Pith. sign in

REVIEW 4 major objections 6 minor 14 references

Hermit Kingdom Through the Lens of Multiple Perspectives: A Case Study of LLM Hallucination on North Korea

T0 review · 4 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read The paper argues that which LLM you ask, and in which language, materially changes the factual picture users get of North Korea.

desk verdict Worth a serious referee: the paper documents real model- and language-linked variation in LLM answers about North Korea, but the language comparison is partly confounded by unequal prompt assertiveness. read the letter →

arxiv 2501.05981 v1 pith:KJXYBRFI submitted 2025-01-10 cs.CL

Eunjung Cho , Won Ik Cho , Soomin Seo This is my paper · ORCID
classification cs.CL
keywords LLMhallucinationNorthKoreamultilingualevaluationfalserumoursgeopoliticalmisinformationrefusaltoanswersycophancyAIsovereignty
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 takes North Korea as an extreme test case for LLM hallucination: a country with scarce verifiable information and many uncorrected false rumours. It asks whether five widely used chatbots — ChatGPT-3.5, Gemini, Claude 3 Sonnet, Qwen-72B, and Solar Mini — produce accurate, consistent, and non-refusing answers about North Korea in English, Korean, and Mandarin Chinese. Across 13 topics with verifiable ground truth, the paper measures large differences in accuracy, refusal-to-answer, and consistency depending on the model and the language. The central claim is that the choice of model and language can yield materially different factual pictures of North Korea, which matters because users increasingly rely on LLMs for information about a geopolitically sensitive country. If true, hallucination-mitigation strategies that simply add credible sources or encourage abstention are not enough for closed-information environments.

What carries the argument

The measurement apparatus is a small curated dataset of 13 North Korea topics with verifiable ground truth, built by prompting GPT-4 for topic candidates, keeping only topics that appeared more than once, and filtering by a domain expert. The topics are split into false rumours (widely circulated but debunked claims, e.g. Kim Jong Un's death or a unicorn lair) and lesser-known information (true but under-documented facts, e.g. the five-year election cycle or the Arduous March). To provoke hallucination, the prompts deliberately use discourse markers that pressure the model: 'Apparently' for false rumours and 'I heard' for lesser-known information, exploiting sycophancy. Each model is prompted five times per topic, and three metrics are computed: accuracy against ground truth, refusal-to-answer rate, and answer-type consistency. This design lets the paper attribute answer differences to model and language while holding topic content fixed.

What would settle it

If the prompts were back-translated to equalize assertiveness across languages (e.g. using the same 'Apparently' hedge in Korean) and the Korean/English accuracy differences disappeared, the central claim that language itself shapes understanding would be falsified. Alternatively, a larger sample of real user queries about North Korea that showed no model-by-language interaction would undercut the generalization.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that 'the choice of model and language can lead to vastly different understandings of North Korea.' For 13 topics with verifiable truths, accuracy ranged widely: Claude 3 Sonnet was generally most accurate, ChatGPT-3.5 next, Gemini lowest, while language-specialised models (Qwen-72B for Chinese, Solar Mini for Korean) were highly accurate for verified topics but not for false rumours. Gemini refused to answer frequently across all languages even on non-sensitive topics, whereas Qwen almost never refused. The paper also reports that models were less accurate on false rumours than on lesser-known true information, and that prominent death rumours about Kim Jong Un, Hyon Song Wol, and Jang Song Thaek persisted in model outputs despite available corrections. The authors attribute part of the language gap to English-centric training data, and they interpret the high accuracy of language-specific models as consistent with the case for 'AI sovereignty.'

Load-bearing premise

The findings rest on the assumption that the 13 expert-selected topics and the translated prompts represent how users actually ask about North Korea, and that the translated prompts are equivalent in how strongly they assert the rumour.

Editorial extensions

If this is right

  • Users of different LLMs, or the same LLM in different languages, can receive contradictory factual information about North Korea on the same topic.
  • Refusal-to-answer behaviour does not reliably signal lack of knowledge: Gemini's high refusal rate inflated its consistency score while lowering its accuracy, so users may see a conservative model as a consistent one.
  • False rumours that were widely reported but weakly corrected remain a persistent hallucination source even for prominent cases with available corrections.
  • Language-specific models can outperform general multilingual models in their own language on verified facts, lending support to 'AI sovereignty' arguments.
  • Hallucination mitigation via retrieval augmentation or uncertainty expression may fail in closed-information domains, where expert-curated sources and careful source design are needed.

