Pith. sign in

REVIEW 4 major objections 5 minor 59 references

Too Big to Fool: Resisting Deception in Language Models

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Larger language models are harder to fool than their smaller counterparts.

desk verdict Larger models may resist deception on average, but the paper's 'consistent' scaling claim doesn't survive its own per-benchmark tables. read the letter →

arxiv 2412.10558 v1 pith:2A4ISMF3 submitted 2024-12-13 cs.CL cs.AIcs.LG

Mohammad Reza Samsami , Mats Leon Richter , Juan Rodriguez , Megh Thakkar , Sarath Chandar , Maxime Gasse This is my paper · ORCID
classification cs.CLcs.AIcs.LG
keywords deceptionmisinformationrelativeaccuracydropin-contextinformationmodelscalingworldinstructionfollowing
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

Large language models must weigh what they learned during training against information written in the prompt. The paper claims that, within the same model family, the larger model keeps more of its accuracy when the prompt contains an incorrect hint, and that this relative advantage is not explained by ignoring hints or by memorization. If true, scaling makes models harder to fool by injected misinformation while still leaving them responsive to genuine instructions.

What carries the argument

The central measurement is the Relative Accuracy Drop, the accuracy loss under a prompt alteration divided by the original accuracy; it allows drops to be compared across models and benchmarks of differing difficulty. The evaluation procedure standardizes every benchmark to the MMLU prompt format and applies four alterations — a deceptive hint, a truthful hint, a directive instruction to answer incorrectly, and removal of the question — so that each comparison isolates one way a model can be steered. The contrast between deceptive and directive conditions is what lets the paper claim scale improves both skepticism and instruction following.

What would settle it

Train or obtain a small and a large model that share the same architecture, data, and instruction-tuning pipeline and differ only in width, depth, or parameter count; if the larger model's Relative Accuracy Drop under a false hint is not smaller than the smaller model's, the claimed scaling law is refuted. A cheaper check is to re-run the deception experiment on a matched pair where the small model was trained on exactly the same data as the large one.

Watch

Extended reading notes

Core claim

Across four open-weight model families, pairing each small model with a larger sibling, the paper finds that larger models show a smaller Relative Accuracy Drop — the fractional loss defined as (original accuracy minus altered accuracy) divided by original accuracy — when a false answer hint is appended to multiple-choice questions. Control experiments show all models exploit truthful hints nearly perfectly, and larger models follow explicit wrong-answer instructions at least as well as smaller ones, ruling out the idea that resilience comes from disregarding prompt content. A contamination experiment, comparing a model with no possible exposure to the benchmark against one deliberately overfitted on the test set, finds both stay above chance when the question is removed, suggesting the resilience reflects inference from choices and world knowledge rather than rote memorization.

Load-bearing premise

The load-bearing premise is that comparing Llama-8B against Llama-70B, Gemma-2B against Gemma-9B, Phi-mini against Phi-medium, and Mistral-7B against Mixtral-8x22B isolates the effect of parameter count alone, even though the paired models differ in architecture, training data, tokenizer, and fine-tuning recipe.

Editorial extensions

If this is right

  • Larger open-weight models will be proportionally less degraded by injected false hints on multiple-choice benchmarks.
  • The resilience is not bought by ignoring prompts: accurate hints still raise accuracy nearly to ceiling in all models, so the improvement is in how hints are screened.
  • Instruction-following improves with scale on this setup, meaning larger models can be directed to wrong answers when explicitly asked, even as they resist unsupported hints.
  • Models without benchmark contamination behave like overfitted ones when the question is removed, so question-removal accuracy is not evidence of memorization; future benchmark audits should control for choice-only inference.
  • Scaling is a partial, not complete, defense against misinformation, and the findings motivate studying malicious hints in open-ended generation.

Reading between the lines

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

  • Inference: the paper's hints are uniform and simple; realistic misinformation is more nuanced, so the size advantage may shrink or reverse on contextually believable false hints, and a targeted study is needed.
  • Inference: because the smaller models' accuracy under deception often falls near chance, the relative drop metric may partly reflect a floor effect; absolute drop plots in the paper show the same direction, but the metric's normalization should be stress-tested on models of matched competence.
  • Inference: one testable extension would be to repeat the deception experiment with the hint presented as an authoritative source (e.g., a domain expert states...) to see whether larger models become more gullible to authority framing.
  • Inference: the same prompt-alteration suite could be applied to open-ended generations with judge-based correctness, rather than multiple choice, to see whether the scaling pattern generalizes.
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.

