A Mathematical Derivation of Aesthetic Vibe from Text
Based on Peli Grietzer's "A Theory of Vibe" (Glass Bead, 2017) and his December 2025 update. This module implements two computational models that mathematically derive the vibe — the ambient aesthetic unity — of a given text or author's corpus.
Table of Contents
- Quick Start
- Usage
- Theoretical Foundation
- Mathematical Framework: Theory of Vibe '17 (VAE)
- Mathematical Framework: Theory of Vibe '25 (Diffusion)
- Architecture & Pipeline
- Metrics Reference
- Interpretation Guide
- References
1. Quick Start
# Clone and set up git clone https://github.com/xraymemory/vibetheory.git cd vibetheory python3 -m venv .venv source .venv/bin/activate pip install -r requirements.txt # Analyze a text file (VAE '17 model, default) python vibetheory.py novel.txt # Analyze with Theory of Vibe '25 (diffusion/manifold model) python vibetheory.py --mode diffusion novel.txt # Analyze inline text python vibetheory.py --text "It was a dark and stormy night..." # Compare two texts (cross-reconstruction) python vibetheory.py --compare kafka.txt stein.txt # Run the built-in demo (both models on Kafka + Stein) python vibetheory.py --demo
2. Usage
CLI Options
| Flag | Type | Default | Description |
|---|---|---|---|
file |
positional | — | Path to text file |
--text, -t |
string | — | Analyze text provided directly |
--compare, -c |
2 files | — | Compare vibes of two text files |
--demo, -d |
flag | — | Run Kafka vs Stein demo |
--mode, -m |
vae / diffusion |
vae |
Model: vae (Theory '17) or diffusion ('25) |
--latent-dim |
int | 16 | Latent dimensions (VAE only) |
--epochs |
int | 500 | Training epochs |
--window-size |
int | 3 | Sentences per phenomenon window |
--canon-percentile |
float | 25.0 | Canon/manifold strictness |
--no-plot |
flag | — | Skip visualization |
--quiet, -q |
flag | — | Suppress training output |
Python API
import vibetheory # Single text analysis result = vibetheory.analyze_vibe( text="Your text here...", title="My Text", mode="diffusion", # or "vae" latent_dim=16, epochs=500, plot=True, verbose=True, ) # Access results metrics = result["metrics"] # VibeMetrics or VibeMetrics25 report = result["report"] # formatted string model = result["model"] # trained LiteraryAutoencoder or ScoreNetwork phenomena = result["phenomena"] # PhenomenaMatrix # VAE-specific canon = result["canon"] # CanonAnalysis (mode="vae" only) # Diffusion-specific manifold = result["manifold"] # ManifoldAnalysis (mode="diffusion" only) # Compare two texts comp = vibetheory.compare_vibes(text_a, text_b, title_a="A", title_b="B") print(vibetheory.format_comparison_report(comp))
Output Files
| Mode | Plot File | Contents |
|---|---|---|
| VAE '17 | vibe_analysis_17.png |
Latent space PCA, error distribution, feature signature, summary |
| Diffusion '25 | vibe_analysis_25.png |
Manifold (score norm coloring), anomaly distribution, tangent spectrum, local comparability distribution |
Example Output: VAE '17
$ python vibetheory.py --mode vae kafka.txt
======================================================================
VIBE THEORY (VAE '17) — Analyzing: Kafka - The Trial
======================================================================
[1/5] Extracting phenomena (textual windows)...
Found 23 phenomena
[2/5] Vectorizing phenomena (TF-IDF)...
Matrix shape: (23, 69)
[3/5] Training literary autoencoder (VAE '17)...
Epoch 100/500 loss=0.237761
Epoch 200/500 loss=0.137229
Epoch 300/500 loss=0.096180
Early stopping at epoch 330
Final loss: 0.085626
Effective latent dim: 5
[4/5] Deriving canon...
Canon size: 6 / 23
[5/5] Computing vibe metrics (VAE '17)...
