Pith.

Machine Learning · arXiv:2611.14683 v1 · submitted 10 Nov 2026 · 36 references

Trait Hitchhiking in LLM Post-Training

Adil Soubki, Elizabeth S.Z. Tan, Miles Cranmer

The paper says

“Reinforcement learning (RL) is now a standard stage of language model post-training, and the compute devoted to it is growing rapidly.”

more of the abstract

While it has been shown to improve capabilities, particularly for tasks with verifiable rewards, it can also change behaviours and internal properties that the reward does not explicitly target. We present a formalism to describe the evolution of "traits" during RL post-training and, in doing so, recover the Price equation from evolutionary biology. We use this to develop a mathematical model for what we call "trait hitchhiking": the amplification or suppression of non-targeted traits as training redistributes probability across completions. We evaluate this Price update expression on language models trained with GRPO and show that the Price estimates closely track the sampled trait dynamics. These results identify trait hitchhiking as a mechanism by which RL post-training can silently amplify undesirable traits.

from the abstract

Pith says

Machine review · deepseek-v4-flash · 13 Nov 2026 · what was read was not recorded · not checked by a person

It can be wrong. Challenge this review or reply to it.

A sound accounting identity applied to RL post-training, backed by genuinely careful empirical work — the sycophancy experiment is the real result, and the paper's honest about its weak spots. This line is wrong

the referee pass did not run

How often objections from reviews like this one are right: not yet measured

Are you an author? What you can do.

Not checked: figures, code, what the 36 references say.

What would settle it

Re-run the activation-level experiment (all-layers condition) with a larger number of importance samples (e.g., N=2048 instead of 512) for the first few transitions. If the Price estimate (selection + transmission) still closely tracks the directly measured change with ESS > 200 for all transitions, the concern about estimator reliability is resolved. If the quantitative agreement degrades or the confidence intervals widen substantially, the claim of 'accurate accounting' would need to be qualified for the early training regime.

from the skeptic's reading

How the review went

3 readings by deepseek-v4-flash · 13 Nov 2026

  1. Reader holds up confidence high
  2. Skeptic partly agrees 9 s · 29,258 tokens read, 847 written
  3. Editor wrote the verdict above 22 s · 30,103 tokens read, 3,097 written

Not run on this paper: the referee, the simulated authors' defence.

Pith did not check this review's objections against the paper.

The review in full

The review in brief, machine-written

The short version: this paper applies the Price equation — a fifty-year-old identity from evolutionary biology — to RL post-training of LLMs, and then does the empirical legwork to show the accounting actually tracks. The derivation is an identity, so nothing new is proved mathematically; the paper knows this and sells itself as an accounting mechanism that attributes trait changes to selection (probability redistribution across completions) versus transmission (the trait scorer moving under the update). That framing is the contribution, and it's a useful one.

The sycophancy experiment is the cleanest result: a trait correlated with reward but never rewarded gets amplified exactly as the covariance term predicts, across a sweep of hint reliabilities. That's a real demonstration, not a rehash. I'd also credit the paper for sample discipline — direct measurements and Price estimates use independent rollouts, so no trivial circularity — for the exact tabular-policy validation with convergence diagnostics, and for the matched all-layers intervention that restores the transmission term. The authors flag the activation-direction rotation problem, the impracticality of frozen measurement layers, and the effective-sample-size issue themselves, and then run an independent endpoint check rather than burying it.

Soft spots, in proportion. First, the stress-test concern about low ESS is real but contained. ESS of 6–11 out of 512 applies to the first 0→3 transition of the all-layers transmission decomposition; the main fixed-trait experiment has ESS in the high hundreds, and its 0.98 correlation survives dropping the first three transitions (0.969). The all-layers full-sample correlation of 0.999 is genuinely leveraged by the early large changes — it falls to 0.897 without 0→3 — but the authors report that and independently confirm the sign pattern with fresh rollouts. This is a caveat on the precision of one decomposition, not a flaw in the central claim. Second, the safety framing runs a bit ahead of the evidence: trait hitchhiking is presented as a mechanism for emergent misalignment, but the demonstrated case is benign sycophancy, and the paper concedes it is checkpoint-to-checkpoint accounting, not forecasting. That sentence deserves softening in revision. Third, no code or locked environment is released; for a paper whose value is partly methodological, that's a real shortcoming.