Referee Report

4 major / 5 minor

Summary. The paper evaluates eight open-source instruction-tuned LLMs from four families (Llama-3.1-8B/70B, Gemma-2-2B/9B, Phi-3 Mini/Medium, Mistral-7B/Mixtral-8x22B) on multiple-choice benchmarks, altering prompts by adding a false hint (Deception), a true hint (Guidance), a directive to answer incorrectly (Directive Instruction), or removing the question (Context Removal). Its central claim is that larger models show smaller Relative Accuracy Drop under Deception and are therefore more resilient to misleading in-context information, and that this resilience is not due to ignoring hints or to memorization/contamination. The paper also discusses world-model interpretations and includes qualitative generation examples.

Significance. If the main claim were established, the paper would be a useful empirical contribution on how scale affects susceptibility to misleading in-context information, with implications for model evaluation and safety. The experimental setup is transparent and inexpensive: it reuses standard benchmarks, uses open models, and the Guidance control (near-perfect accuracy with true hints) is a clean way to show that larger models do not simply ignore prompt content. The Context Removal and overfitting comparison is a constructive idea for probing memorization. However, the evidence as reported does not support the strong, unconditional phrasing in Section 4.1: the paper's own per-benchmark tables contain reversals, and no confidence intervals or significance tests are provided. The causal attribution to parameter scale is also not established by the chosen model pairs.

major comments (4)
  1. [Section 4.1, Figure 2, Tables 10-11] The central claim that "within each model family, we consistently observe that the Relative Accuracy Drop is smaller for larger models" is contradicted by the paper's own Appendix E. From Table 10 (Logiqa2), the Relative Accuracy Drop is larger for the larger model in every family: Llama 8B→70B gives (0.55−0.29)/0.55=0.473 vs. (0.71−0.32)/0.71=0.549; Gemma 2B→9B gives 0.750 vs. 0.923; Phi Mini→Medium gives 0.316 vs. 0.444; Mistral 7B→Mixtral gives 0.615 vs. 0.678. Table 11 (MathQA) reverses for Llama (0.793 vs. 1.000) and Gemma (0.714 vs. 0.955). Because the claim is stated unconditionally and Figure 2 plots only the aggregate mean (with an undefined shaded "deviation" and no confidence intervals or significance tests), the reversals are load-bearing: the result currently holds in the aggregate, not consistently. The authors should report per-benchmark effect sizes with uncertainty, pre-specify the summary statistic, or substantially weaken the claim. Floor effects make this particularly important: in GPQA and MathQA the Deception condition drives several accuracies to or near zero, so the relative-drop ratio is unstable.
  2. [Section 3.5, Equations for Accuracy Drop and Relative Accuracy Drop] The headline metric, Relative Accuracy Drop = (Original Accuracy − Altered Accuracy) / Original Accuracy, is normalized by baseline accuracy. Since larger models generally have higher original accuracy, this metric partially builds in the paper's conclusion: even with exactly equal absolute drops, the larger model will have a smaller relative drop. Absolute drops (Figure 7) are more favorable to the claim but are also not universal; for example, Table 11 (MathQA) shows Llama 8B's absolute drop is 0.23 while Llama 70B's is 0.40, and Table 7 (HellaSwag) shows Phi Medium with a lower original accuracy than Phi Mini, so the baseline ordering itself is not consistent. The analysis should report both metrics with confidence intervals and justify the normalization choice rather than presenting it as the only natural standardization.
  3. [Section 3.3, Models] The paper says the model pairs are chosen "to isolate the effect of scale on model performance," but the pairs differ in more than parameter count. Mistral-7B is a dense model while Mixtral-8x22B is a mixture-of-experts model with a different total parameter count and different active-parameter behavior; the Llama, Gemma, and Phi pairs also differ in training data, instruction-tuning recipes, tokenizers, and possibly architecture details. Parameter scale is therefore confounded with family-specific design choices, so the observed differences cannot be causally attributed to scale alone. The paper should reframe the results as within-family capacity trends, or use a controlled comparison (e.g., same architecture and data with different widths/depths) to support the "as models scale" language in the conclusion.
  4. [Section 4.3, Context Removal and Overfitting Details] The memorization control does not directly test the alternative explanation it targets. The Context Removal experiment and the overfitted-Llama/DCLM-7B comparison show that both a contaminated model and a clean model can answer some MMLU items from the answer choices alone, but they do not show that test-set contamination does not contribute to the larger models' smaller Deception drops. In particular, the overfitted model is not evaluated under the Deception condition, and the comparison is not made between small and large models with matched contamination. The overfitting experiment also uses Llama-3.1-8B only, not the larger models whose resilience is at issue. The paper concedes "we cannot entirely dismiss the possibility that memorization contributes," but Section 5 concludes the resilience is "unlikely due to memorization." This overreach should be fixed by either adding a direct deception experiment on contaminated vs. clean models or by removing the causal claim.
minor comments (5)
  1. [Table 1 and Appendix E] Table 1 lists nine benchmarks, but Appendix E and Figures 6-7 include Logiqa2, which is never described in the benchmark table; this omission should be fixed for reproducibility.
  2. [Figure 7, Logiqa2 panel] The arrow labels in the Logiqa2 panel do not match Table 10: Phi-Mini is labeled −39.7% and Mixtral −17.7%, whereas Table 10 implies drops of approximately 18% and 40%, respectively; the label-to-model mapping appears shifted.
  3. [Section 3.5, metric example] The metric example contains missing spaces: "would be6.25%and 8.33%" should read "would be 6.25% and 8.33%".
  4. [Figure 5 caption and Appendix B] Figure 5 shows training epochs up to 30, while Appendix B states the training loop was executed for 50 epochs; the relationship between these numbers should be clarified, and the caption contains the typo "ALlama-3.1-8B-Instruct".
  5. [Section 4.2 and Abstract] Section 4.2 correctly acknowledges that Gemma is an outlier for instruction-following, but the abstract and Section 5 present instruction-following improvements without that qualifier; the softer claim should be carried through the summary sections.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the adversarial-prompt evaluation and the Relative Accuracy Drop metric are not fitted to the conclusion they support.

