Learning Extrapolative Sequence Transformations from Markov Chains

Aleem Khan; Andrew Wang; Nicholas Andrews; Sophia Hager

REVIEW 5 major objections 5 minor 13 references

Learning Extrapolative Sequence Transformations from Markov Chains

T0 review · 5 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a fine-tuned model can extrapolate beyond the best scores achieved by the MCMC sampler that generated its training data, as shown by protein-stability rates several times higher than the sampler's in the…

desk verdict The core distillation idea is sound and the toy experiment is a genuine proof of concept, but the protein 'beats MCMC' claim rests on an unequal sample budget and post-hoc variant selection. read the letter →

arxiv 2505.20251 v1 pith:2NNWVV3O submitted 2025-05-26 cs.LG cs.CL

Sophia Hager , Aleem Khan , Andrew Wang , Nicholas Andrews This is my paper · ORCID

classification cs.LGcs.CL

keywords extrapolationMarkovchainMonteCarloMetropolis-Hastingsmaskedlanguagemodelsautoregressivesequencegenerationproteinengineeringsentimentcontroltextanonymization

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

The paper argues that the exploratory power of Markov chain Monte Carlo can be distilled into a fast autoregressive model: run MCMC with a masked-language-model proposal, prune each chain to the transitions that most improve a cheap score, and fine-tune a sequence model to reproduce those transitions. The central claim is that the resulting model $q_\theta$ extrapolates as well as or better than the sampler that produced its training data, while needing only a handful of generation steps. In the ACE2 protein-stability task, $q_\theta$ generates mutants below ddG $-5$, $-6$, and $-7$ at far higher rates than MCMC itself (74.8%, 61.6%, and 46.4% versus 27.0%, 4.1%, and 0.5%), and running MCMC ten times longer does not close the gap. On sentiment control the model matches most of MCMC's extrapolation rate with one step instead of roughly 500, and on anonymization it keeps most of the semantic similarity with far fewer iterations. If the claim holds, test-time search compute can be converted into training-time data, giving sample-efficient extrapolation in domains where the oracle is expensive to query.

What carries the argument

The central machinery is a two-phase distillation loop. Phase one builds an energy-based surrogate $\ln p(x) = s(x) - \ln Z$ from a cheap guide scorer plus auxiliary terms, and samples it with Metropolis-Hastings whose proposal distribution is a masked language model that masks a random subset of tokens and infills them. Phase two sub-samples the resulting chains into short training episodes — 'first/best', thinning, or $\Delta$-energy selection — and encodes each episode as a token sequence with special delimiters and observed or predicted scores, then fine-tunes an autoregressive model $q_\theta$ by teacher forcing to predict each next state given the history. At inference $q_\theta$ iteratively revises until it emits a stop token, so the model learns a non-autoregressive iterative transformation embedded in an autoregressive decoder. The $\Delta$-energy episode selection, which keeps transitions with the largest energy improvements, is the variant that gives the largest extrapolation gains in the main protein experiments.

What would settle it

Run a protein-task variant where the guide is trained only on ddG in $[-4,10]$ but an adversarial validation oracle is constructed so that below $-4$ the sequences the guide ranks as best are actually the least stable; if $q_\theta$ trained on $\Delta$-energy episodes then fails to beat MCMC under that oracle, the guide's out-of-range directional signal is the breaking point. Concretely, one would freeze the guide's weights, relabel only the extrapolation region of the evaluation oracle, and compare the FoldX ddG distributions of $q_\theta$ and MCMC on that oracle.

Watch

Extended reading notes

Core claim

The discovery is that a model trained on selected MCMC transitions can learn a transformation operator that keeps improving the target property beyond the score range the guide was trained on, and beyond the best states the sampler itself visited. Starting from a state, $q_\theta$ produces a sequence of revisions, and each revision is trained to imitate a sub-sampled transition that decreased the energy. The paper shows that this operator inverts the usual cost structure: the expensive MCMC search is paid once at training time, and inference is a few autoregressive passes. The strongest evidence is the protein task, where $q_\theta$'s extrapolation-range success rates exceed every baseline and MCMC's, despite MCMC having been the only source of its training data.