======================================================================
VIBE ANALYSIS: Kafka - The Trial
======================================================================
Theory of Vibe '17 (VAE / global latent chart)
Based on Grietzer, 'A Theory of Vibe' (Glass Bead, 2017)
----------------------------------------------------------------------
MODEL STRUCTURE
Phenomena (textual windows): 23
Input dimensions (TF-IDF): 69
Latent dimensions (respects of variation): 5
Compression ratio: 0.0725
VIBE METRICS (global)
Vibe Density: 0.261
(fraction of text in the canon — the 'dense vibe')
Aesthetic Unity (canon): -0.058
(inter-comparability of canonical phenomena)
Full-Text Comparability: -0.027
(inter-comparability of all phenomena)
Vibe Strength: -0.031
(canon coherence minus full-text comparability)
Mean Reconstruction Error: 1.9768
Excess of Reality: 2.1996
(mean error on non-canonical phenomena)
RESPECTS OF VARIATION (latent feature signature)
Active latent dimensions: 5
Top 5 features by mean activation:
Feature 0: activation=0.736 variance=0.341
Feature 2: activation=0.677 variance=0.501
Feature 4: activation=0.638 variance=0.468
Feature 1: activation=0.614 variance=0.460
Feature 3: activation=0.390 variance=0.227
CANON (most vibe-conforming phenomena)
Canon size: 6 / 23 phenomena
Canon threshold (max error): 1.6047
Sample canonical phenomena (best 5):
[1] (err=0.9271) The stranger understood it that way all the same...
[2] (err=1.1929) "No," said the man by the window, throwing his book...
[3] (err=1.3726) Didn't Franz tell you that?" "What is it you want...
[4] (err=1.4366) "You can't leave, you are being held." "So it...
[5] (err=1.5406) But the man did not submit to K.'s gaze for very...
EXCESS OF REALITY (least vibe-conforming phenomena)
Highest-error phenomena (top 3):
[1] (err=3.1960) "And why?" he asked. "We are not authorized to...
[2] (err=2.4803) "We are not authorized to tell you that. Go to...
[3] (err=2.4169) "I want to see Frau Grubach--" said K., and made...
----------------------------------------------------------------------
INTERPRETATION (per Grietzer '17)
LOW AESTHETIC UNITY: The text's phenomena are diverse and
resist compression into a limited generative vocabulary.
This may indicate polyvalence, collage, or an expansive
'lifeworld' that resists schematization.
======================================================================
Example Output: Diffusion '25
$ python vibetheory.py --mode diffusion kafka.txt
======================================================================
VIBE THEORY (Diffusion '25) — Analyzing: Kafka - The Trial
======================================================================
[1/5] Extracting phenomena (textual windows)...
Found 23 phenomena
[2/5] Vectorizing phenomena (TF-IDF)...
Matrix shape: (23, 69)
[3/5] Training score network (Diffusion '25)...
Epoch 100/500 loss=1.018665
Early stopping at epoch 141
Final loss: 0.927987
[4/5] Analyzing implicit manifold...
Computing score norms (anomaly detection)...
Building neighborhood graph...
Computing local tangent spaces (score Jacobian)...
On-manifold: 17 / 23
[5/5] Computing vibe metrics (Diffusion '25)...
======================================================================
VIBE ANALYSIS: Kafka - The Trial
======================================================================
Theory of Vibe '25 (diffusion / implicit manifold)
Based on Grietzer, 'Theory of Vibe' update (Dec 2025)
NOTE: All metrics below are our operationalization of
Grietzer's conceptual framework (his essay has no equations).
----------------------------------------------------------------------
IMPLICIT MANIFOLD STRUCTURE
Phenomena (textual windows): 23
Input dimensions (TF-IDF): 69
MANIFOLD MEMBERSHIP (analog of 'canon')
On-manifold fraction: 0.739
(fraction of phenomena lying on the learned manifold)
Mean anomaly score: 1.4527
(lower = more on-manifold; score norm at sigma_min)
LOCAL MEANING (the '25 innovation)
Mean local comparability: 0.414
(how meaningful comparisons are within neighborhoods)
Local comparability variability: 0.126
(how uniform local meaning is across the manifold)
Global comparability: -0.044
(pairwise similarity in raw input space, for reference)
TANGENT SPECTRUM (local generative vocabulary)
Top 8 tangent singular values (mean across phenomena):
Tangent dir 0: 0.8814
Tangent dir 1: 0.7729
Tangent dir 2: 0.6526
Tangent dir 3: 0.5908
Tangent dir 4: 0.5327
Tangent dir 5: 0.4860
Tangent dir 6: 0.4394
Tangent dir 7: 0.4045
ON-MANIFOLD PHENOMENA (most vibe-conforming)
[1] (score=1.178, local_comp=0.518)
But the man did not submit to K.'s gaze for very long...