full rationale

The paper is an empirical scaling study; it fits no parameters, trains no models whose outputs are then 'predicted,' and invokes no author-specific uniqueness theorem or prior result as the load-bearing premise. The central comparison is measured directly from benchmark accuracies (e.g., Tables 2-11), and the headline metric, Relative Accuracy Drop = (Original - Altered) / Original, is a normalization choice, not a quantity fitted to the data. Although the normalization can interact with higher baselines in larger models, the paper also reports absolute Accuracy Drops (Figure 7) and per-benchmark results, so the conclusion is not forced by the metric alone. The 'world model' framing is supported by external citations (e.g., Gurnee and Tegmark, Li et al.) rather than by the authors' own prior work, and the control experiments (Guidance, Directive Instruction, Context Removal) are independent empirical checks. Whether the per-benchmark reversals in Appendix E undercut the 'consistent' scaling claim is a question of correctness and statistical robustness, not circularity, and therefore does not raise the circularity score.

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

No fitted parameters are used. The assumptions are the comparability of model pairs, the validity of log-likelihood scoring, the contamination-free status of DCLM-7B, and the interpretive world-model framing. No new entities are introduced; world model is adopted from prior literature.

assumptions (5)
  • domain assumption Choosing the answer label with the highest log-likelihood is a valid measure of model performance.
    Section 3.5 uses log-likelihood across choice labels as the response selection rule; the paper cites Lyu et al. (2024) showing this may not match human-perceived performance, so this is an assumption about the evaluation.
  • domain assumption Within each family, models differ in a way that isolates parameter scale.
    Section 3.3 pairs Llama-8B/70B, Gemma-2B/9B, Phi-mini/medium, and Mistral-7B/Mixtral-8x22B, treating them as same-family size pairs, but architectures and training recipes differ, and Mixtral is a mixture of experts.
  • domain assumption DCLM-7B had no prior exposure to MMLU.
    Section 4.3 relies on this guarantee from Li et al. (2024a) to argue memorization is not the cause; if the guarantee is false or the base-versus-instruction-tuned mismatch matters, the control is invalid.
  • domain assumption The MMLU answer choices contain enough implicit information to make context removal a meaningful memorization test.
    Section 4.3 interprets above-chance performance without the question as implying non-memorization mechanisms, but the authors acknowledge answer choices themselves carry information.
  • domain assumption LLMs possess an internal world model, as opposed to being statistical pattern matchers.
    Appendix F grounds the paper in the world-model hypothesis; the empirical measurements do not by themselves prove this interpretation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Too Big to Fool: Resisting Deception in Language Models." pith.science (2026). https://pith.science/paper/2A4ISMF3

@misc{pith2026241210558,
  author       = {Pith},
  title        = {Pith review of: Too Big to Fool: Resisting Deception in Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2A4ISMF3}},
  note         = {Machine review of arXiv:2412.10558}
}
read the original abstract

Large language models must balance their weight-encoded knowledge with in-context information from prompts to generate accurate responses. This paper investigates this interplay by analyzing how models of varying capacities within the same family handle intentionally misleading in-context information. Our experiments demonstrate that larger models exhibit higher resilience to deceptive prompts, showcasing an advanced ability to interpret and integrate prompt information with their internal knowledge. Furthermore, we find that larger models outperform smaller ones in following legitimate instructions, indicating that their resilience is not due to disregarding in-context information. We also show that this phenomenon is likely not a result of memorization but stems from the models' ability to better leverage implicit task-relevant information from the prompt alongside their internally stored knowledge.

