Pith. sign in

REVIEW 3 major objections 5 minor 60 references

What Iterated Self-Feeding Probes of Language Models Measure, and a test that separates the construction from the model

T0 review · 3 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Iterated self-feeding probes of language models read probe and model at once; a variation test separates which is which.

desk verdict A self-critical methods paper that earns its central claim: CRN coupling plus a variation test cleanly separates construction-determined from model-determined readings, and the reproducible artifacts make the soft spots fixable. read the letter →

arxiv 2608.10986 v1 pith:2PKQ5FBO submitted 2026-08-11 cs.CL cs.LG

Nicolás Vera Zúñiga This is my paper · ORCID
classification cs.CLcs.LG
keywords self-feedingprobesdamagespreadingcommonrandomnumbersLyapunovexponentlanguagemodeldynamicsGlauberdirectedpercolationattribution
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Self-feeding a language model its own output is now routine, but what such a probe measures has stayed unclear. This paper argues that an iterated probe reads the model and the probe at once, in quantities that carry the same units and apparent precision but answer to different causes. To make the question sharp it builds a ring of token cells resampled in place by the model's own windowed conditional, and couples two rings that differ in one token under common random numbers, making damage spreading measurable. The central result is a discriminator: hold the construction fixed and vary the model, or hold the model fixed and vary the construction, and see which readings move. Applying it, the paper shows a phase transition it measured to three decimal places belongs to the probe, while a developmental transition in the same instrument belongs to the model.

What carries the argument

The load-bearing object is the common-random-numbers coupling of two rings. Each ring is a set of $N$ token cells resampled in place from the model's windowed conditional $p_r(x_i \mid x_{i\pm r})$ at temperature $T$; two rings are identical except for one flipped token and are advanced with the same stream of uniform variates and the same visit order. Because inverse-CDF sampling against a shared uniform makes identical windows draw identical tokens, undamaged twins diverge by exactly zero, so every difference is attributable to the injected flip. That exact-zero null turns the construction into a damage-spreading instrument: the exponential growth rate of the damage gives the Lyapunov exponent $\lambda_{\mathrm{ca}}$, its saturation gives a damping length, and its spatial extent gives a light cone. The discriminator of Section 5 uses the same idea one level up: vary the construction with the model fixed, or the model with the construction fixed, and classify each reading by whether it moves.

What would settle it

Advance two undamaged rings under common random numbers on a backend with nondeterministic or approximate sampling; if their divergence is ever nonzero, the exact-zero null fails and all damage-spreading readings are contaminated.

Watch

Extended reading notes

Core claim

The paper's central claim is that iterated self-feeding probes do not have a single object: they mix construction-determined quantities with model-determined ones in readings that look alike and cannot be told apart by inspection. In the ring construction, the damage light cone is kinematic, its extent set by the update window rather than by the model, and the radius scaling of the token-space Lyapunov exponent $\lambda_{\mathrm{ca}}(r)$ is model-invariant across 19 models and two scale ladders spanning 70×. Other readings genuinely track the model: $\lambda_{\mathrm{ca}}$ crosses zero at a reproducible point in training, and the attractor share ranks models consistently however the lattice is built. The paper's test for separating the two kinds is a manipulation: vary one factor with the other held fixed and watch which readings move. By that test, a sharply measured absorbing-state phase transition with directed-percolation exponents at a common $T_c \in [0.4343, 0.4391]$ is a property of the probe, not of any language model, while the training-checkpoint crossing is the model. The instrument is validated by reproduction first: on the Domany–Kinzel automaton the common-random-numbers damage field matches an independent prediction bit-exactly.

Load-bearing premise

The exact-zero null, that two rings with identical windows handed the same random numbers draw exactly the same token, is load-bearing; if a backend has any nondeterminism or approximate sampling, undamaged twins will diverge and differences can no longer be attributed solely to the injected flip.