Load-bearing premise

The load-bearing premise is that the cheap guide scorer $s(x)$, trained only on the observed score range, still points in the right direction for sequences outside that range; if its gradient misleads there, the selected 'improvement' transitions would train $q_\theta$ toward spurious targets and the claimed extrapolation would not reproduce under the true oracle.

Editorial extensions

If this is right

$q_\theta$ can match or beat MCMC on extrapolation while needing far fewer inference iterations: 3 versus 83 on protein, 1 versus roughly 496 on sentiment, and 4 versus 4498 on anonymization.
On the protein task, running MCMC for ten epochs (830 steps) does not approach $q_\theta$'s extrapolation performance, so the gain is not simply additional search.
Episode construction matters: fixed-length $\Delta$-energy selection outperforms uniform thinning on protein extrapolation, first/best is the most effective and fluent variant for sentiment, and variable-length $\Delta$-energy preserves the most semantics for anonymization.
$q_\theta$ can be trained to predict scores rather than call the guide at inference, removing the per-step scoring cost with little loss in most configurations.
The pipeline works on extrapolation tasks (protein stability, sentiment) and on an interpolation task (anonymization), so the distillation mechanism is not specific to one domain.

Reading between the lines

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

A direct consequence the paper leaves implicit is that the method inherits the guide scorer's blind spots: whenever the guide's ranking degrades outside its training support, $q_\theta$ would be trained toward spurious energy decreases, so the reported successes depend on the guide's gradient remaining informative in the extrapolation region.
The same distillation recipe should transfer to other expensive search settings with cheap scoring signals and pre-trained infilling models, such as small-molecule design or program synthesis, where MCMC-style exploration is already available but inference-time search is costly.
A testable extension is to close the loop between sampler and model: use $q_\theta$'s own generated states and their oracle scores to re-weight or extend future MCMC episodes, which could push extrapolation further than a single distillation pass.
The approach turns inference-time scaling into training-time scaling, suggesting an unexplored trade-off curve between the MCMC budget spent on episodes and the extrapolation quality of $q_\theta$; mapping that curve would help practitioners decide how much search compute to amortize.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Quick take: the paper has a genuinely useful idea—distill MCMC search trajectories into a cheap autoregressive policy—and the toy example is a clean demonstration that this can extrapolate past the sampler. The FoldX evaluation on the real protein task is real external grounding. But the headline comparison against MCMC is not compute-matched, and the paper oversells it.

What is new: the episode-selection strategies (Δ-energy-based state selection) and the finding that the distilled model can exceed the sampler's best states. The combination of MLM proposals, MCMC, and fine-tuning an AR model on selected transitions is built from known parts—Goyal, ICE, Decision Transformer—but the packaging and the analysis of episode selection is a contribution.

Soft spots, in order of importance. First, the protein claim. qθ generates 10,000 mutants at inference (Appendix B.1); MCMC in Table 1 runs 83 proposals and Figure 2 extends to 830. So qθ sees roughly twelve times more final candidates than even the extended MCMC. The 'further MCMC does not approach qθ' statement in §3.5 is not established for equal compute. If MCMC had 10,000 proposals, it might well match qθ's tail. That matters because the paper's main selling point is extrapolation beyond the sampler, not just amortized speed. Second, the best episode variant is selected after looking at test metrics (Table 4), which inflates the reported numbers. Third, the protein table has no error bars, and the baselines are taken from a paper the authors say they could not reproduce; that makes the baseline comparisons shaky. Fourth, the guide-scorer assumption in §2.1 is acknowledged but not stress-tested; the FoldX results partially validate it, but a systematic exploration of guide failure modes would make the method more robust.