Figures

Figures reproduced from arXiv: 2412.10558 by the authors.

Figure 1
Figure 1. Overview of our evaluation methodology. We begin by selecting a multiple-choice benchmark dataset using the Language Model Evaluation Harness framework (Gao et al., 2024). Samples are then processed through two methods: Prompt Unification, which standardizes the prompt structure using the MMLU format, and Prompt Alteration, where content is added or removed in the prompt (see Section 3.2). Each altered prompt is fin… view at source ↗
Figure 2
Figure 2. Relative Accuracy Drop under the Deception. Bold lines are the main indicators, representing the average Relative Accuracy Drop across all benchmarks, with shaded regions show￾ing the deviation. Thin dashed lines connect smaller and larger models within the same family for each benchmark. The results demonstrate that larger models consistently exhibit a smaller Relative Accuracy Drop, indicating greater robustness t… view at source ↗
Figure 3
Figure 3. Relative Accuracy Drop under the Directive Instruction. Bold lines are the main indicators, representing the average Relative Accuracy Drop across all benchmarks, with shaded regions showing the deviation. Thin dashed lines connect smaller and larger models within the same family for each benchmark. When explicitly instructed to pick a wrong answer instead of the correct one, larger models of each family tend to exh… view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Accuracy Drop under the Context Removal. Accuracy of each model on the original (•) and altered (×) MMLU benchmark, ordered by original performance. The Accuracy Drop is represented by connecting arrows, each labeled with its absolute value. All models except Gemma-2-2…
Figure 5
Figure 5. Figure 5: Overfitting and Context Removal. Models are evaluated by gradually removing portions of the question from MMLU. A Llama-3.1-8B-Instruct model fine-tuned on the evaluation set is assessed over multiple training epochs, illustrating the effects of overfitting. The DCLM-7…
Figure 6
Figure 6. Figure 6: Resilience to Deception across Individual Benchmarks. Relative Accuracy Drop is calculated as original−altered original for each model family, size, and dataset. Each subplot represents one benchmark, with lines connecting models of different sizes within the same fami…
Figure 7
Figure 7. Figure 7: Deception Experiment Accuracy across Individual Benchmarks. Original and altered accuracies on different benchmarks across all models. For each model, the base accuracy is plotted by a •, while the altered accuracy is shown with a ×. The Accuracy Drop is represented by…
Figure 8
Figure 8. Figure 8: Instruction-following across Individual Benchmarks. Relative Accuracy Drop is calculated as original−altered original for each model family, size, and dataset. Each subplot represents one benchmark, with lines connecting models of different sizes within the same family…
Figure 9
Figure 9. Figure 9: Instruction-following Accuracy across Individual Benchmarks. Original and altered accuracies on different benchmarks across all models. For each model, the base accuracy is plotted by a •, while the altered accuracy is shown with a ×. The Accuracy Drop is represented b…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

59 extracted references · 10 canonical work pages

  1. [1]

    GQA : Training generalized multi-query transformer models from multi-head checkpoints

    Joshua Ainslie, James Lee-Thorp, Michiel de Jong, Yury Zemlyanskiy, Federico Lebron, and Sumit Sanghai. GQA : Training generalized multi-query transformer models from multi-head checkpoints. In Houda Bouamor, Juan Pino, and Kalika Bali (eds.), Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pp.\ 4895--4901, Singapor...

    doi:10.18653/v1/2023.emnlp-main.298 2023

  2. [2]

    When benchmarks are targets: Revealing the sensitivity of large language model leaderboards, 2024

    Norah Alzahrani, Hisham Abdullah Alyahya, Yazeed Alnumay, Sultan Alrashed, Shaykhah Alsubaie, Yusef Almushaykeh, Faisal Mirza, Nouf Alotaibi, Nora Altwairesh, Areeb Alowisheq, M Saiful Bari, and Haidar Khan. When benchmarks are targets: Revealing the sensitivity of large language model leaderboards, 2024. URL https://arxiv.org/abs/2402.01781

    arXiv 2024

  3. [3]

    Mathqa: Towards interpretable math word problem solving with operation-based formalisms, 2019

    Aida Amini, Saadia Gabriel, Peter Lin, Rik Koncel-Kedziorski, Yejin Choi, and Hannaneh Hajishirzi. Mathqa: Towards interpretable math word problem solving with operation-based formalisms, 2019. URL https://arxiv.org/abs/1905.13319

    arXiv 2019

  4. [4]

    Self-supervised learning from images with a joint-embedding predictive architecture

    Mahmoud Assran, Quentin Duval, Ishan Misra, Piotr Bojanowski, Pascal Vincent, Michael Rabbat, Yann LeCun, and Nicolas Ballas. Self-supervised learning from images with a joint-embedding predictive architecture. arXiv preprint arXiv:2301.08243, 2023

    arXiv 2023

  5. [5]

