utilizing said knowledge, especially in matters complex, unusual, or otherwise not 1:1 with stuff from their training material. Whether this proves or disproves they are sentient or intelligent or whatever is a matter of semantics, but what it does do is give us a clue as to why we observe certain disparities in their capabilities, and can help inform our expectations about what further capabilities might emerge.

Humans lean on theory, trained pattern spotting, and various heuristics or memorized devices (i.e. king opposition) when playing chess. Memory plays a role to, but outside of maybe Magnus Carlson it is dwarfed by the capacities of LLMs. This is a level of intelligence that can also be employed for creating architecture or symphonies. LLMs lean a lot harder on brute memory recall (although I won't discount entirely their capacity for higher-tier reasoning) through hyper-intensive statistical calculations, and these make it very good for things like discoursing on a broad variety of facts or semantically juggling abstractions, but they do not, apparently, allow LLMs to create complex architecture, symphonies, or do anything else involving the complex interlocking of smaller elements.

The small elements are found in its memory and can be expurgated intact individually, but the LLMs do not possess the intelligence to complexly fit them together. The LLMs do not operate at a level of intelligence that would allow that. They are hyper-intensive exploiters of lower order processes but not high tier ones. That's what's suggested by the fact they can recall 96% of a novel. That they lean on highly scaled relatively brutish methods to repeat stuff verbatim, or close enough.

poor for functionally utilizing said knowledge, especially in matters complex, unusual, or otherwise not 1:1 with stuff from their training material.

Like I said:

Dismissing as mere sophistry novel LLM-discovered software exploits and math theorems is absurd.

"LLMs haven't written a beautiful symphony or designed a beautiful building" is simply moving the goal posts. There's no reason that those are the true test of putting things together and theorems and exploits don't count.

I take the ‘opposite’ view that LLMs are becoming extraordinary intelligences, but I also think the distinction between memory, recall, training set, database etc is unnecessarily importing computer science distinctions into what is a relatively robust colloquial understanding of these models.

If you watch three thousand chess games and then play a chess game and see a move and think “I’ve seen this before, I’m going to do x” and you’re right but you can’t perfectly recall that it was actually a YouTube video of a 2003 Chess regional championship quarter final between… then are you recalling or remembering or did you learn?

This is just not a relevant distinction when it comes to the human concept of memory. I’ll keep pushing this because “actually, an LLM doesn’t have memory of the training set” isn’t really true. It often does have recall of the training set, just like often you really might be able to remember the book you first saw an unusual turn of phrase in or the chess game where you first saw a particular move. And in any case, memory encompasses both that and a relational, situational, partial and often metadata-free recall but it still counts.

The counterargument here isn’t “no LLMs don’t do this”, it’s “so do you”.

relatively robust colloquial understanding of these models.

This doesn't exist, at least on this forum on down. There's at least one person I talked to who really thought that LLMs were looking through the training data at inference time. It turns out that people using sloppy language ""colloquially"" ("joke's on you, I was only pretending to misunderstand LLMs") can cause people to believe the literal meaning if they don't know any better.

This is just not a relevant distinction when it comes to the human concept of memory.

Agreed.

I’ll keep pushing this because “actually, an LLM doesn’t have memory of the training set” isn’t really true.

This isn't what I said. I said it doesn't have access to the training set, in the same way that if you take an exam without "access" to the textbook you're not allowed to bring it in and leaf through it when answering the problems. It doesn't preclude you from reading the textbook a thousand times and memorizing it verbatim though.

I said it doesn't have access to the training set, in the same way that if you take an exam without "access" to the textbook you're not allowed to bring it in and leaf through it when answering the problems.

Again, if we change “it accesses the training set” to “it recalls / accesses / understands [delete as appropriate] the conceptual relationships represented by the training set” what really changes?

