pith. sign in

arxiv: 2605.01870 · v1 · submitted 2026-05-03 · 💻 cs.CL

Maistros: A Greek Large Language Model Adapted Through Knowledge Distillation From Large Reasoning Models

Nikolaos Giarelis , Charalampos Mastrokostas , Nikos Karacapilidis This is my paper

Pith reviewed 2026-05-08 19:02 UTC · model grok-4.3

classification 💻 cs.CL
keywords Greek LLMknowledge distillationlarge reasoning modelsCulturaQAquestion answeringmultilingual NLPopen-weights model
0
0 comments X

The pith

An 8B Greek LLM distilled from large reasoning models outperforms other open models on Greek QA benchmarks.

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

The paper demonstrates that a compact 8-billion-parameter model for Modern Greek can be built by distilling reasoning capabilities from much larger models and then fine-tuning on a purpose-built dataset. The authors introduce CulturaQA, a set of Greek question-answer pairs first generated by large reasoning models and then refined by human curators. Using this data, they produce Maistros 8B and show it leads in accuracy across multiple Greek question-answering tests while also releasing a lightweight evaluation framework. Readers should care because the approach offers a practical route to capable models for languages that lack large native datasets, without requiring the compute budget of training from scratch.

Core claim

Maistros 8B is a state-of-the-art open-weights Greek LLM obtained by knowledge distillation from large reasoning models followed by fine-tuning on CulturaQA, a high-quality LRM-generated and human-curated Greek QA dataset. Evaluation across nine human-curated Greek QA datasets shows Maistros 8B surpassing nine other LLMs, including both general and Greek-specific models.

What carries the argument

Knowledge distillation from large reasoning models into an 8B base model, using the CulturaQA dataset of LRM-generated and human-curated Greek question-answer pairs for fine-tuning.

If this is right

  • Maistros 8B sets a new reference performance level for open Greek LLMs on question answering.
  • CulturaQA provides a reusable training and evaluation resource for future Greek language models.
  • The memory-efficient evaluation framework can be reused for other languages and QA tasks.
  • Targeted distillation allows smaller models to acquire reasoning strengths for specific languages without full-scale pretraining.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same dataset-generation and distillation pipeline could be applied to other low-resource languages by swapping the target language in the LRM prompts.
  • Prioritizing human curation after LRM generation may prove more effective than simply scaling data volume for multilingual adaptation.
  • Future experiments could test whether adding chain-of-thought supervision during distillation further boosts performance on multi-step Greek reasoning tasks.

Load-bearing premise

The CulturaQA dataset, created by large reasoning models and then curated by humans, supplies data of sufficient quality and cultural representativeness for distillation to yield better Greek QA performance than existing models.

What would settle it

If Maistros 8B scores below other open-weights Greek or multilingual models on the nine held-out Greek QA datasets, or if an ablated version trained only on the uncured LRM-generated portion matches its results, the central claim would be refuted.

Figures

Figures reproduced from arXiv: 2605.01870 by Charalampos Mastrokostas, Nikolaos Giarelis, Nikos Karacapilidis.

Figure 1
Figure 1. Figure 1: CulturaQA’s sample distribution per category. The y-axis measures the number of samples; the percentage of samples is written in each bar. To create CulturaQA, we manually curated a list of 180 Greek keyphrases (for the exact phrases, see the code repository) that were grouped into eleven categories. These categories are “πολιτισμός” (“civilization”), “ταξίδια” (“travelling”), "πολιτική" (“politics”), “οικ… view at source ↗
Figure 2
Figure 2. Figure 2: The overall approach for training Maistros 8B. generating it again. We experimented with various training hyperparameters (see view at source ↗
Figure 3
Figure 3. Figure 3: Training and Evaluation loss over steps are trained on step 375 (epoch 3) where the lowest validation and training losses are achieved. This is visualized in view at source ↗
read the original abstract

Large Language Models (LLMs) have substantially advanced the field of Natural Language Processing (NLP), achieving state-of-the-art performance across a wide range of tasks. These improvements have been attributed, in part, to their emerging reasoning capabilities, which are enabled by large-scale training and increased model capacity. However, existing LLMs can generate erroneous responses when addressing complex queries that fall outside their training distribution, due to limited internal knowledge or the need for multi-step reasoning. To address these limitations, recent work has introduced large reasoning models (LRMs), which incorporate explicit internal reasoning processes to improve response accuracy. Additionally, state-of-the-art LRMs often comprise hundreds of billions of parameters and require several seconds per inference, even on advanced multi-GPU systems. These characteristics limit their practicality for deployment in conventional computing environments. Meanwhile, NLP research on multilingual LLMs continues to prioritize high-resource languages. However, these models exhibit limited performance in under-resourced languages, primarily due to insufficient language- and culture-specific training data. In this paper, we focus on Modern Greek, for which only a limited number of question answering (QA) datasets have been proposed, most of which are intended for model evaluation. To address this research gap in Greek QA, we make the following contributions: (i) CulturaQA, a high-quality LRM-generated and human-curated dataset, for Greek LLM training and evaluation; (ii) a memory-efficient LLM evaluation framework adaptable to diverse languages and QA tasks; (iii) Maistros 8B, a state-of-the-art open-weights Greek LLM developed via knowledge distillation and fine-tuning on CulturaQA; and (iv) a comprehensive evaluation of nine LLMs across nine human-curated Greek QA datasets.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper introduces CulturaQA, a new high-quality Greek QA dataset generated using large reasoning models (LRMs) and refined through human curation; a memory-efficient framework for LLM evaluation on QA tasks; and Maistros 8B, an 8B-parameter open-weights Greek LLM obtained via knowledge distillation from LRMs followed by fine-tuning on CulturaQA. It claims Maistros 8B achieves state-of-the-art results on Greek QA and reports a broad evaluation of nine LLMs across nine human-curated Greek QA datasets.