For whom: AI safety researchers and anyone doing RL post-training who wants a formal language for off-target trait drift. It deserves a serious referee; the work is thorough, reproducible on paper, and the limitations are stated rather than concealed. I'd accept with minor revision: soften the emergent-misalignment claim, release the code.

The strongest objection

The central empirical claim is that the Price equation accurately accounts for trait changes during RL post-training. The paper's own all-layers experiment (Appendix E) reveals that for the first policy transition (0→3), the effective sample size (ESS) for the importance-weighted covariance estimate can be as low as 6 out of 512 (Table 4, run 3). This is a regime where the policy changes most, and the importance weights become extremely variable. The paper acknowledges this low ESS (b536) and provides an independent validation of the endpoint components (Appendix E.1), which shows that the sign pattern holds but that the original IS estimates can differ from the direct estimates (b579, b581). The concern is not that the Price equation is wrong, but that the empirical validation of the selection term in the most critical regime—where the policy changes substantially—relies on an estimator with very low effective sample size. The paper's own data show that the direct and IS estimates can disagree (e.g., run 1, feature 27505: IS transmission -0.1507 vs direct interval [-0.1462, -0.1397]), and the paper states that these are not formal two-sample tests (b581). This does not invalidate the overall conclusion, but it means that the quantitative accuracy of the Price equation in the early, high-change regime is less precisely established than the high correlation (0.98) across all transitions might suggest. The correlation is dominated by the large early changes, but those are precisely the ones with the lowest ESS.

The reader's weakest_assumption focuses on the stability of the trait scoring function for activation-level traits and the practicality of freezing layers. This is a valid concern about the applicability of the method, but it is not the most load-bearing concern for the central empirical claim. The paper acknowledges this limitation (b204, b205) and provides an all-layers experiment (Appendix E) that directly addresses it, showing that the Price equation still holds when the transmission term is non-zero. My concern is different: it is about the reliability of the importance-sampling estimator used to compute the selection term in the regime where the policy changes most. The paper's own data show very low effective sample sizes for the first transition in the all-layers experiment (ESS as low as 6 out of 512). This is a technical concern about the strength of the empirical evidence, not about the theoretical derivation. The paper provides an independent validation (Appendix E.1) that the sign pattern holds, but the quantitative accuracy of the Price estimate in this regime is less certain. I recommend UNCHANGED because the paper's central claim is well-supported by the theoretical derivation and the empirical evidence is still strong overall. The concern I raise is a nuance about the precision of the empirical validation, not a fatal flaw. The paper's own limitations section and appendices are transparent about this issue. The concrete test I propose would strengthen the empirical case by demonstrating that the result holds with higher precision in the early regime.

Pith's reading of the paper (a machine's summary, not the authors')

This paper argues that the Price equation, a foundational identity from evolutionary biology, provides an exact accounting framework for how reinforcement learning post-training changes the prevalence of traits in a language model, even when those traits are not explicitly rewarded. The authors define a trait scoring function for any input-output property, derive that the per-step change in the average trait level equals the covariance between the policy reweighting factor and the trait score (the selection term) plus an expected transmission term when the scoring function itself changes. They test this on small LLMs trained with GRPO: for output-level traits like sycophancy, the cumulative covariance estimate closely tracks the directly measured change (final discrepancy < 0.033 across training-hint conditions); for activation-level SAE features, the Price estimate predicts per-transition change with correlation 0.98 ± 0.01 and slope 1.02 ± 0.07. The paper establishes trait hitchhiking as a concrete mechanism by which RL can silently amplify undesirable behaviours, making a case for trait-level auditing alongside reward optimisation.