    Revisiting feature prediction for learning visual representations from video, 2024

    Adrien Bardes, Quentin Garrido, Jean Ponce, Xinlei Chen, Michael Rabbat, Yann LeCun, Mahmoud Assran, and Nicolas Ballas. Revisiting feature prediction for learning visual representations from video, 2024. URL https://arxiv.org/abs/2404.08471

    arXiv 2024

  6. [6]

    Bender and Alexander Koller

    Emily M. Bender and Alexander Koller. Climbing towards NLU : On meaning, form, and understanding in the age of data. In Dan Jurafsky, Joyce Chai, Natalie Schluter, and Joel Tetreault (eds.), Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pp.\ 5185--5198, Online, July 2020. Association for Computational Linguistics...

    doi:10.18653/v1/2020.acl-main.463 2020

  7. [7]

    Bender, Timnit Gebru, Angelina McMillan-Major, and Shmargaret Shmitchell

    Emily M. Bender, Timnit Gebru, Angelina McMillan-Major, and Shmargaret Shmitchell. On the dangers of stochastic parrots: Can language models be too big? In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency , New York, March 2021. Association for Computer Machinery – ACM . doi:10.1145/3442188.3445922. URL https://dl.acm.o...

    arXiv 2021

  8. [8]

    Piqa: Reasoning about physical commonsense in natural language, 2019

    Yonatan Bisk, Rowan Zellers, Ronan Le Bras, Jianfeng Gao, and Yejin Choi. Piqa: Reasoning about physical commonsense in natural language, 2019. URL https://arxiv.org/abs/1911.11641

    arXiv 2019

Show all 59 references
  1. [9]

    Experience grounds language, 2020

    Yonatan Bisk, Ari Holtzman, Jesse Thomason, Jacob Andreas, Yoshua Bengio, Joyce Chai, Mirella Lapata, Angeliki Lazaridou, Jonathan May, Aleksandr Nisnevich, Nicolas Pinto, and Joseph Turian. Experience grounds language, 2020. URL https://arxiv.org/abs/2004.10151

    2020 arXiv

  2. [10]

    Hudson, and Ehsan Adeli et al

    Rishi Bommasani, Drew A. Hudson, and Ehsan Adeli et al. On the opportunities and risks of foundation models, 2022. URL https://arxiv.org/abs/2108.07258

    2022 arXiv

  3. [11]

    Breaking down the defenses: A comparative survey of attacks on large language models, 2024

    Arijit Ghosh Chowdhury, Md Mofijul Islam, Vaibhav Kumar, Faysal Hossain Shezan, Vaibhav Kumar, Vinija Jain, and Aman Chadha. Breaking down the defenses: A comparative survey of attacks on large language models, 2024. URL https://arxiv.org/abs/2403.04786

    2024

  4. [12]

    Think you have solved question answering? try arc, the ai2 reasoning challenge, 2018

    Peter Clark, Isaac Cowhey, Oren Etzioni, Tushar Khot, Ashish Sabharwal, Carissa Schoenick, and Oyvind Tafjord. Think you have solved question answering? try arc, the ai2 reasoning challenge, 2018. URL https://arxiv.org/abs/1803.05457

    2018 arXiv

  5. [13]

    Language modeling is compression, 2024

    Grégoire Delétang, Anian Ruoss, Paul-Ambroise Duquenne, Elliot Catt, Tim Genewein, Christopher Mattern, Jordi Grau-Moya, Li Kevin Wenliang, Matthew Aitchison, Laurent Orseau, Marcus Hutter, and Joel Veness. Language modeling is compression, 2024. URL https://arxiv.org/abs/2309.10668

    2024 arXiv

  6. [14]

    A framework for few-shot language model evaluation, 07 2024

    Leo Gao, Jonathan Tow, Baber Abbasi, Stella Biderman, Sid Black, Anthony DiPofi, Charles Foster, Laurence Golding, Jeffrey Hsu, Alain Le Noac'h, Haonan Li, Kyle McDonell, Niklas Muennighoff, Chris Ociepa, Jason Phang, Laria Reynolds, Hailey Schoelkopf, Aviya Skowron, Lintang S...

    2024

  7. [15]

    Learning and leveraging world models in visual representation learning, 2024

    Quentin Garrido, Mahmoud Assran, Nicolas Ballas, Adrien Bardes, Laurent Najman, and Yann LeCun. Learning and leveraging world models in visual representation learning, 2024. URL https://arxiv.org/abs/2403.00504

    2024 arXiv

  8. [16]

    Are we done with mmlu?, 2024

    Aryo Pradipta Gema, Joshua Ong Jun Leang, Giwon Hong, Alessio Devoto, Alberto Carlo Maria Mancino, Rohit Saxena, Xuanli He, Yu Zhao, Xiaotang Du, Mohammad Reza Ghasemi Madani, Claire Barale, Robert McHardy, Joshua Harris, Jean Kaddour, Emile van Krieken, and Pasquale Minervini...