None of these sink the method. The efficiency story is real: qθ produces useful samples in far fewer iterations than MCMC, and the anonymization results show it can approximate MCMC's quality at a fraction of the cost. The paper would benefit from a compute-matched comparison (e.g., running MCMC for 10,000 proposals) and a rephrased abstract that claims competitive quality with much faster generation rather than outright superiority.

Who should read it: anyone working on controllable generation, protein design, or amortized search. It's a solid empirical paper. My verdict: deserves serious refereeing, but it needs a major revision to fix the compute mismatch and tone down the claims.

Referee Report

5 major / 5 minor

Summary. The paper proposes a method for learning extrapolative sequence transformations. The authors run Metropolis-Hastings MCMC with masked-language-model proposals under a guide energy function, subsample state transitions from the resulting chains using several strategies (uniform thinning, first/best, fixed- and variable-length energy improvement), and fine-tune an autoregressive model qθ on episodes of the form x0, s1, x1, ..., xn, optionally conditioning on intermediate scores. At inference, qθ iteratively revises an input sequence. The method is evaluated on a toy binary-sequence task, ACE2 protein stability (with FoldX as oracle), Yelp sentiment extrapolation, and Reddit anonymization. The central evidence is the protein result (Table 1), where the best qθ variant achieves 74.8% of mutants under ddG -5 versus 27.0% for MCMC, and Figure 2, which shows that 830 MCMC steps do not reach qθ's performance. The paper claims that qθ extrapolates as well as or better than MCMC while using far fewer iterations.

Significance. The idea of distilling MCMC exploration into a conditional autoregressive editor is appealing, and the toy example is a clean proof of concept. The paper ships code, evaluates with external oracles (FoldX, the oracle sentiment scorer), and reports extensive ablations of episode-creation strategies, reward types, and MCMC chain lengths, which is commendable. If the protein result survives matched comparisons, the method would offer a practical way to convert expensive stochastic search into a fast generative model for sequence design. The main caveats are experimental: the headline protein comparison is not sample-matched, some baseline numbers are unreproduced, and the best episode strategy is selected after inspecting test metrics. These issues make the current empirical significance uncertain rather than established.

major comments (5)

[§3.1, Table 1, Figure 2; Appendix B.1] The central protein comparison is not sample-matched. Appendix B.1 states that at inference qθ is prompted with the wildtype sequence and 10,000 mutants are sampled, whereas the MCMC row in Table 1 is based on 83 proposals and Figure 2 extends this to 830 proposals in a single chain. The 'Iterations' column is defined in §3 as the number of revised states in the training episode, not the number of final candidates or a compute measure, so qθ's '3' hides a 10,000-fold replication. Consequently, the claim in §3.5 that further MCMC 'does not begin to approach' qθ, and the abstract's claim of 'significantly higher sample efficiency', are not established for equal sample budget or equal compute. Please add matched comparisons, for example by reporting qθ restricted to 83 and 830 final candidates, running MCMC for a comparable number of proposals (e.g., 10,000), or reporting wall-clock time or FLOPs for both methods.
[§3.4, Tables 4–6] The headline numbers in Tables 1–3 are selected as the best of four episode-creation strategies after examining the test metrics. Table 1 reports fixed-length Δ energy because it is best in Table 4; Table 2 reports first/best because it is best in Table 5; Table 3 reports variable-length Δ energy because it is best in Table 6 under the reported metrics. No validation-based selection of the episode strategy is described; Appendix B.1 only addresses selection of the EBM weights. This selection-on-test inflates the reported effect sizes. Please either fix the episode strategy on a validation split before inspecting test results, or report the selection procedure explicitly and quantify its bias, for example by holding out part of the test set for final evaluation.
[§3.1, Table 1; §3.2, Table 2] The baseline comparisons are taken from Padmakumar et al. (2023) even though §3.1 states that the authors were unable to reproduce those published results with the released code and that the original procedure exhibits large variance. Comparing qθ against unreproduced numbers weakens the claims that qθ 'significantly outperforms our baselines' in Table 1 and that it outperforms baselines in the sentiment task. Please either provide reproduced baseline numbers under the same evaluation protocol, or clearly mark the comparison as previously published and explain why the numbers are still considered directly comparable.
[§3.3, Appendix G.2] The anonymization evaluation may be partly circular. The energy function used to generate MCMC training episodes includes a LUAR stylistic-similarity term, and the EER metric is computed with a LUAR-CRUD author embedding. Appendix G.2 notes that the SBERT checkpoint used in the energy function differs from the one used in evaluation, but it does not state whether the LUAR embedding is the same in both places. If the same authorship representation is used both to guide generation and to measure success, the reported EER improvement partly reflects optimization of the evaluation metric. Please clarify whether the LUAR checkpoints differ, and if they are identical, re-evaluate with a different author-embedding model or a held-out variant.
[§2.1–§2.3] The method relies on the assumption that the guide scorer s(x) is directionally informative outside its training support, because training episodes are selected according to improvements in s (for example, the Δ-energy strategies in §2.3). The paper states this limitation in §2.1, and the oracle/FoldX evaluations of final outputs provide indirect evidence that the assumption holds in the tested cases. However, there is no direct check that the selected MCMC transitions are true improvements according to the oracle. Please add a small validation experiment for at least the protein task in which a sample of selected transitions is scored by both the guide and the oracle, and report the correlation or the fraction of concordant pairs in the extrapolation region. This would turn the weakest assumption into a measured quantity.