What carries the argument. The Price equation identity ΔT_t = Cov_{P_t}(ω_t, s_t) + E_{P_t}[ω_t Δs_t], where ω_t = P_{t+1}/P_t is the importance weight (the ratio of policy probabilities before and after a training step) and s_t is the trait scoring function. The selection term captures how reweighting probability mass toward high-reward completions incidentally changes the average trait level when the trait is correlated with the reward via the shared probability reweighting. The transmission term captures changes in the trait score of a fixed input-output pair when the model's internal representations evolve. The paper's key experimental design choice is to eliminate the transmission term (by using output-level traits or by freezing the measurement layer via LoRA) so that the selection term alone can be directly compared against observed trait changes.

Core claim. The paper discovers that the Price equation, originally derived to describe how average trait values change under natural selection, exactly characterises the evolution of any measurable trait during RL post-training of LLMs. The central identity is ΔT_t = Cov_{P_t}(ω_t, s_t) + E_{P_t}[ω_t Δs_t], where ω_t = P_{t+1}/P_t is the probability reweighting induced by the policy update, s_t is the trait scoring function, and the two terms are called selection and transmission. For output-level traits (e.g., sycophancy measured from responses) the transmission term is zero because the trait score of a fixed prompt–completion pair does not change; for activation-level traits measured at a fixed layer, freezing that layer (via LoRA only on later layers) also forces Δs_t = 0. The paper verifies that the cumulative selection term alone accounts for observed trait changes in both tabular policies and real LLMs trained with GRPO, with quantitative agreement (correlation ~0.98, slope ~1.02). This establishes that non-targeted traits hitchhike on the probability redistribution driven by reward optimisation, exactly as correlated traits do in natural selection.

The premise it rests on. The trait scoring function s_t(x,a) must remain meaningful and consistently measurable across training steps; for activation-level traits, this requires either freezing the measurement layer (which may be impractical) or accepting that the transmission term is non-zero and difficult to estimate with low variance.

If the paper is right (Pith's inferences, not the authors')

RL post-training can silently amplify undesirable traits (e.g., sycophancy, reward-correlated internal features) even when the reward function is correctly specified and does not target those traits.

Trait-level auditing should be performed throughout training, not only at the final checkpoint, because hitchhiker traits can rise and saturate early while task performance continues to improve.

The Price equation provides a practical accounting tool: practitioners can estimate the selection term from importance-weighted samples to detect which non-targeted traits are being amplified or suppressed at each training step.

Modifying the reward function to penalise an undesirable trait (r_aug = r_task - λ s_undesired) is a suggested mitigation, though the paper does not evaluate this intervention.

The framework generalises beyond LLMs: any iterative optimisation that reweights a distribution over input-output pairs (e.g., genetic algorithms, classifier-based distillation) can be analysed via the same covariance decomposition.

The paper's empirical validation uses small models (0.5B and 2B parameters) and a single RL algorithm (GRPO). The quantitative accuracy of the Price equation for larger models or different RL algorithms (PPO, RLOO, REINFORCE with KL penalties) is an open question; the identity itself is exact, but the practical estimator's variance and bias at scale are not yet characterised.

The transmission term for activation-level traits, which the paper mostly eliminates by freezing layers, could be substantial in realistic settings where all layers are updated. The paper's all-layers experiment (Appendix E) shows that transmission can oppose selection, but the estimator's effective sample size is very low for early transitions. Developing low-variance estimators for the transmiss

The paper's sycophancy experiment uses a narrow, synthetic form of sycophancy (switching answers based on user hints in arithmetic problems). Whether real-world sycophancy (e.g., agreeing with user opinions on subjective topics) follows the same hitchhiking dynamics is a plausible but untested extension.

The connection to emergent misalignment is suggestive but not experimentally demonstrated: the paper shows that non-targeted traits can be amplified, but does not show that this leads to the kind of broad, harmful misalignment observed in production RL systems. Establishing that trait hitchhiking is a causal mechanism for emergent misalignment would require additional experiments where a harmful t

Check this review

Pith stored no record of this review's passes and took no fingerprint of it. Reviews made by code from 13 Nov 2026 on carry both.

How often reviews like this one are right

Reviews like this one: written by deepseek-v4-flash, where what was read was not recorded.