[2] (score=1.271, local_comp=0.346)
Through the open window he could again see the old woman...
[3] (score=1.286, local_comp=0.342)
Didn't Franz tell you that?" "What is it you want, then?"...
OFF-MANIFOLD PHENOMENA (highest anomaly)
[1] (score=1.838) At once there was a knock at the door and a man...
[2] (score=1.677) That had never happened before. waited a while...
[3] (score=1.658) He was slim yet solidly built, and he wore a closely...
----------------------------------------------------------------------
INTERPRETATION (per Grietzer '25)
STRONG MANIFOLD: Most phenomena lie on the learned manifold.
The text has a coherent implicit structure — a 'surface' in
phenomenon-space that organizes its diversity.
HIGH LOCAL COMPARABILITY: Neighborhoods on the manifold
are semantically structured. Per Grietzer '25, the vibe
enables 'meaningful bases of comparison in neighborhoods
of V-possible items.'
======================================================================
Example Output: Comparison Mode
$ python vibetheory.py --compare kafka.txt stein.txt --quiet
======================================================================
VIBE COMPARISON
======================================================================
'Kafka - The Trial' vs 'Stein - Tender Buttons'
----------------------------------------------------------------------
Metric A B
----------------------------------- ------------ ------------
Aesthetic Unity 0.061 -0.026
Vibe Density 0.261 0.286
Vibe Strength 0.091 -0.015
Mean Reconstruction Error 1.7893 1.6592
Excess of Reality 1.9999 1.9079
CROSS-RECONSTRUCTION
A's model reconstructing B: error = 1.4727
B's model reconstructing A: error = 1.3953
Vibe Similarity: 1.000
INTERPRETATION: These texts share a strong vibe affinity.
Each model can substantially reconstruct the other's phenomena,
suggesting overlapping 'respects of variation.'
======================================================================
Visualizations
Each analysis generates a 4-panel diagnostic plot.
Theory of Vibe '17 (VAE)
Latent space PCA, reconstruction error distribution, feature activation signature, summary metrics:
Theory of Vibe '25 (Diffusion)
Implicit manifold (score norm coloring), anomaly distribution, tangent spectrum, local comparability distribution:
3. Theoretical Foundation
What Is a Vibe?
Grietzer's central thesis is that a vibe is the aesthetic unity of a set of phenomena that can be compressed into a limited generative vocabulary. When we sense a "Kafkaesque" quality in a bureaucratic encounter, or a "Steinian" quality in a pattern of domestic objects, we are sensing that these diverse phenomena can be approximately reconstructed from a small shared set of structural principles.
More precisely, a vibe is an abstractum that cannot be separated from its concreta — you cannot paraphrase or extract the vibe from the specific phenomena that embody it, yet it is undeniably there as a structural property of the collection.
Grietzer draws an analogy to 64k Intros from demoscene culture: vast, lush visual worlds that fit into 64 kilobytes by exploiting deep internal self-similarity. The objects in a 64k Intro are individually complex but collectively simple — the same principle that defines a vibe.
Two Eras of the Theory
Theory of Vibe '17 models vibe through a variational autoencoder (VAE): a neural network that compresses phenomena into a low-dimensional latent space. The canon — the set of phenomena with zero reconstruction error — is the idealized dense core of the vibe. The meaning of any phenomenon is its position in a single global Euclidean chart.
Theory of Vibe '25 acknowledges that VAEs fail on "surface-diverse but vibe-coherent" datasets (like "all photos taken on the streets of modern Paris"). Modern diffusion models learn an implicit lower-dimensional manifold without a global chart. Meaning becomes local: each phenomenon gets semantic structure through its neighborhood on the manifold, not through a global coordinate. The vibe enables "meaningful bases of comparison in neighborhoods of V-possible items" (Grietzer '25).
4. Mathematical Framework: Theory of Vibe '17 (VAE)
4.1 Phenomena as Input Space
A text is decomposed into phenomena — overlapping sliding windows of
Each phenomenon is embedded into a high-dimensional input space
followed by standard normalization (zero mean, unit variance per feature):
The resulting matrix
4.2 The Literary Autoencoder
The autoencoder consists of two functions:
Feature function (encoder)
The
Decoder function (generative grammar)
Both networks use LayerNorm and GELU activations with dropout regularization.