minor comments (5)

[§2, toy example] The definition 's(x) = exp PL i ri' contains an undefined symbol 'ri' and an unclear sum; it should presumably be 's_i' with an explicit summation over i. Please correct the notation.
[§3.1 and §3.3, Tables 1 and 3] Tables 1 and 3 report no confidence intervals or numbers of independent runs, while Table 2 reports 95% confidence intervals over three test sets. Please add variability estimates for all three tasks so that the reader can judge the stability of the reported gains.
[§3, 'Iterations' definition] The term 'iterations' is defined as the number of revised states in the training episode, which conflates the number of editing steps with computational cost. Please consider renaming this column (e.g., 'steps') and reporting a separate compute measure such as tokens generated or FLOPs.
[Appendix B.1] The sentence 'we prompt with the wildtype sequence and sample 10,000 mutants' should state whether these are independent ancestral samples or include beam search, and how the <stop> symbol is handled in the protein task.
[§3.1] The sentence 'In both cases, we report the better-performing variant with scorer' is ambiguous; please specify which variants were chosen and on what basis they were selected.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: qtheta is trained on in-range guide transitions and evaluated on external oracles; self-citations are not load-bearing for the extrapolation claim.

full rationale

The derivation chain is self-contained in the sense required here. qtheta is trained on MCMC transitions selected by guide-score improvements (Section 2.3), but every headline extrapolation result is measured against an external target: FoldX ddG for protein stability (Section 3.1), the oracle sentiment regressor for Yelp reviews (Section 3.2), and the true reward in the toy example (Section 2). These evaluation functions are not the same object as the training episodes, so the reported extrapolation is not forced by construction. The paper explicitly acknowledges the main assumption on which the method depends: 'This guide will only be robust within the range of training values' (Section 2), rather than smuggling it in as a theorem. The self-citations (Khan et al. 2024 for the anonymization energy, and Khan et al. 2021 / Rivera-Soto et al. 2021 for author-verification evaluation) are tool reuse or benchmark replication; the anonymization task is presented as an interpolation control, and no uniqueness theorem or unverified self-cited premise is load-bearing for the central extrapolation claims. The genuine concerns in the paper, namely the episode-creation strategy appearing to be selected after inspecting test metrics (Tables 1 and 4) and the unequal candidate budget in the protein comparison (10,000 qtheta samples in Appendix B.1 versus 83-830 MCMC proposals in Table 1 and Figure 2), are experimental-validity risks rather than reductions of a prediction to its inputs. Accordingly no circular step can be exhibited.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The method depends on several hand-chosen energy weights, sampling budgets, and episode-selection hyperparameters. These control the MCMC training data and are not derived from first principles. No new physical or mathematical entities are introduced; the invented entities list is empty.

