Update:

So the reason I thought this is a joke is that Erika is just walking through Theory of Vibe step by step with a lot of cursing, going 'but wtf is x??!' or 'how tf can we define y?!!?!' and then assuming the actual definition used in Theory of Vibe in the next section.

And then there's stuff like this which I thought has to be a joke because it replicates section 19 of Theory of Vibe but says 'Theory of Vibe doesn't address this’:

By the time Erika ends she's given everything the formal or informal definition used in the PhD, calling it her own 'careful and nuanced reformulation,' and says the REAL PROBLEM is that Theory of Vibe is an account of a natural dataset’s unity rather than a clustering method:

Which again seemed like it has to be a joke because Erika makes some footnotes crack about ppl 'trying to do AI phenomenology,' so she does have the concept of trying to do phenomenology.

And while there’s no non-comedy evidence that Erika parsed Theory of Vibe sections 9, 10, and 15, it was hard not to assume ‘what could the canon of the Kafkaesque possibly be?!’ is a joke.

The one part that didn’t seem like a joke but like regular confusion is Erika refuting 'every autoencoder canon is the vibe of a worldly domain' while Theory of Vibe (at its most overstated, toy form possible) is actually 'every vibe of a worldly domain is an autoencoder canon'.

Last time I looked Erika responded to a twitter version of these remarks with some mixture of ‘the real problem is that trained autoencoders are so design-dependent that no trained autoencoder can be said to demonstrate any non-trivial properties of its training data’ and ‘the real problem is that you can’t prove that Kafka’s work is mathematically the canon of the Kafkaesque.’ I can’t find the conceptual art or comedy in making this pivot, so I have to conclude Erika’s been in the what-Harry-Frankfurt-called-’bullshit’ business the whole time.

What I think actually happened is Erika looked at Theory of Vibe, thought it makes no sense and would be fun to take down, then ended up making sense of it (=the parts she read) just fine in the process of writing the takedown but went ‘eh whatever’ and pressed ‘post’.

Back in 2015 I was very excited by the idea that what classic bottleneck VAEs trained on some natural dataset learn is a parametrized manifold with multiple “cognitive” functions/interpretations. A classic VAE learns, all at once: an algorithmic compression of approximate reconstructions of a dataset; a dimensionality reduction of the dataset, a global Euclidean chart you can use to meaningfully compare items in the dataset through their difference vector; a generative model of the data distribution; a (weak) decomposition into semantic features; and a (weak) anomaly-detection system.

My PhD thesis (comp lit with an assist from mathematician Tomer Schlank) proposed that these are also properties of the kind of “vibe-knowledge” of the world we live in we produce/express through art and poetry. We learn a very compressible approximate reconstruction of the world we live in, and a latent space and associated generative model that has myriad representation and search and anomaly-detection uses and gives meaning to comparing and contrasting “in-vibe” items. On this VAE-inspired account of vibe, given a vibe 𝑉 the “meaning” of an individual item in 𝑉 is its position in the space of 𝑉-possible items.

Does this hold up in 2025? Not without changes, but I think that on reflection these ten-years-in changes are conceptually robust and not just technical hotfixes.

The problem, from a modern ML POV, is that VAEs aren’t really good with natural datasets that have a lot of variation even at the surface level (pose, angle, scale, lighting, etc.) You can train a good VAE on a dataset of images of faces, but it’s harder to get something worthwhile on a dataset of (say) images of people unrestricted by pose, angle, and size. And VAEs are downright hopeless on something truly surface-diverse (but paradigmatically vibe-coherent) like “the set of all pictures taken in the street in modern Paris.”

You could say “maybe one day we’ll make VAEs work for these surface-diverse, vibe-coherent datasets,” but generative modeling of these kinds of datasets is basically a solved problem in 2025 -- it’s just not VAEs that do it. Modern diffusion and flow-matching models are extremely good at density estimation on surface-diverse natural datasets, but do they also learn a mathematical structure with the right multiplicity of “cognitive” functions/interpretations for Theory of Vibe? I think the answer’s ‘yes’! But with some important modifications to Theory of Vibe that are independently conceptually healthy.

The way I think about diffusion models currently (based on the recent wave of geometric accounts) is that diffusion models learn an implicit lower-dimensional manifold in input space but don’t learn a lower-dimensional Euclidean chart for it. (See for example Farghly et al. ‘25 on how generalization-capable diffusion models behave as if they were projecting the training data toward a lower-dimensional manifold in input space.) And I think this is partly why diffusion models work where VAEs don’t: a complex natural dataset like “the set of all pictures taken in the street in modern Paris” probably doesn’t have an underlying semantics that fits a single global Euclidean chart of reasonable complexity.