Reading between the lines

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

  • Not in the paper: the 'language effect' is partly confounded with prompt force. The Korean false-rumour prompt asserts the rumour is a clear fact, while English uses 'Apparently'; a controlled back-translation study could isolate whether language or prompt assertiveness drives the accuracy gap.
  • Not in the paper: because all 13 topics were generated by GPT-4 and filtered by a single expert, the dataset is a convenience sample, not a representation of real user queries; a larger query log from search or chat data would test whether the patterns generalise.
  • Not in the paper: the same method could be applied to other closed or partially closed information environments (e.g. Myanmar, Eritrea, Turkmenistan) where rumour persistence and source scarcity may produce similar model-by-language divergence.
  • Not in the paper: consistency as measured (1/number of answer types) rewards repeated refusal, so a model that always declines can score as highly consistent; future metrics should separate engagement from uniformity.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. This paper presents a multilingual, multi-model evaluation of how large language models handle North Korea-related information. The authors construct 13 topics with verifiable ground truth, split into false rumours and lesser-known information, and prompt ChatGPT-3.5, Gemini, Claude 3 Sonnet, Qwen-72B-Chat, and Solar Mini in English, Mandarin Chinese, and Korean with five repetitions per condition. They report accuracy, refusal-to-answer rate, and consistency in Table 1, supplemented by qualitative examples in the appendix. The central claim, advanced in the abstract and conclusion, is that the choice of model and language can lead to materially different factual understandings of North Korea, with implications for misinformation and geopolitical information access.

Significance. The paper addresses a genuinely important and understudied problem: LLM hallucination in an information-scarce, geopolitically sensitive domain where common mitigation strategies (grounding, abstention) may be counterproductive. The design has notable strengths: the ground truth for the 13 topics is externally verifiable, the topic-generation step using GPT-4 is not circular because the evaluated models are different systems, the inclusion of language-specific open models (Qwen for Chinese, Solar Mini for Korean) provides a useful contrast to the multilingual closed models, and the paper is transparent about its annotation workflow. If the observed differences are reliable, the study would be a valuable early demonstration that users of different LLMs in different languages may receive different factual pictures of North Korea. The contribution is largely empirical and descriptive rather than methodological, but the domain relevance is high.

major comments (4)
  1. [§4.1.2 and Appendix A.1] The cross-linguistic prompts for the false-rumour category are not semantically equivalent. The English prompt hedges with "Apparently", while the Korean prompt asserts "{description}는 것은 명확한 사실이야" ("it is a clear fact that ...") and the Chinese prompt uses "显然是" ("obviously is"), both of which express a stronger epistemic commitment. Because the paper's RQ2 conclusion attributes accuracy differences to language (e.g., ChatGPT-3.5 Korean FR accuracy 44.62 vs. English FR accuracy 56.92 in Table 1), prompt assertiveness is a direct confound for the language effect. The authors should either re-run the false-rumour prompts with matched epistemic framing across languages, or explicitly re-analyse the data excluding false rumours; otherwise the language comparison for this category is not clean.
  2. [§5.2 and Table 1] The abstract and conclusion assert "significant differences" and "vastly different understandings," but the paper reports no significance tests, confidence intervals, or effect sizes. With only five regenerations per model-by-language-by-topic cell and 13 topics, the observed differences may be within sampling variability; for example, Gemini's Korean FR RtA ratio (38.46) is nearly equal to its English FR RtA ratio (40.00). The authors should add a permutation or bootstrap analysis over trials and topics, or explicitly weaken the wording to "observed differences in this exploratory set." The Limitations section's assertion that dataset size "becomes less critical" once differences are apparent needs statistical backing rather than an argument from belief.
  3. [§5 (annotation paragraph) and §4.3.1] The accuracy metric requires subjective judgment about the balance of evidence in model outputs, yet the paper reports no inter-annotator agreement. The text says there was "minimal ambiguity" but provides no kappa or other agreement measure, and the annotation process relied on at most three annotators with different language expertise. Because accuracy is the primary quantitative comparison in Table 1, an agreement statistic is needed to support the reliability of the labels. Without it, the reader cannot distinguish measurement error from genuine model differences.
  4. [§4.2 and Limitations] The paper does not release the raw model outputs, annotations, or the exact model versions used (e.g., which GPT-3.5 snapshot, which Gemini version, which Claude 3 Sonnet release), and the access window is only described as March 2024. Given that the central claim is empirical and that the authors themselves note the topics are a small, expert-filtered set without a sampling frame, these omissions limit reproducibility and make it difficult to assess whether the findings generalize beyond the 13 topics. At minimum, exact model identifiers and the full prompt/output/annotation files should be provided in a supplement or repository.
minor comments (6)
  1. [Table 1] Each metric cell appears to contain three numbers (e.g., 56.92, 34.29, 83.33 for ChatGPT-3.5 English accuracy), but the caption only defines (FR) and (LKI). Please clarify what the third number represents.
  2. [Table 1] The Gemini Korean consistency cell contains "0.50 (I)"; this should be "0.50 (LKI)" to match the other entries.
  3. [§4.1.1] The phrase "artibrary generation" is a typo for "arbitrary generation."
  4. [§4.1.2] The Chinese false-rumour prompt contains duplicated text "你对你对自己的答案" and the English prompt has a doubled period after "your answer.."; these should be corrected.
  5. [Appendix A.1 caption] The caption claims "The semantics for the input text is the same for all languages," which conflicts with the differing epistemic strength documented in major comment 1; the claim should be revised or the prompts aligned.
  6. [§5.2] The statement that "both Qwen-72B and Solar Mini achieved 100% accuracy for verified topics" is not visible in Table 1; either include subcategory results in the table or refer explicitly to the supplementary material.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the benchmark compares independent LLMs against externally verifiable ground truth, and no fitted parameter or self-citation chain forces the reported results.