free parameters (6)

Protein energy weights = ddG: 500, Hamming: 10
Eq. 1; these weights control the surrogate density used by MCMC and therefore determine which transitions are available as training data for qθ.
Sentiment energy weights = sentiment: 1e5, Hamming: 100
Eq. 2; chosen by hand; the strong sentiment weight drives the sampler toward extreme reviews.
Anonymization energy weights = fluency: 10, Hamming: 1, LUAR: 1e7, SBERT: 5e5
Eq. 3; taken from Khan et al. (2024) with authors overlapping; guide the MCMC episode generation.
Episode selection hyperparameters = lengths 4/5/5; thresholds 20%/2%/1%; thinning factors 2/100/3
Table 12; these control the training episodes for each task and are chosen by validation or by hand.
MCMC sampling epochs = protein 1, sentiment 8, anonymization 40
Table 12; the amount of exploration used to generate training data; directly affects qθ quality.
Reward type = predicted
Main text uses predicted scores instead of oracle or real guide scores at inference, adding a learned score-prediction component to qθ.

assumptions (5)

domain assumption MCMC with an MLM proposal can discover states that improve the guide score
The paper assumes the sampler finds enough improving transitions to make useful training episodes; Section 2.1 and 2.3.
domain assumption The guide scorer provides a reliable directional signal outside its training range
Central premise: improving s(x) is assumed to correlate with improving oracle(x) even in the extrapolation region; Section 2.1.
domain assumption FoldX is a valid oracle for protein stability (ddG)
All protein claims are evaluated against FoldX; Section 3.1.
domain assumption LUAR and SBERT embeddings capture the stylometric and semantic properties needed for anonymization
Used both in the energy function (G.2) and in evaluation (Section 3.3).
standard math Metropolis-Hastings acceptance theory
The sampler uses the standard MH acceptance rule; correctness of the stationary distribution is assumed though not needed for exploration.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Learning Extrapolative Sequence Transformations from Markov Chains." pith.science (2026). https://pith.science/paper/2NNWVV3O

@misc{pith2026250520251,
  author       = {Pith},
  title        = {Pith review of: Learning Extrapolative Sequence Transformations from Markov Chains},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2NNWVV3O}},
  note         = {Machine review of arXiv:2505.20251}
}

read the original abstract

Most successful applications of deep learning involve similar training and test conditions. However, tasks such as biological sequence design involve searching for sequences that improve desirable properties beyond previously known values, which requires novel hypotheses that \emph{extrapolate} beyond training data. In these settings, extrapolation may be achieved by using random search methods such as Markov chain Monte Carlo (MCMC), which, given an initial state, sample local transformations to approximate a target density that rewards states with the desired properties. However, even with a well-designed proposal, MCMC may struggle to explore large structured state spaces efficiently. Rather than relying on stochastic search, it would be desirable to have a model that greedily optimizes the properties of interest, successfully extrapolating in as few steps as possible. We propose to learn such a model from the Markov chains resulting from MCMC search. Specifically, our approach uses selected states from Markov chains as a source of training data for an autoregressive model, which is then able to efficiently generate novel sequences that extrapolate along the sequence-level properties of interest. The proposed approach is validated on three problems: protein sequence design, text sentiment control, and text anonymization. We find that the autoregressive model can extrapolate as well or better than MCMC, but with the additional benefits of scalability and significantly higher sample efficiency.

Figures

Figures reproduced from arXiv: 2505.20251 by the authors.