What changes is that one is completely wrong and the other is defensible. One leads you to think that an LLM recognizing Shakespeare (or a lesser author) from a sample of writing is an unremarkable feat, an information retrieval problem that's been solved for forty years. The other causes you to realize that what's going on here is much deeper.

It doesn't have a "database," this is a fundamental misunderstanding of what's going on under the hood.

Maybe I am using the wrong word. What do you call the set of data used to train an LLM? Is it just "training data"?

I'm puzzled that the discourse remains around "it's just doing what it's seen before"

I think a more accurate statement is "It's just making predictions based on what it's seen before." Of course the word "just" might not being doing justice to the capabilities of an LLM. Because they are definitely very impressive.

But anyway, my point is that it's possible for an LLM to make legal chess moves without actually modeling chess. Do you dispute this?

What do you call the set of data used to train an LLM? Is it just "training data"?

The point is that the training data is not accessible at inference time. To the extent that being trained on chess data gives the LLM information about how to respond to a particular opening, it's because the LLM has learned that information, similarly to how a human studying openings has.

But anyway, my point is that it's possible for an LLM to make legal chess moves without actually modeling chess. Do you dispute this?

Sure, in the same way that it's possible for a human to make legal chess moves without modeling chess:

  • you could just get lucky and make random moves that happen to be legal
  • you might know how all the pieces move and that the goal is a checkmate but have basically no understanding of strategy (I am here btw)
  • the above, but you might have studied a book on chess openings and endgames

It's unclear to me at which point even a human can be said to "model" chess.

The point is that the training data is not accessible at inference time. To the extent that being trained on chess data gives the LLM information about how to respond to a particular opening, it's because the LLM has learned that information, similarly to how a human studying openings has.

I'm not sure I understand your point here.

Here's a claim I am making:

A possible reason why an LLM makes a legal chess move is that it simply makes a good (but imperfect) guess as to what's likely to be the next move after a sequence of moves, as a result of all machine learning from all the chess games in the training data.

Do you dispute this claim?

It's unclear to me at which point even a human can be said to "model" chess.

Here's an example of what I think it means to "model" chess. Suppose you are playing postal chess with someone -- instead of sitting at a chess board, you send each other postcards with chess moves written on them. After a few postcards go back and forth, you decide it would be helpful to set up a chessboard in order to keep track of what's going on and for each move, you make a corresponding move on the chessboard. Or, if you are really talented, you envision a chessboard in your head. Those are models. They are representations of the game which enable you to analyze the game.

Do you dispute this claim?

Only to the extent that this claim applies to humans too, so it's not clear to me how this is supposed to draw a line between what humans do and what LLMs do.

After a few postcards go back and forth, you decide it would be helpful to set up a chessboard in order to keep track of what's going on and for each move

Okay. But we know that LLMs can keep track of the game by printing the current state of the board and updating each time you or it make a move. So in what way do LLMs not model chess?

But we know that LLMs can keep track of the game by printing the current state of the board and updating each time you or it make a move.

Is this what the LLM is doing or is this what the agentic software harness around the LLM is doing? You previously pointed out how colloquial information pollutes and poisons understanding of the technical process that is actually occurring. You just tripped over that yourself. The LLM is not doing any of this.

I admit I couldn't find examples of people testing LLMs' ability to continually update e.g. a FEN representation of the board. However, I did find that (certain) LLMs have actually been able to play chess just given the moves made so far (no actual visualization of the board, no harness) and make legal next moves since 2023. If you have an explanation of how it's doing that with no internal model of the board (as shown in this paper for a toy model), I'm all ears.

I'm making a very restrained comment, nothing in regards to whether or not LLMs have an internal model of the board, just that the agentic harness is not the LLM and attributing features of that harness to the LLM is adding to the overall level of confusion about what LLMs are doing by the lay folk.

I'll look through the paper, I owe someone a LLM-world-model answer based on technical understanding of what a world model actually is, sounds like the findings are related.

You don't need agentic harnesses to play chess with an LLM.