Editorial extensions

If this is right

  • Every reading taken from a self-feeding loop, such as agreement rates, convergence speeds, or self-correction success, must be classified by the variation test before it is attributed to the model.
  • A sharp, reproducible, precisely measured transition in an iterated probe can be entirely a property of the probe; precision and reproducibility do not imply that the model is the source.
  • $\lambda_{\mathrm{ca}}$ is a developmental quantity: within one model's training it crosses zero reproducibly, but across models it does not rank them consistently.
  • The attractor share is the model-determined reading that survives construction variation, ranking models consistently across lattice constructions.
  • Estimator defects of a recognizable class, including saturated predictors, noise denominators, and undefined values treated as zero, can produce confident wrong numbers, so estimators must be gated at their own geometry.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the exact-zero null is backend-dependent, every damage-spreading reading inherits that dependence: moving to approximate or nondeterministic sampling would require re-validating all conclusions, not just the null.
  • The same two-way variation test could be applied to existing self-feeding practices such as self-consistency, iterated refinement, and agentic loops, deciding which of their readouts are construction artifacts.
  • The bit-exact Domany–Kinzel reproduction suggests a general template for model-dynamics instruments: calibrate on a system with a known answer before reading anything off a language model.
  • One testable extrapolation is that the diversity–$\lambda_{\mathrm{ca}}$ relation, which the paper finds holds across temperatures and checkpoints, might serve as a cheap proxy for training progress in model families without public checkpoints; that is an editorial extension, not the paper's claim.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper studies iterated self-feeding probes of language models by constructing a ring of token cells resampled in place from a windowed conditional, with two nearly identical rings advanced under common random numbers. It claims that the exact-zero null makes damage spreading measurable, that a sharp low-temperature absorbing-state transition is a property of the probe rather than of any language model, and that a variation test can separate construction-determined from model-determined readings. The instrument is calibrated by bit-exact reproduction on the Domany-Kinzel automaton, and the paper reports several retracted verdicts and estimator-gating guards. All code and per-run result files are released.

Significance. If the main claims hold, the paper is a valuable methodological contribution: it demonstrates that a sharp, reproducible phase transition can be an artifact of the probe, and it offers a simple, transferable test for classifying readings of self-feeding loops. The bit-exact Domany-Kinzel reproduction, the explicit retractions, and the reproducible package are concrete strengths that support the paper's central message. The universal wording in places goes beyond the finite evidence, but the discrimination test itself does not depend on that universalization.

major comments (3)
  1. [§2, Eq. (1)] The symmetric-window conditional in Eq. (1) is well-defined for masked language models, but no mechanism is given by which autoregressive models (Pythia, GPT-2, Granite) provide p_r(x_i | x_{i−r}, …, x_{i−1}, x_{i+1}, …, x_{i+r}) with the centre masked. A causal LM computes only left-context probabilities, so the implementation must use some approximation such as a bidirectional attention variant, a permutation scheme, an infilling head, or a heuristic combination. This is not a reproducibility footnote: if the autoregressive numbers come from a left-context-only or heuristic computation, then λ_ca, the developmental transition, and the attractor-share results are measurements of that approximation, not of the stated construction. The paper should specify the exact computation or identify the precise code path before the discriminator demonstrations can be taken as evidence about the construction.
  2. [§2, exact-zero null] The exact-zero null is stated as holding on every backend because inverse-CDF sampling against a shared uniform makes agreement exact. That guarantee is not valid for every possible backend: nondeterministic sampling kernels, approximate sampling, or thread-level RNG divergence would break exact agreement. The bit-exact Domany-Kinzel reproduction in Rung 1 validates the null on the used backend, so the central measurements survive, but the universal claim should be scoped to the validated backend and to the code path that implements the shared uniform stream.
  3. [§5.1] The term 'model-invariant' for λ_ca(r) is stronger than the evidence supports. The demonstration covers 19 models and two scale ladders, which is a substantial sample, but it does not establish invariance across models in general, especially because §5.2 reports a cross-model spread of 0.051 against a range of 0.122–0.804 produced by varying radius and temperature alone. I recommend saying 'invariant across the tested models and constructions' and giving a quantitative criterion for when a spread counts as 'not moving' relative to the construction-driven range.
minor comments (5)
  1. [Abstract] The notation p_r(x_i | x_{i+-r}) in the abstract is nonstandard and ambiguous; the intended symmetric window appears in Eq. (1), and the abstract should use the same notation.
  2. [Figure 3 caption] The sentence 'Monochrome by marker and dash throughout' is unclear; please state explicitly how the five arms are distinguished if color is not used.
  3. [§7] The symbol T* and the phrase 'greedy degeneration' are used without definition, which makes the predictive-result paragraph hard to evaluate; please define both before the claims that rely on them.
  4. [§5.2] The sentence 'its model ordering is seed-stable at 0.848 against 0.030' is ambiguous; please specify whether 0.848 is a correlation, an agreement rate, or another statistic, and state the comparison being made.
  5. [§5.2] The paper says 'ten models spanning six families and four architecture classes' but does not list them; a small table or appendix listing the models, families, and parameter counts would make the invariance and ordering claims easier to check.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central claims are empirical, calibrated against independent known-answer systems, and the one definitional null is not fitted to any target reading.