4.3 Training Objective
The autoencoder is trained to minimize the sum of reconstruction error and a sparsity penalty on the latent codes:
where
Optimization uses AdamW with weight decay
The effective latent dimensionality is automatically scaled:
4.4 Canon Derivation
The canon
Per-phenomenon reconstruction error:
The canon is defined by a percentile threshold
Per Grietzer: "The canon is the set of all the objects that a given trained autoencoder can imagine or conceive of whole, without approximation or simplification."
The non-canonical phenomena — those with high reconstruction error — represent what Grietzer calls the excess of material reality over the gestalt-systemic logic of autoencoding.
4.5 Vibe Metrics (Global)
Vibe Density
The fraction of phenomena lying within the canon:
Higher density means the generative vocabulary covers more of the text.
Aesthetic Unity
The mean pairwise cosine similarity among canonical phenomena in the latent space:
This measures what Grietzer calls comparability (section 18): "the global intercomparability of the manifold's objects is not just technically but conceptually and phenomenologically inseparable from the representation of a system."
Full-Text Comparability
The same cosine similarity measure but over all phenomena (not just canon):
Vibe Strength
The degree to which the canon is more coherent than the full text:
Positive
Kolmogorov⁺ Complexity Proxy
The compression ratio approximates how much the generative vocabulary compresses the input space:
Per Grietzer (section 8–9): the Kolmogorov⁺ complexity of an object relative to a schema measures the "depth of pattern." A low ratio means the schema achieves strong compression — the text's phenomena are "individually complex but collectively simple."
Excess of Reality
The mean reconstruction error on non-canonical phenomena:
This quantifies what the generative vocabulary cannot capture — the material reality that exceeds the gestalt.
Feature Activation Signature
The mean absolute activation of each latent dimension across canonical phenomena:
and the variance:
High activation with high variance indicates a latent dimension that is actively differentiating phenomena — a semantically loaded "respect of variation."
4.6 Cross-Reconstruction (Comparison)
Per Grietzer (section 11): "a trained autoencoder and its canon are effectively mathematically equivalent." We exploit this to compare the vibes of two texts
Text
The vibe similarity measures how much worse cross-reconstruction is compared to self-reconstruction:
When similarity
5. Mathematical Framework: Theory of Vibe '25 (Diffusion)
Provenance note: Grietzer's "Theory of Vibe '25" (Substack, Dec 2025) is a short conceptual essay containing no equations, theorems, or formal definitions. He argues that diffusion models learn an implicit manifold with local (not global) semantic structure, citing three papers by abbreviated author-year only. All equations in this section are our operationalization of his conceptual framework, using standard techniques from diffusion model theory. We are explicit below about what comes from where.
5.1 Motivation: Why VAEs Fail
Per Grietzer (December 2025):
"VAEs aren't really good with natural datasets that have a lot of variation even at the surface level... 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."
The '25 model replaces the explicit bottleneck with an implicit manifold learned by a score-based denoising network. The manifold gives meaning locally — through tangent spaces and neighborhood structure — rather than through a global coordinate system.
5.2 The Score Function
Standard diffusion model theory (Song & Ermon, 2019; not from Grietzer).
Given the data distribution
At a noise level
and the noised score is
5.3 Score Network Architecture
Our implementation choice; not from any cited paper.
A time-conditional neural network
The noise level
The score prediction concatenates the input with the time embedding and passes through the main network:
where Net is a 4-layer MLP with LayerNorm, GELU activations, and dropout.
5.4 Denoising Score Matching
Standard technique (Vincent, 2011; not from Grietzer).
Training uses the denoising score matching objective. For a clean data point
The true score at
The training loss with
where the expectations are over
5.5 The Implicit Manifold
Grietzer '25 writes that "generalization-capable diffusion models behave as if they were projecting the training data toward a lower-dimensional manifold in input space," citing a paper he calls "Farghly et al. '25." The actual paper is Wang et al. '24 (arXiv:2409.02426), which proves — under a mixture of low-rank Gaussians (MoLRG) assumption with a specific DAE parameterization — that the learned denoiser acts as a scaled orthogonal projection onto low-dimensional subspaces, and that training is equivalent to PCA / subspace clustering. Their theoretical guarantees hold for their specific parameterization; the empirical evidence on real architectures (U-Net on CIFAR-10, CelebA, etc.) is suggestive but not proven.