**Figure 1.** The sentiment extrapolation task (§3.2) requires generating reviews with ratings beyond the range observed at training time. The search process is illustrated using a toy 1D representation of the features (x-axis) and rating (y-axis). Monte Carlo exploration can produce reviews that extrapolate, but many steps are required. However, once good state sequences have been discovered, we can sub-sample the transitions t… view at source ↗

**Figure 2.** In the protein engineering task, comparing MCMC performance (solid line) over ten epochs, or 830 steps, compared to the performance of qθ (dotted line) trained on MCMC data generated on one epoch, or 83 steps. We find that MCMC does not approach the performance of qθ and does not notably improve after even two epochs. 1083 steps, the sequence length 7 [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

13 extracted references · 6 canonical work pages

[4]

doi: 10.18653/v1/2021.naacl-main.415

Association for Computational Linguistics. doi: 10.18653/v1/2021.naacl-main.415. URL https:// aclanthology.org/2021.naacl-main.415. Khan, A., Wang, A., Hager, S., and Andrews, N. Learning to generate text in arbitrary writing styles, 2024. Krishna, K., Song, Y ., Karpinska, M., Wieting, J. F., and Iyyer, M. Paraphrasing evades detectors of AI-generated te...

doi:10.18653/v1/2021.naacl-main.415 2021
[5]

Lewis, M., Liu, Y ., Goyal, N., Ghazvininejad, M., Mo- hamed, A., Levy, O., Stoyanov, V ., and Zettlemoyer, L

URL https://openreview.net/forum? id=WbFhFvjjKj. Lewis, M., Liu, Y ., Goyal, N., Ghazvininejad, M., Mo- hamed, A., Levy, O., Stoyanov, V ., and Zettlemoyer, L. BART: Denoising sequence-to-sequence pre-training for natural language generation, translation, and com- prehension. In Jurafsky, D., Chai, J., Schluter, N., and Tetreault, J. (eds.), Proceedings o...

2020
[9]

Schymkowitz, J., Ferkinghoff-Borg, J., Stricher, F., Nys, R., Rousseau, F., and Serrano, L

URL https://openreview.net/forum? id=KbYevcLjnc. Schymkowitz, J., Ferkinghoff-Borg, J., Stricher, F., Nys, R., Rousseau, F., and Serrano, L. The FoldX web server: An online force field. Nucleic acids research, 33:W382–8, 08 2005. doi: 10.1093/nar/gki387. Sevgen, E., Moller, J., Lange, A., Parker, J., Quigley, S., Mayer, J., Srivastava, P., Gayatri, S., Ho...

doi:10.1093/nar/gki387 2005
[12]

Trabucco, B., Geng, X., Kumar, A., and Levine, S

URL https://openreview.net/forum? id=6ruVLB727MC. Trabucco, B., Geng, X., Kumar, A., and Levine, S. Design- bench: Benchmarks for data-driven offline model-based optimization. In International Conference on Machine Learning, pp. 21658–21676. PMLR, 2022. Wang, A. and Cho, K. BERT has a mouth, and it must speak: BERT as a Markov random field language model....

2022
[31]

acl-long.31

URL https://aclanthology.org/2022. acl-long.31. 11 Learning Extrapolative Sequence Transformations from Markov Chains OpenAI, Achiam, J., Adler, S., Agarwal, S., Ahmad, L., Akkaya, I., Aleman, F. L., Almeida, D., Altenschmidt, J., Altman, S., Anadkat, S., Avila, R., Babuschkin, I., Bal- aji, S., Balcom, V ., Baltescu, P., Bao, H., Bavarian, M., Belgum, J....

arXiv 2022
[115]

Make this {positive, negative }:

URL https://aclanthology.org/2020. findings-emnlp.115. Zhang, X., Zhao, J., and LeCun, Y . Character-level convolu- tional networks for text classification. In Proceedings of the 28th International Conference on Neural Information Processing Systems - Volume 1, NIPS’15, pp. 649–657, Cambridge, MA, USA, 2015. MIT Press. 14 Learning Extrapolative Sequence T...

work page 2024
[280]

findings-emnlp.280/

URL https://aclanthology.org/2022. findings-emnlp.280/. Reimers, N. and Gurevych, I. Sentence-BERT: Sentence embeddings using Siamese BERT-networks. In Inui, K., Jiang, J., Ng, V ., and Wan, X. (eds.), Proceed- ings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th Interna- tional Joint Conference on Natural Language P...