full rationale

The paper's central derivation—that iterated self-feeding probes mix construction-determined and model-determined readings, and that the Section 5 variation test separates them—is not circular. The exact-zero null is definitional within the implementation (inverse-CDF sampling against a shared uniform makes identical windows draw identical tokens), not a parameter fitted to the readings it later explains. The damage light cone and the model-invariant radius scaling of λca are derived consequences of the update rule and are explicitly classified as construction-determined rather than presented as model predictions. The model-determined readings (the training-checkpoint crossing of λca and the attractor-share ordering) are empirical results, and the instrument is independently calibrated: the Domany–Kinzel damage field is reproduced bit-exactly with zero mismatching cells against an independent prediction, elementary CA rules recover a known ordered/non-ordered boundary, and known transition matrices are recovered by the attractor census. The paper also explicitly identifies and retracts its own earlier manufactured phase transition, which is the opposite of importing a conclusion from its inputs. The only notable gap, that the symmetric-window conditional is 'imposed' on autoregressive models without specifying the mechanism, is a definitional/implementation concern about what quantity is measured rather than a circular reduction: no equation equates a claimed prediction to a fitted parameter or to a self-citation. If backend nondeterminism violated the exact-zero null, that would affect generality, not circularity.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The paper is explicit about its constructional axioms. No fitted free parameters enter the central claim; temperature and radius are scanned as experimental controls, not as fit constants. No new physical entities are postulated; all quantities are observables of the defined process.

assumptions (5)
  • standard math Exact-zero null: two rings with identical windows and a shared uniform stream draw the same token via inverse-CDF sampling.
    Invoked in Section 2 as the foundation of damage attribution; if this fails, the central measurement collapses.
  • domain assumption The language model is treated as a fixed conditional distribution p_r and the ring is closed after initialization.
    Section 2, Eq. (1); no external text enters, so the loop dynamics are purely the model's conditional plus the construction.
  • standard math The Domany-Kinzel identity is accepted as external ground truth for calibration.
    Section 3, Rung 1; the instrument is validated against this known answer.
  • domain assumption The discriminator assumes that varying one factor while holding the other fixed isolates construction-determined versus model-determined readings.
    Section 5 and Table 1; the entire separation argument rests on this causal design.
  • domain assumption The construction is a well-defined stochastic process: periodic ring, random visit order per replica, symmetric window with masked center.
    Section 2 defines the state space and update rule; all measurements depend on this definition.