How often objections from reviews like this one are right: not yet measured

Every value above, in the review's public record (JSON)

Reviews by people

This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.

Request a human review

How Pith checks papers · The full audit of this paper

The paper

Abstract

Reinforcement learning (RL) is now a standard stage of language model post-training, and the compute devoted to it is growing rapidly. While it has been shown to improve capabilities, particularly for tasks with verifiable rewards, it can also change behaviours and internal properties that the reward does not explicitly target. We present a formalism to describe the evolution of "traits" during RL post-training and, in doing so, recover the Price equation from evolutionary biology. We use this to develop a mathematical model for what we call "trait hitchhiking": the amplification or suppression of non-targeted traits as training redistributes probability across completions. We evaluate this Price update expression on language models trained with GRPO and show that the Price estimates closely track the sampled trait dynamics. These results identify trait hitchhiking as a mechanism by which RL post-training can silently amplify undesirable traits.

The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. No sentence in this record was written by a model.

abstract compiled · 2026-11-13

The authors' main claim, as the classifier picked it out 1.00

“These results identify trait hitchhiking as a mechanism by which RL post-training can silently amplify undesirable traits.”

How the main claim is stated: asserted 0.99

What the abstract itself states, as probabilities: claims first 0.06 · states limitation 0.12 · human participants 0.03 · controlled comparison 0.06 · randomized assignment 0.02 · states quantitative result 0.12 · compares with prior results 0.21 · mentions code or data availability 0.03

The abstract, sentence by sentence

This table labels the abstract only (arXiv listing abstract v1). The number is the probability the classifier gave the label.

SentenceRoleThe authors' own workStates the main conclusionStrength
Reinforcement learning (RL) is now a standard stage of language model post-training, and the compute devoted to it is growing rapidly. background 1.00 0.05 0.02 asserted 1.00
While it has been shown to improve capabilities, particularly for tasks with verifiable rewards, it can also change behaviours and internal properties that the reward does not explicitly target. background 0.98 0.08 0.07 asserted 0.90
We present a formalism to describe the evolution of "traits" during RL post-training and, in doing so, recover the Price equation from evolutionary biology. contribution 0.93 0.95 0.07 asserted 1.00
We use this to develop a mathematical model for what we call "trait hitchhiking": the amplification or suppression of non-targeted traits as training redistributes probability across completions. contribution 0.77 0.87 0.11 asserted 0.94
We evaluate this Price update expression on language models trained with GRPO and show that the Price estimates closely track the sampled trait dynamics. result 0.98 0.96 0.08 asserted 1.00
These results identify trait hitchhiking as a mechanism by which RL post-training can silently amplify undesirable traits. interpretation 0.99 0.88 0.89 asserted 0.99

What kind of paper the classifier reports this is: theoretical result 0.97 · evidence from simulation or computation 0.48

What was examined, and what was not

Sources read

  • arXiv listing abstract v1

Not examined

  • full text
  • figures and tables
  • references and citing sentences
  • relations to other papers

Levels run

  • P abstract

Not run

  • S sections
  • T body sentences
  • C citations
  • R cross-paper relations

How the labels were made

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-11-13: 1 request, 4,938 input tokens.