    2024

  9. [17]

    Gemma: Open models based on gemini research and technology, 2024

    Google. Gemma: Open models based on gemini research and technology, 2024. URL https://arxiv.org/abs/2403.08295

    2024 arXiv

  10. [18]

    Changing answer order can decrease mmlu accuracy, 2024

    Vipul Gupta, David Pantoja, Candace Ross, Adina Williams, and Megan Ung. Changing answer order can decrease mmlu accuracy, 2024. URL https://arxiv.org/abs/2406.19470

    2024 arXiv

  11. [19]

    Language models represent space and time, 2024

    Wes Gurnee and Max Tegmark. Language models represent space and time, 2024. URL https://arxiv.org/abs/2310.02207

    2024 arXiv

  12. [20]

    World models

    David Ha and Jürgen Schmidhuber. World models. 2018. doi:10.5281/ZENODO.1207631. URL https://zenodo.org/record/1207631

    2018

  13. [21]

    Learning latent dynamics for planning from pixels

    Danijar Hafner, Timothy Lillicrap, Ian Fischer, Ruben Villegas, David Ha, Honglak Lee, and James Davidson. Learning latent dynamics for planning from pixels. In International Conference on Machine Learning, pp.\ 2555--2565, 2019

    2019

  14. [22]

    Reasoning with language model is planning with world model

    Shibo Hao, Yi Gu, Haodi Ma, Joshua Hong, Zhen Wang, Daisy Wang, and Zhiting Hu. Reasoning with language model is planning with world model. In Houda Bouamor, Juan Pino, and Kalika Bali (eds.), Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processi...

    2023 doi

  15. [23]

    Measuring massive multitask language understanding, 2021

    Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. Measuring massive multitask language understanding, 2021

    2021

  16. [24]

    Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen

    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

    2021 arXiv

  17. [25]

    Albert Q. Jiang, Alexandre Sablayrolles, Antoine Roux, Arthur Mensch, Blanche Savary, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Emma Bou Hanna, Florian Bressand, Gianna Lengyel, Guillaume Bour, Guillaume Lample, Lélio Renard Lavaud, Lucile Saulnier, Marie-Anne...

    2024 arXiv

  18. [26]

    Strengthening llm trust boundaries: A survey of prompt injection attacks

    Surender Suresh Kumar, Missy Cummings, and Alexander Stimpson. Strengthening llm trust boundaries: A survey of prompt injection attacks. In IEEE 4th International Conference on Human Machine Systems (ICHMS), 02 2024

    2024

  19. [27]

    A path towards autonomous machine intelligence version 0.9

    Yann LeCun. A path towards autonomous machine intelligence version 0.9. 2, 2022-06-27. Open Review, 62 0 (1): 0 1--62, 2022

    2022

  20. [28]

    Li, Maxwell Nye, and Jacob Andreas

    Belinda Z. Li, Maxwell Nye, and Jacob Andreas. Implicit representations of meaning in neural language models, 2021. URL https://arxiv.org/abs/2106.00737

    2021 arXiv

  21. [29]

    Dimakis, Yair Carmon, Achal Dave, Ludwig Schmidt, and Vaishaal Shankar

    Jeffrey Li, Alex Fang, Georgios Smyrnis, Maor Ivgi, Matt Jordan, Samir Gadre, Hritik Bansal, Etash Guha, Sedrick Keh, Kushal Arora, Saurabh Garg, Rui Xin, Niklas Muennighoff, Reinhard Heckel, Jean Mercat, Mayee Chen, Suchin Gururangan, Mitchell Wortsman, Alon Albalak, Yonatan ...

    2024 arXiv

  22. [30]

    Hopkins, David Bau, Fernanda Viégas, Hanspeter Pfister, and Martin Wattenberg

    Kenneth Li, Aspen K. Hopkins, David Bau, Fernanda Viégas, Hanspeter Pfister, and Martin Wattenberg. Emergent world representations: Exploring a sequence model trained on a synthetic task, 2024 b . URL https://arxiv.org/abs/2210.13382

    2024 arXiv

  23. [31]

    Truthfulqa: Measuring how models mimic human falsehoods

    Stephanie Lin, Jacob Hilton, and Owain Evans. Truthfulqa: Measuring how models mimic human falsehoods. In Smaranda Muresan, Preslav Nakov, and Aline Villavicencio (eds.), ACL, pp.\ 3214--3252. Association for Computational Linguistics, 2022. ISBN 978-1-955917-21-6