how reviews work

0 comments
Cite this review

Pith. "Pith review of What Iterated Self-Feeding Probes of Language Models Measure, and a test that separates the construction from the model." pith.science (2026). https://pith.science/paper/2PKQ5FBO

@misc{pith2026260810986,
  author       = {Pith},
  title        = {Pith review of: What Iterated Self-Feeding Probes of Language Models Measure, and a test that separates the construction from the model},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2PKQ5FBO}},
  note         = {Machine review of arXiv:2608.10986}
}
read the original abstract

A growing class of methods probes a language model by feeding it its own output: self-consistency, iterated refinement, agentic loops. We ask what such a probe measures, in a construction chosen to make the question sharp: a ring of token cells resampled in place by the model's own windowed conditional p_r(x_i | x_{i+-r}). The substrate is Glauber dynamics on token sequences and is not new; what we change is the coupling. Advancing two rings that differ in one token under common random numbers makes undamaged copies diverge by exactly zero, so damage spreading becomes measurable where a maximal coupling gives mixing times instead. The answer is that it measures two different things at once, in readings that look alike. Some quantities are fixed by the construction: the damage light cone is kinematic, and the radius scaling of the token-space Lyapunov exponent lambda_ca(r) is model-invariant across 19 models and two scale ladders spanning 70x. Others genuinely track the model: lambda_ca crosses zero at a reproducible point in training, and the attractor share ranks models consistently however the lattice is built. Left undistinguished, the first kind is readily mistaken for the second -- we did so ourselves for four months, and report a phase transition we measured to three decimal places that belongs to the probe rather than to any language model. We give the test that separates them: hold the construction fixed and vary the model, or hold the model fixed and vary the construction, and see which readings move. We validate the instrument by reproduction first, recovering a Domany-Kinzel damage field bit-exactly against an independent prediction, and we report the estimator failures that this discipline caught -- four retracted verdicts, each on a quantity that looked like a measurement. The methodology ships as a package.

Figures

Figures reproduced from arXiv: 2608.10986 by the authors.

Figure 1
Figure 1. The construction. A ring of token cells is resampled in place by the model’s own windowed [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Validation by reproduction. The instrument is calibrated against systems whose answers [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. The transition belongs to the probe. Left (r = 2): under the autoregressive construction pythia-410m collapses onto a single token as T falls, reaching a top-1 share of 0.744 at T = 0.02, and crosses half the ring at the measured Tc. Three controls do not. Prepending one beginning-of￾sequence token to the same model drops it to 0.241 — the map’s domain changes, not its parameters. gpt2-medium, whose argmax map has n… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: A model-determined reading. With the automaton byte-identical at every checkpoint, [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

60 extracted references · 48 canonical work pages

  1. [1]

    a fer, and Niels L \

    Julian Arnold, Flemming Holtorf, Frank Sch \"a fer, and Niels L \"o rch. Phase transitions in the output distribution of large language models, 2024

    work page 2024

  2. [2]

    Damage spreading and lyapunov exponents in cellular automata

    Franco Bagnoli, Ra \'u l Rechtman, and Stefano Ruffo. Damage spreading and lyapunov exponents in cellular automata. Physics Letters A, 172 0 (1--2): 0 34--38, 1992

    work page 1992

  3. [3]

    Equivalence of cellular automata to I sing models and directed percolation

    Eytan Domany and Wolfgang Kinzel. Equivalence of cellular automata to I sing models and directed percolation. Physical Review Letters, 53 0 (4): 0 311--314, 1984

    work page 1984

  4. [4]

    Model equality testing: Which model is this API serving? In International Conference on Learning Representations (ICLR), 2025

    Irena Gao, Percy Liang, and Carlos Guestrin. Model equality testing: Which model is this API serving? In International Conference on Learning Representations (ICLR), 2025

    work page 2025

  5. [5]

    Auditing question-order effects in large language models with the QQ equality: Mechanism characterization and a saturation caveat, 2026