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. This record as JSON

Figures (12)
Figure 1 of the paper
Figure 1. Trait hitchhiking follows the same law in evolution and in RL. Left: long tails are correlated with other traits that make a widowbird ‘fit’, although long tails themselves do not lead to increased fitness of a widowbird. Right: RL leads to an increased probability for outputs that achieve a high reward. Behaviours or traits that appear consistently in highly rewarded outputs will thus also gain a greater prevalence, even though the reward function does not necessarily reward these traits. The change in the average trait exhibited by the model is the covariance of reinforcement…
Figure 2 of the paper
Figure 2. One GRPO step (top) and the held￾out samples used to measure a trait (bottom). The mean of st,i estimates the trait level Tt, and the sample covariance of ωt,i and st,i estimates the selection term (Section 3.2). Shaded circles are observed variables; the shaded diamond is fixed at step t and the open diamond is computed from its parents. For activation-level traits, st,i also depends on θt. It is useful to split up these two types of traits because the scoring function for activation level traits can change after an RL training step. If we determine that a specific activation direc
Figure 3 of the paper
Figure 3. Expected trait level Tt (blue), expected reward Rt (green), and the exact Price equation reconstruction (orange) across initial trait prevalences p (rows) and trait–quality correlations γ (columns). Although only reward is optimised, the trait rises for γ > 0 and falls for γ < 0, while the reconstruction overlaps the observed trajectory. Curves show means over five seeds with standard-error bands; gray dashed lines mark T0. appearing in the reward function. This creates a toy scenario where quality is the causal variable for maximizing reward and the trait is external. We vary γ,…
Figure 4 of the paper
Figure 4. Cumulative change in wrong hint influence throughout training across conditions with different hint
Figure 5 of the paper
Figure 5. Example training prompt. The suggested answer, D, is correct. The reward depends only on answer
Figure 6 of the paper
Figure 6. Cumulative activation-trait change for three layer-12 SAE features during GRPO on GSM8K: initially positive
Figure 7 of the paper
Figure 7. Convergence of the sampled Price selection term in the tabular policy. Blue shows the exact cumulative trait
Figure 8 of the paper
Figure 8. Cumulative activation-trait change for all 15 tracked layer-12 SAE features, grouped by positive, negative, or
Figure 9 of the paper
Figure 9. Activation-level Price predictions against independently measured trait changes for all adjacent GRPO
Figure 10 of the paper
Figure 10. Activation-trait dynamics when LoRA updates all decoder layers and the transmission term is non-zero.
Figure 11 of the paper
Figure 11. Full Price predictions against independently measured activation-trait changes for the all-layers intervention.
Figure 12 of the paper
Figure 12. Training-batch correctness during the activation-level experiments. Each point averages the

Figures are the authors', reproduced from the paper.