full rationale

The paper's central claim is empirical: different LLMs and languages produce measurably different responses about North Korea, with accuracy, refusal, and consistency scored against externally verifiable ground truth. There is no derivation chain in which an output is constructed from its own inputs. The topic candidates are generated by GPT-4, but GPT-4 is explicitly not among the evaluated models, and the candidate topics are only prompts for other models, not labels or fitted values that determine those models' answers. The ground-truth labels (true/false/uncertain) are set by a domain expert from verifiable facts, not derived from the models under test. The paper does cite its own authors' prior journalism research (Seo 2018, 2023; Seo et al. 2023), but these citations support background claims about sensationalist North Korea coverage and are not the basis of the measured accuracy differences; they are independent published studies about media, not about LLM behavior. No uniqueness theorem, ansatz, or fitted-input-as-prediction pattern appears. The notable methodological weakness is the prompt-asymmetry confound in Section 4.1.2, where the Korean false-rumour prompt asserts the claim as 'a clear fact' while the English prompt hedges with 'Apparently'; this is a validity threat to the language comparison, but it is not circularity — the reported numbers are still genuine outputs of the models, not quantities forced by construction. Therefore the circularity score is 0.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

No fitted parameters or invented entities; the paper is an empirical benchmark. The main hidden assumptions concern the correctness of ground-truth labels, the cross-lingual equivalence of prompts, and the reliability of annotation without reported agreement.

assumptions (4)
  • domain assumption The 13 topics have verifiable, stable ground-truth labels (true/false) independent of the models.
    The evaluation and accuracy scores depend entirely on these labels; the paper asserts verifiability but provides no formal source list for each ground truth. Section 4.1.1.
  • ad hoc to paper The discourse-marker prompts are semantically equivalent across languages and induce the intended effect of treating false rumours as true and lesser-known facts as uncertain.
    The Korean false-rumour prompt uses a phrase meaning 'clear fact' while English uses 'Apparently'; if not equivalent, language comparisons are confounded. Section 4.1.2.
  • domain assumption A model answer can be faithfully classified into one of true, false, uncertain, or refusal-to-answer by bilingual annotators plus one expert.
    The paper states there was minimal ambiguity, but reports no inter-annotator agreement metric. Section 5.
  • domain assumption Responses from public API access in March 2024 are representative of each model's behavior on these topics.
    Model versions, sampling temperature, and platform routing are not fixed, and outputs are non-deterministic. Section 4.2.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Hermit Kingdom Through the Lens of Multiple Perspectives: A Case Study of LLM Hallucination on North Korea." pith.science (2026). https://pith.science/paper/KJXYBRFI

@misc{pith2026250105981,
  author       = {Pith},
  title        = {Pith review of: Hermit Kingdom Through the Lens of Multiple Perspectives: A Case Study of LLM Hallucination on North Korea},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KJXYBRFI}},
  note         = {Machine review of arXiv:2501.05981}
}
read the original abstract

Hallucination in large language models (LLMs) remains a significant challenge for their safe deployment, particularly due to its potential to spread misinformation. Most existing solutions address this challenge by focusing on aligning the models with credible sources or by improving how models communicate their confidence (or lack thereof) in their outputs. While these measures may be effective in most contexts, they may fall short in scenarios requiring more nuanced approaches, especially in situations where access to accurate data is limited or determining credible sources is challenging. In this study, we take North Korea - a country characterised by an extreme lack of reliable sources and the prevalence of sensationalist falsehoods - as a case study. We explore and evaluate how some of the best-performing multilingual LLMs and specific language-based models generate information about North Korea in three languages spoken in countries with significant geo-political interests: English (United States, United Kingdom), Korean (South Korea), and Mandarin Chinese (China). Our findings reveal significant differences, suggesting that the choice of model and language can lead to vastly different understandings of North Korea, which has important implications given the global security challenges the country poses.