The following anomaly detection construction is ours, not from Grietzer or Wang et al.:
Anomaly detection via score norm:
Low
The on-manifold set (analogous to the '17 canon):
5.6 Local Tangent Spaces via the Score Jacobian
This construction is ours. See provenance note below.
We extract local semantic structure from the Jacobian of the score function at each data point. The score Jacobian at point
computed column-by-column via automatic differentiation. Taking the SVD:
where
The rows of
Provenance note on the Jacobian: Park et al. '23 (NeurIPS) compute the Jacobian of an intermediate U-Net activation (an encoder Jacobian defining a pullback Riemannian metric), not the score Jacobian. Kwon et al. '23 (ICLR) show that U-Net bottleneck activations are semantically meaningful but do not compute any Jacobians. Wang et al. '24 analyze the DAE Jacobian (predicted
$\mathbf{x}_0$ ), not the score. We use the score Jacobian because our small MLP score network has no intermediate activations to extract. The score Jacobian (related to the Hessian of$\log p$ ) does encode local geometric information, but it is a mathematically different object from what Park et al. compute. Grietzer '25 mentions "learning local tangent spaces around data points" but provides no mathematical specification of how.
5.7 Local Comparability (Neighborhood Semantics)
Our operationalization of Grietzer's '25 concept: "V enables the construction of meaningful bases of comparison in neighborhoods of V-possible items."
For each phenomenon
Project into the local tangent space using the top
Compute pairwise cosine similarity among projected neighbors:
The absolute value ensures we measure alignment of directions of variation, not just correlated/anti-correlated positioning.
The global mean local comparability:
5.8 Tangent Spectrum (Local Generative Vocabulary)
The global tangent spectrum aggregates the singular value profiles across all phenomena:
This is the average "strength" of each tangent direction — a summary of the local generative vocabulary's structure across the text.
6. Architecture & Pipeline
6.1 Full Pipeline
Theory of Vibe '17 (VAE)
========================
Raw Text
|
v
[1] Sentence splitting + sliding windows (w=3, stride=1)
|
v
[2] TF-IDF vectorization (ngrams 1-2, sublinear TF, max 512 features)
+ StandardScaler normalization
|
v
X in R^{N x D} (phenomena matrix)
|
v
[3] Train LiteraryAutoencoder
| encoder: D -> hidden -> hidden/2 -> d (tanh)
| decoder: d -> hidden/2 -> hidden -> D
| loss: MSE + 0.01 * L1(z)
|
v
f_theta, g_phi (feature function + decoder function)
|
v
[4] Derive canon: {p_i : ||x_i - g(f(x_i))|| <= threshold}
|
v
[5] Compute vibe metrics:
vibe density, aesthetic unity, comparability,
vibe strength, excess of reality, feature signature
Theory of Vibe '25 (Diffusion)
===============================
Raw Text
|
v
[1] Sentence splitting + sliding windows (w=3, stride=1)
|
v
[2] TF-IDF vectorization + StandardScaler normalization
|
v
X in R^{N x D} (phenomena matrix)
|
v
[3] Train ScoreNetwork via denoising score matching
| score_net: (x, sigma) -> s_psi(x, sigma) in R^D
| loss: E[sigma^2 * ||s_psi(x+sigma*eps, sigma) + eps/sigma||^2]
|
v
s_psi (learned score function)
|
v
[4] Analyze implicit manifold:
| a. Score norms -> anomaly detection -> on-manifold set
| b. k-nearest neighbors in input space
| c. Score Jacobian at each point -> SVD -> local tangent spaces
| d. Project neighbors into tangent space -> local comparability
|
v
[5] Compute vibe metrics:
on-manifold fraction, mean local comparability,
tangent spectrum, global comparability (reference)
6.2 Neural Network Architectures
LiteraryAutoencoder (VAE '17)
Encoder: Decoder:
Linear(D, H) Linear(d, H/2)
LayerNorm(H) LayerNorm(H/2)
GELU GELU
Dropout(0.1) Dropout(0.1)
Linear(H, H/2) Linear(H/2, H)
LayerNorm(H/2) LayerNorm(H)
GELU GELU
