Simulacra Levels and the misunderstood L4

February 2024

I found the Simulacra Levels framework very useful. If you haven't read that summary, go do it, otherwise this post is not going to make much sense. A quick review of the different levels taken from that summary:

  1. Level 1: You are describing things as you see them.
  2. Level 2: You are describing things so that they have an intended effect on somebody else, regardless of whether they are true or not
  3. Level 3: You are describing things to match what a group believes and thus make yourself belong to that group.
  4. Level 4: In the posts it is described as "seems to have gone strangely meta or off the rails entirely. The symbolic representations are mostly of the associations and vibes of other symbols. The whole thing seems more like a stream of words, associations and vibes. It sounds like GPT-4.".

As you can see, L4 is described as word salad – phrases that have no connection to reality and don't need a connection. Consider what these other posts say about L4:

Level 4: This person thinks that making this statement will advance their interests in some other way, perhaps indirectly.

Note that the Level 4 actor has in an important sense lost the ability to think or plan.

But I think there is a lot more to L4 than what that suggests.

The original post has a seed of what I'll cover today in there:

  1. Finally, the pseudostructure itself becomes perceptible as an object that can be manipulated, the pseudocorrespondence breaks down, and all assertions are nothing but moves in an ever-shifting game where you’re trying to think a bit ahead of the others (for positional advantage), but not too far ahead.

But it was not clear enough for me to understand, which is why I am writing this.

You speak in L4 when you are trying to change the battlefield where arguments fight.

The battlefield where arguments fight

In different societies, different "types of facts" or different "types of arguments" resonate more or less. For example, in Chinese society, what the elders say carriers more weight than what the youth say.

If you imagine debates as a battlefield, arguments coming from the elders have the "high ground advantage". Elder arguments have an unfair advantage beyond other merits. The youth are in the low ground and likely to lose.

The shape of this battlefield is itself arbitrary. It doesn't have to be that way and in other contexts, it is not.

If you were a young person in China, you might notice this and dislike it. How would you change society to change the battlefield and reduce the elders' high ground advantage?

You would use L4 language. For example:

In the past, when we've listened to the elders we ended up in bad situations. X, Y, and Z are examples. In the future, we shouldn't listen to them anymore.

And here is another example:

We don't discriminate ideas based on whether men or women say them right? That is sexists and bad. Guess what? When we discriminate ideas based on the age of the person that says them. That is ageist and also bad.

Notice how neither statement describes the world directly. Instead we are reshaping the battleground in which future debates will play out. If this L4 speech succeeds, future arguments of the type "X is important because the elders said so" will have lower ground and be more likely to lose.


Some of the Simulacra Levels post were motivated by how complicated talking about COVID in the early days was. Here is a seemingly factual statement:

There is a virus in China and it might make it to the US

But that statement can have important implications on future arguments:

If we say that there are bad things in China, we are conceding some high ground to people that want to say that China itself is bad and that Chinese people are bad. This is xenophobic and we should never give xenophobia high ground.

This article from Yale School of Medicine spells that point of view very clearly and goes further:

DO: Talk about people "acquiring" or "contracting" COVID-19.

DO NOT: Talk about people "transmitting COVID-19," "infecting others," or "spreading the virus" as it implies intentional transmission and assigns blame. Criminalizing or dehumanizing terminology creates the impression that those with the disease have somehow done something wrong or are less human than the rest of us, feeding stigma, undermining empathy, and potentially fueling wider reluctance to seek treatment or attend screening, testing and quarantine.

Notice how this prioritizes L4 speech ("dehumanizing terminology") over L1 ("COVID is not spread by people, it is simply acquired").

Moral progress depends on L4 speech

Today we think "X is legitimate because people voted on it". But that is only because certain philosophers convinced others that "People voting on X gives X legitimacy." This wasn't the case that long ago!

Structures in level 4 speech

People try to change the discourse battlefield in many ways. I am certainly not a master of this but here are some structures I've observed.

With vibes

X arguments make me feel good. That is enough for us to elevate them

X arguments make me feel bad. That is enough for us to discount them.

At the end of the day, everything is vibes. So what follows are refinement on those vibes.