References (36)
  1. [1] Charles Darwin. On the Origin of Species by Means of Natural Selection, or the Preservation of Favoured Races in the Struggle for Life. John Murray, London, 5th edition, 1869 on Pith
  2. [2] C. W. Mackworth-Praed and C. H. B. Grant. Birds of Eastern and North Eastern Africa, volume 2 of African Handbook of Birds, Series One. Longmans, Green and Co., London, second edition, 1960 on Pith
  3. [3] Malte Andersson. Female choice selects for extreme tail length in a widowbird. Nature, 299 0 (5886): 0 818--820, October 1982. doi:10.1038/299818a0 on Pithdoi:10.1038/299818a0
  4. [4] Charles Darwin. The Descent of Man, and Selection in Relation to Sex. John Murray, London, 1871 on Pith
  5. [5] Ronald A. Fisher. The Genetical Theory of Natural Selection. Clarendon Press, Oxford, 1930 on Pith
  6. [6] George R. Price. Selection and covariance. Nature, 227 0 (5257): 0 520--521, August 1970. doi:10.1038/227520a0 on Pithdoi:10.1038/227520a0
  7. [7] Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. Lora: Low-rank adaptation of large language models, 2021. URL https://arxiv.org/abs/2106.09685 on PithPith's reviewarXiv
  8. [8] Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, Christopher Hesse, and John Schulman. Training verifiers to solve math word problems. arXiv preprint arXiv:2110.14168, 2021 on PithPith's reviewarXiv
  9. [9] Robert Huben, Hoagy Cunningham, Logan Smith, Aidan Ewart, and Lee Sharkey. Sparse autoencoders find highly interpretable features in language models. In International Conference on Learning Representations, 2024. URL https://arxiv.org/abs/2309.08600 on PithPith's reviewarXiv
  10. [10] George R. Price. Extension of covariance selection mathematics. Annals of Human Genetics, 35 0 (4): 0 485--490, April 1972. doi:10.1111/j.1469-1809.1957.tb01874.x on Pithdoi:10.1111/j.1469-1809.1957.tb01874.x
  11. [11] Andy Gardner. Price's equation made clear. Philosophical Transactions of the Royal Society B: Biological Sciences, 375 0 (1797): 0 20190361, 03 2020. ISSN 0962-8436. doi:10.1098/rstb.2019.0361. URL https://doi.org/10.1098/rstb.2019.0361 on Pithdoi:10.1098/rstb.2019.0361
  12. [12] Steven A. Frank. The Price equation, Fisher 's fundamental theorem, kin selection, and causal analysis. Evolution, 51 0 (6): 0 1712--1729, 1997. doi:10.1111/j.1558-5646.1997.tb05096.x on Pithdoi:10.1111/j.1558-5646.1997.tb05096.x
  13. [13] Jussi Lehtonen. The Price equation and the unity of social evolution theory. Philosophical Transactions of the Royal Society B, 375 0 (1797): 0 20190362, 2020. doi:10.1098/rstb.2019.0362 on Pithdoi:10.1098/rstb.2019.0362
  14. [14] Troy Day and Sylvain Gandon. Applying population genetic models in theoretical evolutionary epidemiology. Ecology Letters, 10 0 (10): 0 876--888, 2007. doi:10.1111/j.1461-0248.2007.01091.x on Pithdoi:10.1111/j.1461-0248.2007.01091.x
  15. [15] Sylvain Gandon and Troy Day. The evolutionary epidemiology of vaccination. Journal of the Royal Society Interface, 4: 0 803--817, 2007. doi:10.1098/rsif.2006.0207 on Pithdoi:10.1098/rsif.2006.0207
  16. [16] Michel Loreau and Andy Hector. Partitioning selection and complementarity in biodiversity experiments. Nature, 412: 0 72--76, 2001. doi:10.1038/35083573 on Pithdoi:10.1038/35083573
  17. [17] Jeremy W. Fox. Using the Price equation to partition the effects of biodiversity loss on ecosystem function. Ecology, 87 0 (11): 0 2687--2696, 2006. doi:10.1890/0012-9658(2006)87[2687:UTPETP]2.0.CO;2 on Pithdoi:10.1890/0012-9658(2006)87
  18. [18] Claire El Mouden, Jean-Baptiste Andr \'e , Olivier Morin, and Daniel Nettle. Cultural transmission and the evolution of human behaviour: A general approach based on the Price equation. Journal of Evolutionary Biology, 27 0 (2): 0 231--241, 2014. doi:10.1111/jeb.12296 on Pithdoi:10.1111/jeb.12296
  19. [19] Sinéad Collins and Andy Gardner. Integrating physiological, ecological and evolutionary change: A Price equation approach. Ecology Letters, 12 0 (8): 0 744--757, 2009. doi:10.1111/j.1461-0248.2009.01340.x on Pithdoi:10.1111/j.1461-0248.2009.01340.x
  20. [20] Steven A. Frank. The Price equation reveals a universal force--metric--bias law of algorithmic learning and natural selection. Entropy, 27 0 (11): 0 1129, October 2025. doi:10.3390/e27111129. URL https://www.mdpi.com/1099-4300/27/11/1129 on Pithdoi:10.3390/e27111129