2022
[703]

Approximate Inference with Amortised MCMC

URL https://aclanthology.org/2020. acl-main.703/. Li, Y ., Turner, R. E., and Liu, Q. Approximate inference with amortised MCMC. arXiv preprint arXiv:1702.08343, 2017. Liu, Y ., Ott, M., Goyal, N., Du, J., Joshi, M., Chen, D., Levy, O., Lewis, M., Zettlemoyer, L., and Stoyanov, V . Ro{bert}a: A robustly optimized {bert} pretraining ap- proach, 2020. URL h...

work page Pith review arXiv 2020

Show all 13 references

[2021]

Chen, L., Lu, K., Rajeswaran, A., Lee, K., Grover, A., Laskin, M., Abbeel, P., Srinivas, A., and Mordatch, I

URL https://openreview.net/forum? id=NCDMYD2y5kK. Chen, L., Lu, K., Rajeswaran, A., Lee, K., Grover, A., Laskin, M., Abbeel, P., Srinivas, A., and Mordatch, I. De- cision transformer: reinforcement learning via sequence modeling. In Proceedings of the 35th International Con- f...

2024 doi arXiv
[2022]

G´omez-Rodr´ıguez, C

URL https://openreview.net/forum? id=yCZRdI0Y7G. G´omez-Rodr´ıguez, C. and Williams, P. A confederacy of models: a comprehensive evaluation of LLMs on creative writing. In The 2023 Conference on Empirical Methods in Natural Language Processing, 2023. 10 Learning Extrapolative ...

2023
[2023]

Chakrabarty, T., Laban, P., Agarwal, D., Muresan, S., and Wu, C.-S

URL https://openreview.net/forum? id=9krnQ-ue9M. Chakrabarty, T., Laban, P., Agarwal, D., Muresan, S., and Wu, C.-S. Art or artifice? large language models and the false promise of creativity. In Proceedings of the CHI Conference on Human Factors in Computing Systems, pp. 1–34...

2024
[2024]

Tang, Y ., Tran, C., Li, X., Chen, P.-J., Goyal, N., Chaudhary, V ., Gu, J., and Fan, A

URL https://openreview.net/forum? id=dhFHO90INk. Tang, Y ., Tran, C., Li, X., Chen, P.-J., Goyal, N., Chaudhary, V ., Gu, J., and Fan, A. Multilingual translation with extensible multilingual pretraining and finetuning. 2020. Tay, Y ., Dehghani, M., Tran, V . Q., Garcia, X., W...

2020
[2025]

Swanson, B., Mathewson, K., Pietrzak, B., Chen, S., and Dinalescu, M

URL https://openreview.net/forum? id=4FWAwZtd2n. Swanson, B., Mathewson, K., Pietrzak, B., Chen, S., and Dinalescu, M. Story centaur: Large language model few shot learning as a creative writing tool. In Proceed- ings of the 16th Conference of the European Chapter of the Assoc...