With association or equivalence

We agree that X is a good type of argument. Y is actually just like X, so Y is good too.

We agree that X is a bad type of argument. Y is like X, so Y is bad too.

With bullying

We find X to be very compelling type of argument. If you don't, you are bad and don't belong with us

Those we hate make arguments like X. If you use arguments like X, we'll hate you too.

When people use arguments that already work to belong, that is L3 speech. But when they try to change what those arguments should be in the first place (as the quotes above), then that is L4 speech.

With implication

When we use arguments like X, we reach conclusions that we already agreed we like. So, do use X arguments.

When we use arguments like X, we reach conclusions that we already agreed we don't like. So don't use X arguments.

Notice how none of the claims have "merit" in of themselves. And that is on purpose because merit itself is the thing being debated.

Some constructive L4 statements from Scott Alexander

Beware the man of one study refines the statement "We should listen to scientific evidence" with "We should listen to broad scientific evidence because individual studies are subject to all sorts of errors and statistical artifacts". And in particular, "Discount 'scientific' evidence from people that got the results they wanted to get".

Beware of isolated demands for rigor refines "X doesn't have rigorous evidence, therefore X is to be discounted" with "But disregard people that have learned to weaponize rigor as a way to discount arguments they don't like and are happy to be unrigorous for arguments they do like".

The categories were made for man, not man for the categories diminishes arguments of the form "X belongs to Y category and therefore we should do Z" by pointing out that categories are meant to serve purposes and maybe the Y category is unrelated to Z.

Through this lens, the entire rationality community can be viewed as L4 speech that is trying to elevate a certain type of argument.

Day to day L4 speech

So far, we've only considered L4 statements about Big Topics. For well-intentioned, more parochial example, consider daily uses of L4 speech inside of a company.

Stripe is a payment processor and accepts many payment methods (credit cards, bank transfers, etc.). Before 2019, Stripe internally referred to payment methods that are not cards as "Alternative Payment Methods".

When planning what to build next, Alternative Payment Methods were often deprioritized. Of course they weren't, the fact that they are secondary is in the name itself.

To fix this, an important product manager went on a crusade to rename them as Local Payment Methods. This was amazing L4 speech. They changed the internal corporate discourse so that payment methods would have higher ground in prioritization debates. It worked! After that, Stripe committed to building many more local payment methods.

But this kind of L4 speech can also be weaponized. Before performance reviews, one can start talking about "the types of contributions that are valuable", getting ahead of the actual debate. Then on the day of the debate, it turns out that the contributions you or your team happened to make are exactly the types of contributions that we recently agreed are valuable. This equivalent to the war general choosing the precise battlefield where his army would have the high ground.

Why is L4 often word salad?

I see what the original Simulacra Levels posts describe L4 as word salad – it often is. This includes the original postmodern texts from where the terms Simulacra itself comes from.

But the fact that you and I perceive it as word salad doesn't mean everybody perceives it as word salad:

It is not meant for you. They are using meta arguments that don't resonate with you. "Militarizing perestroika would be neo-Stalinist!" means nothing to me but that was Gorbachev's said to shoot down USSR reforms he didn't like. L4 strategies depend on you agreeing about certain things already (e.g. "neo-Stalinism is bad"). If you lack the necessary context, or you have it but it has no power over you, it doesn't work.

It is incompetent L4 speech. One can be incompetent at all levels of speech, including L4. For example, you can fail at L1 speech meant to describe reality by saying "The Earth is flat because when you reach Antartica you can keep going forever". Maybe what you thought was word salad was incompetent L4 speech (and therefore it merits being labeled "word salad").

We are in L4 right now

The original Simulacra Levels posts, the post I am writing, and everything we say about Simulacra Levels is itself L4 language. By exploring these concepts, we are implicitly changing what kind of future arguments we want to value. We are reinforcing that we only want to be valuing:

X is a good argument because it lets us see reality as it is and make accurate predictions about the future

And we are also changing the battlefield by saying:

Beware of speech patterns Y and Z mean to make you value other types of arguments.

If you like the idea of Simulacra Levels or this post, that simply means that it was effective L4 speech. But I bet many people that started reading this post, already closed the tab thinking "what kind of word salad was that?!".