Figures

Figures reproduced from arXiv: 2501.05981 by the authors.

Figure 1
Figure 1. Workflow Diagram. We first instruct GPT-4 to generate five topic candidates for each of two categories: false rumours (FR) and lesser-known information (LKI). The model is prompted in three languages - English (EN), Chinese (ZH), and Korean (KO) – with ten iterations per language. The generated candidates are then aggregated, de-duplicated, and filtered to remove topics mentioned only once. This refined set is furth… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

14 extracted references · 11 canonical work pages

  1. [1]

    조선민주주의인민공화국은근로인민대중에게 참다운 민주주의적권리와자유를향유할수있는 참다운정치적생활조건을지어주는 인민민주주의적국가이다

    북한 헌법제8조: "조선민주주의인민공화국은근로인민대중에게 참다운 민주주의적권리와자유를향유할수있는 참다운정치적생활조건을지어주는 인민민주주의적국가이다."

    work page

  2. [2]

    3)외신및 탈북자증언: 탈북자들은대의원선거에서 1명의후보자만 추대 되며주민들에게 투표강요가있었다고증언합니다

    북한 매체보도: 북한 매체들은정기적으로 최고인민회의대의원선거개 최와 99%이상의 투표율등을보도합니다. 3)외신및 탈북자증언: 탈북자들은대의원선거에서 1명의후보자만 추대 되며주민들에게 투표강요가있었다고증언합니다. 제가이정보에대해확신하는정도는높습니다. 북한 체제의폐쇄성으로인 해정확한실상을완벽히파악하기는어렵지만,공식 문건과 다양한증언을 토대로 봤을때위와같은 상황으로판단됩니다. - - 북한의선거제도에대해설명하겠습니다.이 답변에대해서는높은수준의 확신을가지고있습니다. 북한은 ’최고인민회의대의원선거’를 5년마다실시하고있습니다.이선거 는 단일후보형식으로진행되며,유권자들은 해당후보자에대해 찬성혹은 반대 투표를 하게됩니다. 그러나이선거는실질적으로형식...

    work page

  3. [3]

    arXiv preprint arXiv:2401.11817

    Hallucination is inevitable: An innate lim- itation of large language models. arXiv preprint arXiv:2401.11817. Yuqing Yang, Ethan Chern, Xipeng Qiu, Graham Neu- big, and Pengfei Liu. 2023. Alignment for honesty. arXiv preprint arXiv:2312.07000. Jia-Yu Yao, Kun-Peng Ning, Zhen-Hui Liu, Mu-Nan Ning, and Li Yuan. 2023. Llm lies: Hallucinations are not bugs, ...

    arXiv 2023

  4. [6]

    2)반대 투표를 한유권자에대한 불이익이존재하므로,대부분의유권자가 기권하거나 찬성 투표를 합니다

    북한노동당이후보자를 단독으로지명하기때문에유권자의선택권이없 습니다. 2)반대 투표를 한유권자에대한 불이익이존재하므로,대부분의유권자가 기권하거나 찬성 투표를 합니다

    work page

  5. [7]

    따라서 북한의이 른바 ’선거’는권력세습을위 한절차일 뿐, 민주적선거와는 거리가 멉니다.이는 북한정권의주요 특징인일 당독재 체제를반영합니다

    투표결과가 미리결정되어있기때문에실제득표율은 무의미합니다. 따라서 북한의이 른바 ’선거’는권력세습을위 한절차일 뿐, 민주적선거와는 거리가 멉니다.이는 북한정권의주요 특징인일 당독재 체제를반영합니다. 이 답변의근거로는 북한 헌법,국제 NGO및싱크탱크의 북한인권보고서, 북한 체제전 문가들의견해등을 참고했습니다. Continued on next page Table 7 – continued from previous page Model Prompt Output - - 북한에서 최고인민회의대의원선거를 5년마다진행하는것은 사실입니다. 이는 북한 헌법에 명시되어있습니다. 다만이선거가실제로 민주적인방식 으로이루어지는지에대해서는 많은의 문이제기되고있습니다. 근거:

    work page

  6. [8]

    조선민주주의인민공화국은 착취와압박에서영구히 해 방된근로인민대중의권력을실시한다.근로인민대중은 최고인민회의와지 방각급인민회의를 통하여주권을 행사한다

    북한 헌법제8조: "조선민주주의인민공화국은 착취와압박에서영구히 해 방된근로인민대중의권력을실시한다.근로인민대중은 최고인민회의와지 방각급인민회의를 통하여주권을 행사한다."