2021

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [4]

doi: 10.18653/v1/2021.naacl-main.415

Association for Computational Linguistics. doi: 10.18653/v1/2021.naacl-main.415. URL https:// aclanthology.org/2021.naacl-main.415. Khan, A., Wang, A., Hager, S., and Andrews, N. Learning to generate text in arbitrary writing styles, 2024. Krishna, K., Song, Y ., Karpinska, M., Wieting, J. F., and Iyyer, M. Paraphrasing evades detectors of AI-generated te...

doi:10.18653/v1/2021.naacl-main.415 2021

[2] [5]

Lewis, M., Liu, Y ., Goyal, N., Ghazvininejad, M., Mo- hamed, A., Levy, O., Stoyanov, V ., and Zettlemoyer, L

URL https://openreview.net/forum? id=WbFhFvjjKj. Lewis, M., Liu, Y ., Goyal, N., Ghazvininejad, M., Mo- hamed, A., Levy, O., Stoyanov, V ., and Zettlemoyer, L. BART: Denoising sequence-to-sequence pre-training for natural language generation, translation, and com- prehension. In Jurafsky, D., Chai, J., Schluter, N., and Tetreault, J. (eds.), Proceedings o...

2020

[3] [9]

Schymkowitz, J., Ferkinghoff-Borg, J., Stricher, F., Nys, R., Rousseau, F., and Serrano, L

URL https://openreview.net/forum? id=KbYevcLjnc. Schymkowitz, J., Ferkinghoff-Borg, J., Stricher, F., Nys, R., Rousseau, F., and Serrano, L. The FoldX web server: An online force field. Nucleic acids research, 33:W382–8, 08 2005. doi: 10.1093/nar/gki387. Sevgen, E., Moller, J., Lange, A., Parker, J., Quigley, S., Mayer, J., Srivastava, P., Gayatri, S., Ho...

doi:10.1093/nar/gki387 2005

[4] [12]

Trabucco, B., Geng, X., Kumar, A., and Levine, S

URL https://openreview.net/forum? id=6ruVLB727MC. Trabucco, B., Geng, X., Kumar, A., and Levine, S. Design- bench: Benchmarks for data-driven offline model-based optimization. In International Conference on Machine Learning, pp. 21658–21676. PMLR, 2022. Wang, A. and Cho, K. BERT has a mouth, and it must speak: BERT as a Markov random field language model....

2022

[5] [31]

acl-long.31

URL https://aclanthology.org/2022. acl-long.31. 11 Learning Extrapolative Sequence Transformations from Markov Chains OpenAI, Achiam, J., Adler, S., Agarwal, S., Ahmad, L., Akkaya, I., Aleman, F. L., Almeida, D., Altenschmidt, J., Altman, S., Anadkat, S., Avila, R., Babuschkin, I., Bal- aji, S., Balcom, V ., Baltescu, P., Bao, H., Bavarian, M., Belgum, J....

arXiv 2022

[6] [115]

Make this {positive, negative }:

URL https://aclanthology.org/2020. findings-emnlp.115. Zhang, X., Zhao, J., and LeCun, Y . Character-level convolu- tional networks for text classification. In Proceedings of the 28th International Conference on Neural Information Processing Systems - Volume 1, NIPS’15, pp. 649–657, Cambridge, MA, USA, 2015. MIT Press. 14 Learning Extrapolative Sequence T...

work page 2024

[7] [280]

findings-emnlp.280/

URL https://aclanthology.org/2022. findings-emnlp.280/. Reimers, N. and Gurevych, I. Sentence-BERT: Sentence embeddings using Siamese BERT-networks. In Inui, K., Jiang, J., Ng, V ., and Wan, X. (eds.), Proceed- ings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th Interna- tional Joint Conference on Natural Language P...

2022

[8] [703]

Approximate Inference with Amortised MCMC

URL https://aclanthology.org/2020. acl-main.703/. Li, Y ., Turner, R. E., and Liu, Q. Approximate inference with amortised MCMC. arXiv preprint arXiv:1702.08343, 2017. Liu, Y ., Ott, M., Goyal, N., Du, J., Joshi, M., Chen, D., Levy, O., Lewis, M., Zettlemoyer, L., and Stoyanov, V . Ro{bert}a: A robustly optimized {bert} pretraining ap- proach, 2020. URL h...

work page Pith review arXiv 2020