That said, I’ve been seeing more and more work from the last couple of years that suggests the implicit lower-dimensional manifold learned by diffusion models is semantically meaningful in a more indirect way. Per recent work (e.g. Kwon et al. ‘23, Kim et al. ‘25), taking the intermediate activations in a diffusion model as a latent space enables meaningfully training sparse autoencoders, or learning local tangent spaces around data points, or doing local PCA. In other words, you can use the internals of diffusion models to learn local lower-dimensional Euclidean charts covering the implicit lower-dimensional manifold in input space, and extract good “local” semantic feature vectors for the neighborhood of each data point.

So what happens if I start thinking about “vibe” along the lines of the implicit manifold of a diffusion model instead of the manifold of a VAE? We get a Theory of Vibe ‘25: vibe is a manifold that has generative, compressive, and weak anomaly-detection functions but gives individual data points “meaning” in a more local and indirect way than a classic VAE does. Instead of a vibe 𝑉 directly grounding meaningful global comparisons between any two 𝑉-possible items, 𝑉 enables the construction of meaningful bases of comparison in neighborhoods of 𝑉-possible items.

This is, I think, a much more natural account than Theory of Vibe ‘17 of how “vibe” -- and the kind of world-making we get from art and poetry -- is an enabling condition for more determinate cognitive operations. And it produces of its own accord a (very Kantian) contrast between this enabling vibe/determinate-cognition relationship and the resistance of vibe to global expression by determinate cognition!

1. For the past ten years, I’ve been saying that my second book is going to be 'something about Bowie, dialectics, BPD, romantic irony, thrownness, autism, social reality'. I haven’t written a first book.

2. I had, at one time, got a serious Borderline Personality Disorder diagnosis from a very thoughtful psychiatrist/analyst. This confused everyone since I am mostly pleasant in a crisis and amicable in disappointment and frustration but it made sense in specific ways, and I developed the lay theory that my pathological deal is that my emotions are insane but never have the power to convince me about much. My emotions are vicious rumors, muttering gossip, hissing vipers that can break my body but can’t tell me what anything is. My emotions can break my body but can’t have my spirit. Spirit, so to speak, floats over the waters of the void like a powerless god.

3. Heidegger has this idea of ‘thrownness’ – that we experience, right in the matter of sensation, a world of unchosen meanings and stakes and commitments. I don’t think I’ve ever had thrownness, and I can’t tell whether this is a personal quirk or neural atypicality or rational response to growing up in capitalism or in a scientifically disenchanted world or in a childhood. I always had a lot of feelings, but knew early on that as a child I don’t participate in constructing the premises or language of the day-to-day life structuring and activating and attuning and interpreting my feelings, and knew later on that as a neural evolved ape brain-chemistry cluster of particles whatever electricity my feelings are a mix of misplaced messages about irrelevant evolutionary whatever and electric or chemical or whatnot noise, then later on knew the capitalism version of the childhood thing, and all this knowing kept my feelings from ever attaining that translucent, risen form where they are not just feelings but a presentation of meanings or stakes or commitments or a world.

4. All through my teenage years I thought the central fact of human life is that the deeper and more beautiful the feeling, the more its reality-claim is a pantomime. Somewhere between the crisp taste of an apple saying healthful nurture and the nausea of a crush saying the godhead of the Good, the Beautiful, and True goes to your high-school, feeling’s claim on reason turns vestigial. Life’s lesson, again and again, was that any sufficiently advanced experience is indistinguishable from fantasy – what goes for Narnia and Middle-earth at twelve goes for a crush at fifteen, and goes on for every feeling born of ecstasy or art or drugs or daydreaming.

5. There’s something we might call the cunning of art: Art takes you in by promising that what our feelings lack in world-making authority they can make up for in raw force of being, shifting life’s gravity to the subjective. Art gives the would-be meanings, would-be stakes, would-be commitments exposed as feelings a second-order, over-their-head meaning that subsists in their contingent, textured, patterned, causal hijinks as flora and fauna of the soul. At the same time, because this meaning that art introduces – call it beauty – is itself something felt, art also demonstrates in its own workings the idea of feeling that really is meaning-grasping, stakes-grasping, commitment-grasping, world-grasping.

6. Everything Bowie wrote after 1971 is a monologue of a stage of spirit cognizing that it has transcendental conditions (ways the world has to be or has to let you assume that it is in order for your language to keep holding together as a language) that it didn’t see when it got started and are not going to pan out. Life On Mars is different, a before-life song in a corpus where almost every song is a death song. A song to those who aren’t yet persons, and to those who came undone and now again aren’t yet persons but are looking for an in.

7. All feelings want to be believed: Pain screams about bodily harm. Contempt preaches superiority. Excitement chatters that good times will roll. Anxiety foretells the world withdrawing. Despair tells a sickness unto death. Beauty says that something about this life or mind or language is a sacred mystery, known and unknown, a ground and a horizon, structured and ineffable, beyond all words and infinitely speakable. And, past all shame, it says that if the glass between feeling and thought can break this once, for art, maybe it can break always and for everything.

8. Is there life on Mars?