An example would be improvised jazz, the musicians need to bend the rules, but they still need some sense of key and rhythm to make it coherent.
Of course everyone has world models. Otherwise people would wander into traffic like headless chickens, if they'd even be capable of that. What he likely means is that not everyone explicitly things of possibilities in terms of probabilities that are a function of Bayesian updating. That does not imply the absence of world models.
You could argue that some people have simpler world models, but claiming the absence of world models in others is extremely arrogant.
Roughly 4% of the population are said to have aphantasia (lacking a "mind's eye"). Around 10% (numbers vary) don't have an internal monologue.
Unfortunately there's almost no research on the consequences of things which many would consider prerequisites for evaluating truth-claims about the world around them, but obviously it's not quite so stark, they are capable of abstract reasoning.
So, if someone with aphantasia reads a truth claim 'X is true' and they can't visualise the evidence in their mind, what then? Perhaps they bias their beliefs on social signals in such circumstances. Personally, this makes sense to me as a way to explain why highly socially conformist people perceive the world; they struggle to imagine anything which would get them in to trouble.
When does having aphantasia mean someone doesn't have a world model? Ditto for an internal monologue? Also the data on subjective experiences is notoriously flaky. I.e. it's highly likely that many people don't even know what an internal monologue actually means when they do in fact have something approximating that description.
Similarly for aphantasia. In fact, you can see a list of notable people with Aphantasia where you can see it includes professional sportspeople, writers, tech founders etc. I.e. you can have no "minds eye" and still reach the highest heights in our society, again, meaning that the mind is still constructing some model of the world and in fact our own understanding of how our brain works is just incredibly limited and basic.
In my opinion, everyone person has a model of the world (kind of obviously) but our brains are more idiosyncratic when we suppose and we represent things very differently to each other, and there is no "right brain" or "wrong brain".
I’d be interested in seeing a study of similar people but in this sample size (n=1), visualising evidence isn’t needed to evaluate it. I’m perfectly comfortable thinking about things without needing an image of it in my head or in front of me.
For example: should we allow big game hunting as a way to fund wildlife conservation? Whoa, not sure. Let me google an image of an elephant so I can remind myself what they look like.
You don't need either of those to have a world model. A world model is a representation of reality that you can use and manipulate to simulate or predict the outcome of your actions. If you are able to discriminate that one of the actions of accepting a $ 1000000 unconditional gift is better than moving in front of a moving train you have a world model.
You can question the sophistication of world models in people — that's essentially what intelligence represents — but not their existence.
>First, much like LLMs, lots of people don’t really have world models.
God you are so convinced of your own brilliance aren't you?
>aphantasia reads a truth claim 'X is true' and they can't visualise the evidence in their mind
That's not what aphantasia is. It's just visual imagery, it says nothing about one's capacity to reason through hypotheticals or counterfactuals.
Compare for instance to a blind person using sound, touch, memorization, signals from a guide dog to navigate.
But if racist uncle talks to his other racist uncle friends who have similar insular lifestyles, the ideas will quickly spread. Until they become big enough to e.g. affect voting behaviour.
As nice as that would be, its only marginally less true.
> everyone without my political beliefs is a model-free slop machine that just goes by vibes.
Nah, some of them are evil on purpose.
but like, in all seriousness. Politics is downstream of a world-model right? And the two predominant world models are giving very different predictions, right? So what are the odds that both models are somehow equally valid, equally wrong (even if its on different cases that somehow happen to add to the same 'moral value')? And we also know that one of the models predicts that climate change isn't real? at some point, a world-model is so bad that it is indistinguishable being a model-free slop machine.
Politics is (if systematically grounded, which for many individuals it probably isn't-and this isn't a statement about one faction or another, it is true across factions) necessarily downstream of a moral/ethical value framework. If that is a consequentialist framework, it necessarily also requires a world model. If it is a deontological framework, a world model may or may not be necessary.
> And the two predominant world models are giving very different predictions, right?
I...don't agree with the premise of the question that there are "two dominant world models". Even people in the same broad political faction tend to have a wide variety of different world models and moral frameworks; political factions are defined more by shared political conclusions than shared fundamental premises, whether of model or morals; and even within a system like the US where there are two broad electoral coalitions, there more than two identifiable political factions, so even if factions were cohesive around world models, partisan duopoly wouldn't imply a limitation to two dominant world models.
Yeah, I agree with this.
> necessarily downstream of a moral/ethical value framework. If that is a consequentialist framework, it necessarily also requires a world model. If it is a deontological framework, a world model may or may not be necessary.