    Pilsung Kang. Auditing question-order effects in large language models with the QQ equality: Mechanism characterization and a saturation caveat, 2026

    work page 2026

  6. [6]

    Lieb and Derek W

    Elliott H. Lieb and Derek W. Robinson. The finite group velocity of quantum spin systems. Communications in Mathematical Physics, 28 0 (3): 0 251--257, 1972

    work page 1972

  7. [7]

    Self-refine: Iterative refinement with self-feedback, 2023

    Aman Madaan, Niket Tandon, Prakhar Gupta, Skyler Hallinan, Luyu Gao, Sarah Wiegreffe, Uri Alon, Nouha Dziri, Shrimai Prabhumoye, Yiming Yang, Shashank Gupta, Bodhisattwa Prasad Majumder, Katherine Hermann, Sean Welleck, Amir Yazdanbakhsh, and Peter Clark. Self-refine: Iterative refinement with self-feedback, 2023

    2023

  8. [8]

    The hydra effect: Emergent self-repair in language model computations, 2023

    Thomas McGrath, Matthew Rahtz, Janos Kramar, Vladimir Mikulik, and Shane Legg. The hydra effect: Emergent self-repair in language model computations, 2023

    2023

Show all 60 references
  1. [9]

    Mathematical contributions to the theory of evolution.--- XI

    Karl Pearson. Mathematical contributions to the theory of evolution.--- XI . on the influence of natural selection on the variability and correlation of organs. Philosophical Transactions of the Royal Society of London. Series A, 200: 0 1--66, 1903. doi:10.1098/rsta.1903.0001

    1903

  2. [10]

    When LLMs play the telephone game: Cultural attractors as conceptual tools to evaluate LLMs in multi-turn settings, 2024

    J \'e r \'e my Perez, Grgur Kova c , Corentin L \'e ger, C \'e dric Colas, Gaia Molinaro, Maxime Derex, Pierre-Yves Oudeyer, and Cl \'e ment Moulin-Frier. When LLMs play the telephone game: Cultural attractors as conceptual tools to evaluate LLMs in multi-turn settings, 2024

    2024

  3. [11]

    Explorations of self-repair in language models

    Cody Rushing and Neel Nanda. Explorations of self-repair in language models. In International Conference on Machine Learning (ICML), 2024

    2024

  4. [12]

    Sackett and Hyuckseung Yang

    Paul R. Sackett and Hyuckseung Yang. Correction for range restriction: An expanded typology. Journal of Applied Psychology, 85 0 (1): 0 112--118, 2000. doi:10.1037/0021-9010.85.1.112

    doi

  5. [13]

    Mixing times of glauber dynamics on masked language models, 2026

    Suvadip Sana, Sami Wolf, Neer Mehta, Alina Shah, Aitzaz Shaikh, Janna Goodman, and Lionel Levine. Mixing times of glauber dynamics on masked language models, 2026

    2026

  6. [14]

    Self-consistency improves chain of thought reasoning in language models, 2022

    Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. Self-consistency improves chain of thought reasoning in language models, 2022

    2022

  7. [15]

    Unveiling attractor cycles in large language models: A dynamical systems view of successive paraphrasing, 2025

    Zhilin Wang, Yafu Li, Jianhao Yan, Yu Cheng, and Yue Zhang. Unveiling attractor cycles in large language models: A dynamical systems view of successive paraphrasing, 2025

    2025

  8. [16]

    Chain-of-thought prompting elicits reasoning in large language models, 2022

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc Le, and Denny Zhou. Chain-of-thought prompting elicits reasoning in large language models, 2022

    2022

  9. [17]

    Which model is actually serving you? IRIS : Budgeted black-box auditing of model substitution and routing dilution in LLM gateways, 2026

    Yuewei Zhang, Zhi-Hai Zhang, and Hanzhang Qin. Which model is actually serving you? IRIS : Budgeted black-box auditing of model substitution and routing dilution in LLM gateways, 2026