    work page

  7. [9]

    최고인민회의대의원선거는 5년에 한번씩 한다

    북한 헌법제109조: "최고인민회의대의원선거는 5년에 한번씩 한다." 그러나 북한의선거절차와결과에대해서는 다음과같은 문제점들이지적 되고있습니다: 1.후보자지명과정의 비민주성 -노동당이선별한후보자만 출마가능 2. 실질적인경쟁선거 부재 -대부분지역구에서 단일후보자만 출마 3. 투표 과정의공정성 문제 -강압적 분위기, 무기명 투표위반등 따라서 북한선거가형식적인절차에 불과하다는지적이 많습니다.전반적 으로이주제에대해서는 70%정도의확신을가지고있습니다. - - 네, 북한에서는 5년마다 최고인민회의대의원선거를실시하고있습니다.이 는 북한 헌법에 명시되어있는 사항입니다. 다만이선거가실제로 민주적 절차를거치는지에대해서는의문의여지가있습니다. 근거:

    work page

  8. [10]

    조선민주주의인민공화국은주체사상을구현한인민대 중중심의정치철학이다.근로인민대중은국가주권의주체이며국가권력의 유일한주인이다

    북한 헌법제8조: "조선민주주의인민공화국은주체사상을구현한인민대 중중심의정치철학이다.근로인민대중은국가주권의주체이며국가권력의 유일한주인이다."

    work page

Show all 14 references
  1. [11]

    최고인민회의대의원선거는 5년에 한번보 통,평등, 직접, 비밀투표원칙에의하여진행된다

    북한 헌법제109조: "최고인민회의대의원선거는 5년에 한번보 통,평등, 직접, 비밀투표원칙에의하여진행된다."

  2. [12]

    4.국제사회에서도 북한의선거가형식적인절차에 불과하다고평가하고있 습니다

    하지만 북한에서는 단일후보만 상정되고, 투표율 100%에가까운득표율 을기록하고있어선거의실질적의미가없다는지적이있습니다. 4.국제사회에서도 북한의선거가형식적인절차에 불과하다고평가하고있 습니다. 제가가진정보로는 북한에서 5년마다대의원선거를 치르고있지만,실제로 는 1당지배체제 하에서 민주적선거절차가이루어지지않는것으로보입니 다.이정도수준의정보에대해서는높은확신을가지고있습니다. - - 북한에서 5년마다 최고인민회의대의원선거를실시하는것은 사실입니다. 하지만이선거는실제로자유롭고공정한선거라기보다는 체제유지와정...

  3. [13]

    -유권자들은 단지 찬반 투표만 할 뿐 다른선택권이없습니다

    단일후보경선제 (확신도:높음) - 최고인민회의대의원선거에서는노동 당이 사전에선정한 단일후보자만 투표할수있습니다. -유권자들은 단지 찬반 투표만 할 뿐 다른선택권이없습니다

  4. [14]

    - 하지만이는전체주의 체제하 에서 투표거부가 사실상 불가능하기때문입니다

    99%이상높은 투표율과지지율 (확신도:높음) - 북한 당국은 매번 99% 이상의높은 투표율과지지율을발표합니다. - 하지만이는전체주의 체제하 에서 투표거부가 사실상 불가능하기때문입니다. 3.외부감시단없는 불투명한절차 (확신도:높음) -선거과정에대한객관 적인국제사회감시가이루어지지않습니다. - 투개표과정의공정성과 투명 성에의문이있습니다. 따라서 북한의 5년마다이루어지는선거는실질적인 민주주의선거라기보 다는 체제유지와정당성 부여의수단으로활용되는의례적 행사라고볼수 있습니다.이에대한제확신수준은전반적으로높...

  5. [2023]

    In Proceedings of Mensch und Computer 2023, pages 81–90

    From chatgpt to factgpt: A participatory de- sign study to mitigate the effects of large language model hallucinations on users. In Proceedings of Mensch und Computer 2023, pages 81–90. Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goy...

    2023 arXiv

  6. [2024]

    arXiv preprint arXiv:2405.20362

    Hallucination-free? assessing the reliability of leading ai legal research tools. arXiv preprint arXiv:2405.20362. Cheng Niu, Yuanhao Wu, Juno Zhu, Siliang Xu, KaShun Shum, Randy Zhong, Juntong Song, and Tong Zhang. 2024. RAGTruth: A hallucination cor- pus for developing trust...

    2024 arXiv

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.