    2022

  24. [32]

    Beyond probabilities: Unveiling the misalignment in evaluating large language models, 2024

    Chenyang Lyu, Minghao Wu, and Alham Fikri Aji. Beyond probabilities: Unveiling the misalignment in evaluating large language models, 2024. URL https://arxiv.org/abs/2402.13887

    2024 arXiv

  25. [33]

    Openeqa: Embodied question answering in the era of foundation models

    Arjun Majumdar, Anurag Ajay, Xiaohan Zhang, Pranav Putta, Sriram Yenamandra, Mikael Henaff, Sneha Silwal, Paul Mcvay, Oleksandr Maksymets, Sergio Arnaud, Karmesh Yadav, Qiyang Li, Ben Newman, Mohit Sharma, Vincent Berges, Shiqi Zhang, Pulkit Agrawal, Yonatan Bisk, Dhruv Batra,...

    2024

  26. [34]

    The llama 3 herd of models, 2024

    Meta. The llama 3 herd of models, 2024. URL https://arxiv.org/abs/2407.21783

    2024 arXiv

  27. [35]

    Phi-3 technical report: A highly capable language model locally on your phone, 2024

    Microsoft. Phi-3 technical report: A highly capable language model locally on your phone, 2024. URL https://arxiv.org/abs/2404.14219

    2024 arXiv

  28. [36]

    Emergent linear representations in world models of self-supervised sequence models, 2023

    Neel Nanda, Andrew Lee, and Martin Wattenberg. Emergent linear representations in world models of self-supervised sequence models, 2023. URL https://arxiv.org/abs/2309.00941

    2023 arXiv

  29. [37]

    Gpt-4 technical report, 2024

    OpenAI. Gpt-4 technical report, 2024. URL https://arxiv.org/abs/2303.08774

    2024 arXiv

  30. [38]

    Mapping language models to grounded conceptual spaces

    Roma Patel and Ellie Pavlick. Mapping language models to grounded conceptual spaces. In International conference on learning representations, 2022

    2022

  31. [39]

    David Rein, Betty Li Hou, Asa Cooper Stickland, Jackson Petty, Richard Yuanzhe Pang, Julien Dirani, Julian Michael, and Samuel R. Bowman. GPQA: A Graduate-Level Google-Proof Q&A Benchmark , 2023

    2023

  32. [40]

    An early categorization of prompt injection attacks on large language models, 2024

    Sippo Rossi, Alisia Marianne Michel, Raghava Rao Mukkamala, and Jason Bennett Thatcher. An early categorization of prompt injection attacks on large language models, 2024. URL https://arxiv.org/abs/2402.00898

    2024 arXiv

  33. [41]

    Are emergent abilities of large language models a mirage? In Thirty-seventh Conference on Neural Information Processing Systems, 2023

    Rylan Schaeffer, Brando Miranda, and Sanmi Koyejo. Are emergent abilities of large language models a mirage? In Thirty-seventh Conference on Neural Information Processing Systems, 2023. URL https://openreview.net/forum?id=ITw9edRDlD

    2023

  34. [42]

    Integrated architectures for learning, planning, and reacting based on approximating dynamic programming

    Richard S Sutton. Integrated architectures for learning, planning, and reacting based on approximating dynamic programming. In Machine learning proceedings 1990, pp.\ 216--224. Elsevier, 1990

    1990

  35. [43]

    C ommonsense QA : A question answering challenge targeting commonsense knowledge

    Alon Talmor, Jonathan Herzig, Nicholas Lourie, and Jonathan Berant. C ommonsense QA : A question answering challenge targeting commonsense knowledge. In Jill Burstein, Christy Doran, and Thamar Solorio (eds.), Proceedings of the 2019 Conference of the North A merican Chapter o...

    2019 doi

  36. [44]

    Bin Wang, Chengwei Wei, Zhengyuan Liu, Geyu Lin, and Nancy F. Chen. Resilience of large language models for noisy instructions, 2024 a . URL https://arxiv.org/abs/2404.09754

    2024 arXiv

  37. [45]

    Adversarial glue: A multi-task benchmark for robustness evaluation of language models

    Boxin Wang, Chejian Xu, Shuohang Wang, Shuohang Wang, Zhe Gan, Yu Cheng, Jianfeng Gao, Ahmed Awadallah, and Bo Li. Adversarial glue: A multi-task benchmark for robustness evaluation of language models. In J. Vanschoren and S. Yeung (eds.), Proceedings of the Neural Information...