I kinda think that deontological frameworks are basically vibes? And if you start to smuggle in enough context about the precise situation where the framework is being applied, it starts to look a lot like just doing consequentialism.
> I...don't agree with the premise of the question that there are "two dominant world models". Even people in the same broad political faction tend to have a wide variety of different world models and moral frameworks; political factions are defined more by shared political conclusions than shared fundamental premises, whether of model or morals; and even within a system like the US where there are two broad electoral coalitions, there more than two identifiable political factions, so even if factions were cohesive around world models, partisan duopoly wouldn't imply a limitation to two dominant world models.
A 'world-model' is a matter of degree and, at a minimum, pluralities of people in any faction don't really have something that meets the bar. And sure, at the limit you could say that reality is entirely subjective because every individual has a unique to them 'world-model'. But I think that goes a bit too far. And I think there's a pretty strong correlation between the accuracy of a given individual's world model and the party they vote for.
But I think both are true to an extent.
That's why they turn 180 or radicalize badly when exposed to sufficiently strong social or usual media.
Also, no one has a "world model" that is purely based on experiment and reason. Everyone gets their beliefs via other people first and foremost. Some get it from few people, some get it from many people (many people can still be wrong!).
For code, you may have the model of what it does strictly from reason and experience - but probably only if you're the only author. And you can still damn well be wrong, as we all know.
This is the same if you tried asking a general populace ethical questions in a vacuum sneakily. You're going to be dismayed after collecting the set of approved behaviors per culture.
There's not really a way to evaluate one of these.
These blind as a bat Codex pieces are the hamster wheel of silicon valley, he never gets at the problems, which are ad hoc, and only addresses them as post hoc.
Language is FOR deception, this is its endstate, and a particularly juicy one where language is used to pretend the problems aren't related to langauge.
This is all pyschological high comedy. Hide the source of psychosis when it stares us right in the face.
> Humans don't have world models
To
> Of course humans have world models
To
> You fools, there is no such thing as a "world model" and you are all hamsters!
Classic Socratic dialogue.
You can argue that's they're not the governing principle of cognition, but it seems farcical to say they don't even exist, when we are trying to explain them to eachother all the time.
“We refute (based on empirical evidence) claims that humans use linguistic representations to think.” Ev Fedorenko Language Lab MIT 2024
Edit: Also come to think of it, that quote is odd, like it's rather late to the party. The NPC meme is several years old and came from a study that most people don't have an inner voice - that they don't think with words.
What is there then?
What words (heh) do you use to distinguish between someone who makes more accurate predictions about the world than someone else?
In terms of words, they barely represent and never reference. Any statement like that serves primarily status gain, not know knowledge transmission (I proved this from the first statement above as well).
The reality is CS built a math model from totally false premises as it relates to communication and knowledge. It works for efficient value trading using symbols in place of actions. Does it have a future, no.
The problem is how do we shift to a real neurodynamic system of sharing?
https://docs.google.com/document/d/1cXtU97SCjxaHCrf8UVeQGYaj...
The ideas you mention sound interesting, but I’m not sure what the point is.
All the symptoms from LLM failure rates stem from their reliance in arbitrary forms to extract value, and they are no different than the errors we experience in reality in climate, politics etc. CS didn't solve the initial conditions, it's maxing them out as errors.
If a person primarily evaluates the truth content of a statement based on identity or something instead of math/physics/etc then that person has no "world model", and vice versa.
Possibly the same idea: lots of people at work don't appear to think about what they are trying to achieve and just execute tasks very mechanically. The most likely explanation is they are just lazy or bored, and so intentionally or not just haven't thought about the implications of what they do and just do a task someone gave them. Some people appear to be like that in other aspects of life too, they just don't think so don't form any mental model about whatever subject, basically out of laziness or disinterest.
There's lots of subjects I don't care about, say celebrities, that I would not question anything someone told me about them or their lives, even if e.g Taylor swift did something contradictory to the model a fan had of her behavior, I wouldn't question it.
I do wonder about how someone could be simultaneously passionate about something and also not have a model of it. But I think for e.g. some wacky conspiracy, one might be interested in the people involved, but completely disinterested in physics or history or whatever so have views that are consistent with how they think Hillary Clinton or whoever would behave but inconsistent with some other common sense world model in an area they never think about.
This is interesting and something I never considered in a broad sense.
I have noticed how the majority of programmers I worked with do not have a mental model of the code or what it describes – it's basically vibing without an LLM, the result accidental. This is fine and perfectly workable. You need only a fraction of devs to purposefully shape the architecture so that the rest can continue vibing.
But never have I stopped to think whether this extends to the world at large.