    2026

  10. [18]

    Physics Letters A , volume=

    Damage spreading and Lyapunov exponents in cellular automata , author=. Physics Letters A , volume=

  11. [19]

    Communications in Mathematical Physics , volume=

    The finite group velocity of quantum spin systems , author=. Communications in Mathematical Physics , volume=

  12. [20]

    Physical Review B , volume=

    Butterfly effect and spatial structure of information spreading in a chaotic cellular automaton , author=. Physical Review B , volume=. 2021 , eprint=

    2021

  13. [21]

    Physica D: Nonlinear Phenomena , volume=

    Computation at the edge of chaos: Phase transitions and emergent computation , author=. Physica D: Nonlinear Phenomena , volume=. 1990 , publisher=

    1990

  14. [22]

    Neural Computation , volume=

    Real-time computation at the edge of chaos in recurrent neural networks , author=. Neural Computation , volume=. 2004 , publisher=

    2004

  15. [23]

    Physica D: Nonlinear Phenomena , volume=

    Computational mechanics of cellular automata: An example , author=. Physica D: Nonlinear Phenomena , volume=. 1997 , publisher=

    1997

  16. [24]

    2026 , eprint=

    Mixing Times of Glauber Dynamics on Masked Language Models , author=. 2026 , eprint=

    2026

  17. [25]

    2025 , eprint=

    Recurrent Self-Attention Dynamics: An Energy-Agnostic Perspective from Jacobians , author=. 2025 , eprint=

    2025

  18. [26]

    2026 , eprint=

    Spectral Origins of the Self-Correction Blind Spot in Autoregressive Generation , author=. 2026 , eprint=

    2026

  19. [27]

    2025 , eprint=

    Cognitive Activation and Chaotic Dynamics in Large Language Models: A Quasi-Lyapunov Analysis of Reasoning Mechanisms , author=. 2025 , eprint=

    2025

  20. [28]

    International Conference on Learning Representations (ICLR) , year=

    Intelligence at the Edge of Chaos , author=. International Conference on Learning Representations (ICLR) , year=. 2410.02536 , archivePrefix=

    arXiv

  21. [29]

    2023 , eprint=

    The Hydra Effect: Emergent Self-repair in Language Model Computations , author=. 2023 , eprint=

    2023

  22. [30]

    International Conference on Machine Learning (ICML) , year=

    Explorations of Self-Repair in Language Models , author=. International Conference on Machine Learning (ICML) , year=. 2402.15390 , archivePrefix=

    arXiv

  23. [31]

    2026 , eprint=

    QUIVER: A Formal Framework for Quantifying Perturbation Propagation and Bifurcation in Compound AI Systems , author=. 2026 , eprint=

    2026

  24. [32]

    2024 , eprint=

    Phase transition in large language models and the criticality of natural languages , author=. 2024 , eprint=

    2024

  25. [33]

    2024 , eprint=

    Phase Transitions in the Output Distribution of Large Language Models , author=. 2024 , eprint=

    2024

  26. [34]

    2026 , eprint=

    Generative Criticality in Large Language Model Temperature Scaling , author=. 2026 , eprint=

    2026

  27. [35]

    2025 , eprint=

    Phase Transitions in Large Language Models and the O(N) Model , author=. 2025 , eprint=

    2025

  28. [36]

    (Un)biased data and spin glasses reveal clustering for

    George, Jackson and Yusaf, Zachariah and Zoltick, Stephanie and Huynh, Linh , year=. (Un)biased data and spin glasses reveal clustering for. 2505.02879 , archivePrefix=

  29. [37]

    2026 , eprint=

    Transformer Field Theory: A Response-Theoretic Approach to Mechanistic Interpretability , author=. 2026 , eprint=

    2026

  30. [38]