Linear(H/2, d) Linear(H, D)
Tanh
ScoreNetwork (Diffusion '25)
Time Embedding: Score Network:
Linear(1, d_t) Linear(D + d_t, H)
GELU LayerNorm(H)
Linear(d_t, d_t) GELU
Dropout(0.05)
Linear(H, H)
LayerNorm(H)
GELU
Linear(H, H/2)
LayerNorm(H/2)
GELU
Linear(H/2, D)
6.3 Hyperparameter Scaling
Both models automatically scale their architecture to the data:
| Parameter | VAE '17 | Diffusion '25 |
|---|---|---|
| Effective latent dim | min(d, max(4, N/4), D/4) | N/A (implicit) |
| Effective hidden dim | min(H, D) | min(H, 2D) |
| Batch size | min(32, N) | min(32, N) |
| TF-IDF features | min(512, 3N) | min(512, 3N) |
7. Metrics Reference
7.1 Theory of Vibe '17 Metrics
| Metric | Symbol | Description | Range |
|---|---|---|---|
| Vibe Density | Fraction of text in the canon | [0, 1] | |
| Aesthetic Unity | Mean pairwise cosine sim in latent space (canon) | [-1, 1] | |
| Full Comparability | Mean pairwise cosine sim in latent space (all) | [-1, 1] | |
| Vibe Strength | Canon coherence above baseline: |
[-2, 2] | |
| Complexity Ratio | Compression ratio: |
(0, 1) | |
| Mean Recon Error | Mean L2 reconstruction error | [0, inf) | |
| Excess of Reality | Mean error on non-canonical phenomena | [0, inf) | |
| Feature Activation | Mean abs activation per latent dim (canon) | [0, 1] | |
| Feature Variance | Variance of latent codes per dim (canon) | [0, inf) |
7.2 Theory of Vibe '25 Metrics
All metrics below are our operationalization (see Section 5 provenance notes).
| Metric | Symbol | Description | Range |
|---|---|---|---|
| On-Manifold Fraction | — | Fraction of phenomena on the manifold | [0, 1] |
| Mean Anomaly Score | Mean score norm (lower = more on-manifold) | [0, inf) | |
| Local Comparability | Mean neighborhood cosine sim (tangent-projected) | [0, 1] | |
| Local Comp Variability | — | Std of local comparability across points | [0, inf) |
| Global Comparability | Pairwise cosine sim in raw input space (reference) | [-1, 1] | |
| Tangent Spectrum | Mean singular values across all points | [0, inf) |
8. Interpretation Guide
8.1 Reading the VAE '17 Report
High aesthetic unity (
Moderate aesthetic unity (
Low aesthetic unity (
Strong vibe strength (
Diffuse vibe (
8.2 Reading the Diffusion '25 Report
Strong manifold (on-manifold > 70%): Most phenomena lie on the learned manifold. The text has a coherent implicit structure.
High local comparability (
Moderate local comparability (
Low local comparability (
9. References
- Grietzer, Peli. "A Theory of Vibe." Glass Bead Site 1: Logic Gate (2017). glass-bead.org/article/a-theory-of-vibe
- Grietzer, Peli. "Theory of Vibe '25." Second Balcony (Substack), December 4, 2025. peligrietzer.substack.com/p/theory-of-vibe-25
- Wang, P., Zhang, H., Zhang, Z., Chen, S., Ma, Y., & Qu, Q. "Diffusion Models Learn Low-Dimensional Distributions via Subspace Clustering." arXiv:2409.02426 (2024).
- Kwon, M., Jeong, J., & Uh, Y. "Diffusion Models Already Have a Semantic Latent Space." ICLR (2023).
- Park, Y.-H., Kwon, M., Choi, J., Jo, J., & Uh, Y. "Understanding the Latent Space of Diffusion Models through the Lens of Riemannian Geometry." NeurIPS (2023). arXiv:2307.12868
- Ngai, Sianne. Ugly Feelings. Cambridge, MA: Harvard University Press (2005).
- Flatley, Jonathan. Affective Mapping: Melancholia and the Politics of Modernism. Cambridge, MA: Harvard University Press (2008).
- Lewis, David. "New Work for a Theory of Universals." Australasian Journal of Philosophy 61, no. 4 (1983): 343-377.
- Heidegger, Martin. The Fundamental Concepts of Metaphysics: World, Finitude, Solitude. Trans. William McNeill and Nicholas Walker. Bloomington: Indiana University Press (1995).
- Batuman, Elif. "What Am I Doing Here?" The Guardian, April 25, 2008. theguardian.com/books/2008/apr/26/popandrock
"A vibe is an abstractum that cannot be separated from its concreta." — Peli Grietzer