More comments

Only to the extent that this claim applies to humans too, so it's not clear to me how this is supposed to draw a line between what humans do and what LLMs do.

Well the claim I was responding to is that LLMs MUST be modelling chess, because otherwise they would not be able to make legal moves at a rate better than chance. This claim pretty clearly seems to be incorrect.

Beyond that, I don't really understand your point. Here's an example to show what I mean:

There used to be these books you could buy, I think they were called "Informers." They contained records of all IM or higher level chess games for some time period. In theory, you could buy a set of them and have a big library of Informers. Ok, suppose you are playing postal chess with someone and you observe that they make a series of legal moves. Most likely, the person has a chess board set up in their house which they are using to analyze the game. Possibly, they have no chessboard set up and they are just looking up similar games in the Informers and playing whatever moves most masters played in similar positions.

So regardless of whether you are playing a human or playing an LLM, it's potentially possible for your opponent to make legal moves, even a series of legal moves, without modeling the game.

In my non-expert view, LLMs don't create sophisticated models the way humans do. Perhaps chess isn't the best example of this since there was no chess in the ancestral environment. But they definitely can and do create rudimentary models and I think that there's a good chance this will improve a lot in the future.

But we know that LLMs can keep track of the game by printing the current state of the board and updating each time you or it make a move.

I don't know that, but I'm certainly willing to agree that's potentially possible. That's basically how the LLM modeled the simple game I had invented for purposes of testing it. Once it started doing creating a rudimentary model along these lines, it stopped making illegal moves.

So in what way do LLMs not model chess?

By not modeling chess. I mean, even if one allows that an LLM can set up a rudimentary model along the lines you describe, it's not carved into stone anywhere that they must do so.

By not modeling chess. I mean, even if one allows that an LLM can set up a rudimentary model along the lines you describe, it's not carved into stone anywhere that they must do so.

I genuinely don't understand what we're talking about anymore. It's not carved into stone that a human must set up a rudimentary model to play a game of chess either. Is there any distinction, in your view, between LLMs and humans in the chess scenario?

I genuinely don't understand what we're talking about anymore.

Agreed. Let me ask you this: Now that you understand what I meant with the word "database," do you disagree with anything I have said?

Is there any distinction, in your view, between LLMs and humans in the chess scenario?

Yes, as far as I know, at the moment humans are capable of making and using more sophisticated models of chess than are LLMs. (To the extent that LLMs are capable of making models along the lines you described.)

Agreed. Let me ask you this: Now that you understand what I meant with the word "database," do you disagree with anything I have said?

I think we agree that LLMs learn from seeing chess games. I don't agree that LLMs aren't modeling chess at all - there seems to be no evidence for this. A thinking trace of a chess game with an LLM would certainly show the LLM trying to think about possible moves and evaluating their trade offs, even in the absence of a game board that it uses to keep state. I'm sure they are more likely to make mistakes if they aren't keeping track of the board, but so are humans.

To be honest, I'm not sure what kind of evidence would change your mind here if existing thinking traces are not enough. Anthropic researchers found that models are doing complex modeling for tasks as simple as deciding when to start a new line when writing text. Looking into a model to really understand how it's modeling something as complicated as a chess game would require a research budget.

Edit: there was a dude who trained a small model on next token prediction and found that it maintains a board state internally. So I guess now we can put this to bed?

Yes, as far as I know, at the moment humans are capable of making and using more sophisticated models of chess than are LLMs. (To the extent that LLMs are capable of making models along the lines you described.)

Sure, in the sense that good human players can beat LLMs at chess. That said, the best models play at a class C level and do significantly better than the average chess.com player, which I think is hard to explain away by appealing to the models having memorized a bunch of openings.

More comments

It's unclear to me at which point even a human can be said to "model" chess.