    2021

  38. [46]

    Look at the text: Instruction-tuned language models are more robust multiple choice selectors than you think, 2024 b

    Xinpeng Wang, Chengzhi Hu, Bolei Ma, Paul Röttger, and Barbara Plank. Look at the text: Instruction-tuned language models are more robust multiple choice selectors than you think, 2024 b . URL https://arxiv.org/abs/2404.08382

    2024 arXiv

  39. [47]

    Mmlu-pro: A more robust and challenging multi-task language understanding benchmark, 2024 c

    Yubo Wang, Xueguang Ma, Ge Zhang, Yuansheng Ni, Abhranil Chandra, Shiguang Guo, Weiming Ren, Aaran Arulraj, Xuan He, Ziyan Jiang, Tianle Li, Max Ku, Kai Wang, Alex Zhuang, Rongqi Fan, Xiang Yue, and Wenhu Chen. Mmlu-pro: A more robust and challenging multi-task language unders...

    2024

  40. [48]

    Unveiling selection biases: Exploring order and token sensitivity in large language models, 2024

    Sheng-Lun Wei, Cheng-Kuang Wu, Hen-Hsen Huang, and Hsin-Hsi Chen. Unveiling selection biases: Exploring order and token sensitivity in large language models, 2024. URL https://arxiv.org/abs/2406.03009

    2024 arXiv

  41. [49]

    Liu, and Matt Gardner

    Johannes Welbl, Nelson F. Liu, and Matt Gardner. Crowdsourcing multiple choice science questions. In NUT@EMNLP, 2017

    2017

  42. [50]

    Auto-gpt for online decision making: Benchmarks and additional opinions, 2023

    Hui Yang, Sifu Yue, and Yunzhong He. Auto-gpt for online decision making: Benchmarks and additional opinions, 2023. URL https://arxiv.org/abs/2306.02224

    2023 arXiv

  43. [51]

    Assessing prompt injection risks in 200+ custom gpts, 2024

    Jiahao Yu, Yuhang Wu, Dong Shu, Mingyu Jin, Sabrina Yang, and Xinyu Xing. Assessing prompt injection risks in 200+ custom gpts, 2024. URL https://arxiv.org/abs/2311.11538

    2024 arXiv

  44. [52]

    Hellaswag: Can a machine really finish your sentence?, 2019

    Rowan Zellers, Ari Holtzman, Yonatan Bisk, Ali Farhadi, and Yejin Choi. Hellaswag: Can a machine really finish your sentence?, 2019. URL https://arxiv.org/abs/1905.07830

    2019 arXiv

  45. [53]

    Large language models are not robust multiple choice selectors, 2024

    Chujie Zheng, Hao Zhou, Fandong Meng, Jie Zhou, and Minlie Huang. Large language models are not robust multiple choice selectors, 2024. URL https://arxiv.org/abs/2309.03882

    2024 arXiv

  46. [54]

    Promptbench: A unified library for evaluation of large language models

    Kaijie Zhu, Qinlin Zhao, Hao Chen, Jindong Wang, and Xing Xie. Promptbench: A unified library for evaluation of large language models. J. Mach. Learn. Res., 25: 0 254:1--254:22, 2024. URL https://jmlr.org/papers/v25/24-0023.html

    2024

  47. [55]

    Fool your (vision and) language model with embarrassingly simple permutations, 2024

    Yongshuo Zong, Tingyang Yu, Ruchika Chavhan, Bingchen Zhao, and Timothy Hospedales. Fool your (vision and) language model with embarrassingly simple permutations, 2024. URL https://arxiv.org/abs/2310.01651

    2024 arXiv

  48. [56]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

  49. [57]

    @esa (Ref

    \@ifxundefined[1] #1\@undefined \@firstoftwo \@secondoftwo \@ifnum[1] #1 \@firstoftwo \@secondoftwo \@ifx[1] #1 \@firstoftwo \@secondoftwo [2] @ #1 \@temptokena #2 #1 @ \@temptokena \@ifclassloaded agu2001 natbib The agu2001 class already includes natbib coding, so you should ...

  50. [58]

    \@lbibitem[] @bibitem@first@sw\@secondoftwo \@lbibitem[#1]#2 \@extra@b@citeb \@ifundefined br@#2\@extra@b@citeb \@namedef br@#2 \@nameuse br@#2\@extra@b@citeb \@ifundefined b@#2\@extra@b@citeb @num @parse #2 @tmp #1 NAT@b@open@#2 NAT@b@shut@#2 \@ifnum @merge>\@ne @bibitem@firs...

  51. [59]

    [o??] x o7L

    @open @close @open @close and [1] URL: #1 \@ifundefined chapter * \@mkboth \@ifxundefined @sectionbib * \@mkboth * \@mkboth\@gobbletwo \@ifclassloaded amsart * \@ifclassloaded amsbook * \@ifxundefined @heading @heading NAT@ctr thebibliography [1] @ \@biblabel @NAT@ctr \@bibset...

    1996

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.