    2026 , eprint=

    Dynamics of the Transformer Residual Stream: Coupling Spectral Geometry to Network Topology , author=. 2026 , eprint=

    2026

  31. [39]

    2026 , eprint=

    LOGOS-CA: A Cellular Automaton Using Natural Language as State and Rule , author=. 2026 , eprint=

    2026

  32. [40]

    Perez, J. When. 2024 , eprint=

    2024

  33. [41]

    2025 , eprint=

    Unveiling Attractor Cycles in Large Language Models: A Dynamical Systems View of Successive Paraphrasing , author=. 2025 , eprint=

    2025

  34. [42]

    2026 , eprint=

    Is Your Diffusion Sampler Actually Correct? A Sampler-Centric Evaluation of Discrete Diffusion Language Models , author=. 2026 , eprint=

    2026

  35. [43]

    2023 , eprint=

    Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling , author=. 2023 , eprint=

    2023

  36. [44]

    Neural networks trained with

    Refinetti, Maria and Ingrosso, Alessandro and Goldt, Sebastian , booktitle=. Neural networks trained with. 2023 , eprint=

    2023

  37. [45]

    Equivalence of Cellular Automata to

    Domany, Eytan and Kinzel, Wolfgang , journal =. Equivalence of Cellular Automata to. 1984 , publisher =

    1984

  38. [46]

    Kohring, G. A. and Schreckenberg, M. , journal =. The

  39. [47]

    Journal of Statistical Physics , volume =

    Are damage spreading transitions generically in the universality class of directed percolation? , author =. Journal of Statistical Physics , volume =

  40. [48]

    Journal of Statistical Physics , volume =

    An algorithm-independent definition of damage spreading---application to directed percolation , author =. Journal of Statistical Physics , volume =

  41. [49]

    Advances in Physics , volume =

    Non-equilibrium critical phenomena and phase transitions into absorbing states , author =. Advances in Physics , volume =

  42. [50]

    Zebende, G. F. and Penna, T. J. P. , journal =. The

  43. [51]

    Physical Review Research , volume=

    Universal Scaling Laws of Absorbing Phase Transitions in Artificial Deep Neural Networks , author=. Physical Review Research , volume=. 2025 , eprint=

    2025

  44. [52]

    2023 , eprint=

    Are Emergent Abilities of Large Language Models a Mirage? , author=. 2023 , eprint=

    2023

  45. [53]

    Model Equality Testing: Which Model Is This

    Gao, Irena and Liang, Percy and Guestrin, Carlos , booktitle=. Model Equality Testing: Which Model Is This. 2025 , eprint=

    2025

  46. [54]

    Which Model Is Actually Serving You?

    Zhang, Yuewei and Zhang, Zhi-Hai and Qin, Hanzhang , year=. Which Model Is Actually Serving You?. 2607.20860 , archivePrefix=

    arXiv

  47. [55]

    2022 , eprint=

    Chain-of-Thought Prompting Elicits Reasoning in Large Language Models , author=. 2022 , eprint=

    2022

  48. [56]

    2022 , eprint=

    Self-Consistency Improves Chain of Thought Reasoning in Language Models , author=. 2022 , eprint=

    2022

  49. [57]

    2023 , eprint=

    Self-Refine: Iterative Refinement with Self-Feedback , author=. 2023 , eprint=

    2023

  50. [58]

    Auditing Question-Order Effects in Large Language Models with the

    Kang, Pilsung , year=. Auditing Question-Order Effects in Large Language Models with the. 2607.17219 , archivePrefix=

    arXiv

  51. [59]

    Mathematical contributions to the theory of evolution.---

    Pearson, Karl , journal=. Mathematical contributions to the theory of evolution.---. 1903 , doi=

    1903

  52. [60]

    Journal of Applied Psychology , volume=

    Correction for range restriction: An expanded typology , author=. Journal of Applied Psychology , volume=. 2000 , doi=

    2000

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.