Many humans of course do openings in a somewhat similar way; they memorize a bunch. The modelling comes in that a (competent) human will have memorized a number of opening variations, and will play into one that matches what he wants for the midgame; the LLM has essentially memorized a number of opening variations and then picks one using an element of randomness.

It's certainly possible to play good chess without memorizing openings; time constraints are the main reason to do so.

You can say: "Hmm, e4 -- he wants to dominate the centre with that pawn. I need to contest it; e5 would work -- or I could do it indirectly, like Nf6? But then he will just advance the pawn and threaten my knight; seems like a wasted move. Better stick with e5."

This takes much longer than "let's go for the Italian Game", but it's the kind of modelling that you need to do once beyond your memorized opening; LLMs don't do anything like that ever.

This argument smells like the old canard of LLMs not being able to do anything novel, not being able to do anything that they haven't seen before. Again, I think this can be dismissed out of hand now that LLMs are solving open math problems.

LLMs don't do anything like that ever.

LLMs don't make plans while evaluating tradeoffs and then do things to put those plans into action? I don't know how you can even believe that in May 2026. Have you never used a coding agent and had it plan a solution, seen it analyze different approaches with their respective tradeoffs, and seen it propose the option it thinks is best?

This argument smells like the old canard of LLMs not being able to do anything novel, not being able to do anything that they haven't seen before.

Well it's not -- LLMs are clearly useful systems, but it is equally clear that the way they accomplish things does not involve modelling the world.

LLMs don't make plans while evaluating tradeoffs and then do things to put those plans into action? I don't know how you can even believe that in May 2026. Have you never used a coding agent and had it plan a solution, seen it analyze different approaches with their respective tradeoffs, and seen it propose the option it thinks is best?

Using anthropomorphized language to describe something doesn't make it so. Does your car "analyze different approaches with their respective tradeoffs and implement the solution it thinks best" when you apply the anti-lock braking system?

Does your car "analyze different approaches with their respective tradeoffs and implement the solution it thinks best" when you apply the anti-lock braking system?

Can't say my car has ever presented me with multiple options when I press the brakes.

No, it chooses the best one itself -- autonomous agents at your fingertips man.

Maybe the issue is that "interpolation and limited extrapolation paired with a reward signal" is really what intelligence is, and although it's all LLMs are doing, it's also all that humans do too.

What is this place?

This website is a place for people who want to move past shady thinking and test their ideas in a court of people who don't all share the same biases. Our goal is to optimize for light, not heat; this is a group effort, and all commentators are asked to do their part.

The weekly Culture War threads host the most controversial topics and are the most visible aspect of The Motte. However, many other topics are appropriate here. We encourage people to post anything related to science, politics, or philosophy; if in doubt, post!

Check out The Vault for an archive of old quality posts. You are encouraged to crosspost these elsewhere.


Why are you called The Motte?

A motte is a stone keep on a raised earthwork common in early medieval fortifications. More pertinently, it's an element in a rhetorical move called a "Motte-and-Bailey", originally identified by philosopher Nicholas Shackel. It describes the tendency in discourse for people to move from a controversial but high value claim to a defensible but less exciting one upon any resistance to the former. He likens this to the medieval fortification, where a desirable land (the bailey) is abandoned when in danger for the more easily defended motte. In Shackel's words, "The Motte represents the defensible but undesired propositions to which one retreats when hard pressed."

On The Motte, always attempt to remain inside your defensible territory, even if you are not being pressed.


New post guidelines

If you're posting something that isn't related to the culture war, we encourage you to post a thread for it. A submission statement is highly appreciated, but isn't necessary for text posts or links to largely-text posts such as blogs or news articles; if we're unsure of the value of your post, we might remove it until you add a submission statement. A submission statement is required for non-text sources (videos, podcasts, images).

Culture war posts go in the culture war thread; all links must either include a submission statement or significant commentary. Bare links without those will be removed.

If in doubt, please post it!


Rules


Recommended Posts And Communities

Recommended Realtime Chats