show 16 more
  1. [21] Lee Altenberg. The schema theorem and Price 's theorem. In L. Darrell Whitley and Michael D. Vose, editors, Foundations of Genetic Algorithms 3, pages 23--49. Morgan Kaufmann, San Francisco, CA, 1995. doi:10.1016/B978-1-55860-356-1.50006-6 on Pithdoi:10.1016/b978-1-55860-356-1.50006-6
  2. [22] Reilly Raab, Luca de Alfaro, and Yang Liu. Conjugate natural selection, 2022. URL https://arxiv.org/abs/2208.13898 on PithPith's reviewarXiv
  3. [23] Tom Everitt, Marcus Hutter, Ramana Kumar, and Victoria Krakovna. Reward tampering problems and solutions in reinforcement learning: A causal influence diagram perspective. Synthese, 198 0 (Suppl 27): 0 6435--6467, 2021. doi:10.1007/s11229-021-03141-4 on Pithdoi:10.1007/s11229-021-03141-4
  4. [24] Dario Amodei, Chris Olah, Jacob Steinhardt, Paul Christiano, John Schulman, and Dan Mané. Concrete problems in AI safety. arXiv preprint arXiv:1606.06565, 2016 on PithPith's reviewarXiv
  5. [25] Jan Leike, Miljan Martic, Victoria Krakovna, Pedro A. Ortega, Tom Everitt, Andrew Lefrancq, Laurent Orseau, and Shane Legg. AI safety gridworlds. arXiv preprint arXiv:1711.09883, 2017 on PithPith's reviewarXiv
  6. [26] Joar Skalse, Nikolaus Howe, Dmitrii Krasheninnikov, and David Krueger. Defining and characterizing reward gaming. In Advances in Neural Information Processing Systems, volume 35, 2022. doi:10.52202/068431-0687 on Pithdoi:10.52202/068431-0687
  7. [27] Leo Gao, John Schulman, and Jacob Hilton. Scaling laws for reward model overoptimization. arXiv preprint arXiv:2210.10760, 2022 on PithPith's reviewarXiv
  8. [28] Monte MacDiarmid, Benjamin Wright, Jonathan Uesato, Joe Benton, Jon Kutasov, Sara Price, Naia Bouscal, Sam Bowman, Trenton Bricken, Alex Cloud, Carson Denison, Johannes Gasteiger, Ryan Greenblatt, Jan Leike, Jack Lindsey, Vlad Mikulik, Ethan Perez, Alex Rodrigues, Drake Thomas... on PitharXiv
  9. [29] Jan Betley, Daniel Chee Hian Tan, Niels Warncke, Anna Sztyber-Betley, Xuchan Bao, Mart \'i n Soto, Nathan Labenz, and Owain Evans. Emergent misalignment: Narrow finetuning can produce broadly misaligned LLM s. In Proceedings of the 42nd International Conference on Machine Lear... on Pith
  10. [30] Lyudmila N. Trut. Early canid domestication: The farm-fox experiment. American Scientist, 87 0 (2): 0 160--169, 1999. doi:10.1511/1999.2.160 on Pithdoi:10.1511/1999.2.160
  11. [31] Devvrit Khatri, Lovish Madaan, Rishabh Tiwari, Rachit Bansal, Sai Surya Duvvuri, Manzil Zaheer, Inderjit S. Dhillon, David Brandfonbrener, and Rishabh Agarwal. The art of scaling reinforcement learning compute for LLMs , 2025. URL https://arxiv.org/abs/2510.13786 on PithPith's reviewarXiv
  12. [32] Osborne Reynolds. On the dynamical theory of incompressible viscous fluids and the determination of the criterion. Philosophical Transactions of the Royal Society of London. A, 186: 0 123--164, 1895. doi:10.1098/rsta.1895.0004 on Pithdoi:10.1098/rsta.1895.0004
  13. [33] Christian Kuehn. Moment closure---a brief review. In Eckehard Sch \"o ll, Sabine H. L. Klapp, and Philipp H \"o vel, editors, Control of Self-Organizing Nonlinear Systems, Understanding Complex Systems, pages 253--271. Springer, Cham, 2016. doi:10.1007/978-3-319-28028-8_13 on Pithdoi:10.1007/978-3-319-28028-8_13
  14. [34] Sydney Chapman and T. G. Cowling. The Mathematical Theory of Non-Uniform Gases. Cambridge University Press, 3 edition, 1970 on Pith
  15. [35] Russell Lande. Natural selection and random genetic drift in phenotypic evolution. Evolution, 30 0 (2): 0 314--334, 1976. doi:10.2307/2407703 on Pithdoi:10.2307/2407703
  16. [36] Karthik Duraisamy, Gianluca Iaccarino, and Heng Xiao. Turbulence modeling in the age of data. Annual Review of Fluid Mechanics, 51 0 (1): 0 357--377, 2019. doi:10.1146/annurev-fluid-010518-040547 on Pithdoi:10.1146/annurev-fluid-010518-040547

Challenges and discussion

Disagree with the review, or know something it missed? Leave a comment, or challenge this review formally: a challenge becomes a permanent entry in this paper's record, and its decision is published here.

No comments yet.

Sign in to comment