Significance. If the performance claims and dataset quality are substantiated with quantitative evidence, the work would provide a useful open-weights model and training resource for Modern Greek, helping close the gap for under-resourced languages. The distillation pipeline and evaluation framework could serve as a template for similar adaptations in other languages. The significance is currently limited by the absence of supporting metrics.

major comments (2)
  1. [Abstract] Abstract: the central claim that Maistros 8B is state-of-the-art is stated without any quantitative metrics, baseline comparisons, or error analysis, which is load-bearing for the primary contribution and cannot be assessed from the given description.
  2. [Abstract] Contributions (i): CulturaQA is described as high-quality LRM-generated and human-curated data sufficient to enable superior distillation performance, yet no validation statistics (e.g., inter-annotator agreement, LRM hallucination rates before/after curation, lexical diversity, or comparison to existing Greek QA corpora) are referenced, leaving open the possibility that any gains are attributable to unverified data quality rather than the method.
minor comments (1)
  1. [Abstract] The abstract lists evaluation of nine LLMs on nine datasets but does not name them; adding this information would improve clarity even if details appear later.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed feedback. We address each major comment below and describe the revisions we will implement to improve the clarity and substantiation of our claims.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that Maistros 8B is state-of-the-art is stated without any quantitative metrics, baseline comparisons, or error analysis, which is load-bearing for the primary contribution and cannot be assessed from the given description.

    Authors: We agree that the abstract, in its current concise form, does not include specific quantitative metrics or direct references to baseline comparisons and error analysis. The full manuscript contains these details in the evaluation section, reporting results across nine Greek QA datasets with comparisons to nine other LLMs. We will revise the abstract to incorporate key performance figures (e.g., accuracy improvements on CulturaQA and other benchmarks) and a brief mention of the comparative evaluation to make the state-of-the-art claim immediately verifiable from the abstract alone. revision: yes

  2. Referee: [Abstract] Contributions (i): CulturaQA is described as high-quality LRM-generated and human-curated data sufficient to enable superior distillation performance, yet no validation statistics (e.g., inter-annotator agreement, LRM hallucination rates before/after curation, lexical diversity, or comparison to existing Greek QA corpora) are referenced, leaving open the possibility that any gains are attributable to unverified data quality rather than the method.

    Authors: The abstract summarizes the contribution at a high level. The full paper provides the requested validation statistics in the CulturaQA construction section, including inter-annotator agreement, pre/post-curation hallucination rates from the LRMs, lexical diversity metrics, and direct comparisons against prior Greek QA corpora. To address the concern directly in the abstract, we will add a brief clause referencing these supporting statistics so that the data-quality claims are substantiated without requiring the reader to consult the body text. revision: yes

Circularity Check

0 steps flagged

No circularity: standard empirical pipeline with no self-referential derivations

full rationale

The paper describes a conventional empirical workflow: LRM-generated and human-curated CulturaQA dataset creation, followed by knowledge distillation and fine-tuning to produce Maistros 8B, then evaluation of nine LLMs on nine Greek QA datasets. No equations, parameters, or derivations appear in the abstract or described contributions that reduce performance claims to fitted inputs, self-definitions, or self-citation chains. No uniqueness theorems, ansatzes, or renamings of known results are invoked. The central claim rests on dataset construction and benchmarking rather than any load-bearing loop back to the paper's own inputs by construction. Absence of quantitative data-quality metrics is a potential evidence gap but does not constitute circularity under the defined patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the effectiveness of knowledge distillation for Greek and the quality of the newly created CulturaQA dataset; no explicit free parameters or invented physical entities are stated in the abstract.

axioms (1)
  • domain assumption Knowledge distillation from large reasoning models transfers useful reasoning capabilities to smaller models for a low-resource language.
    Invoked implicitly when claiming Maistros reaches SOTA via distillation on CulturaQA.
invented entities (1)
  • CulturaQA dataset no independent evidence
    purpose: High-quality training and evaluation resource for Greek QA
    Newly generated and curated in this work; no independent evidence outside the paper is provided.

pith-pipeline@v0.9.0 · 5630 in / 1177 out tokens · 53987 ms · 2026-05-08T19:02:04.420930+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

63 extracted references · 63 canonical work pages · 4 internal anchors

  1. [1]

    Naveed, H.et al.A comprehensive overview of large language models.ACM Trans. Intell. Syst. Technol.16, DOI: 10.1145/3744746 (2025). 3.Bommasani, R.et al.On the opportunities and risks of foundation models (2022). 2108.07258

    work page doi:10.1145/3744746 2025

  2. [2]

    Llama 2: Open Foundation and Fine-Tuned Chat Models

    Brown, T. B.et al.Language models are few-shot learners. InProceedings of the 34th International Conference on Neural Information Processing Systems, NIPS ’20 (Curran Associates Inc., Red Hook, NY , USA, 2020). 5.Touvron, H.et al.Llama 2: Open foundation and fine-tuned chat models (2023). 2307.09288. 6.Singh, A.et al.Openai gpt-5 system card (2025). 2601....

    work page internal anchor Pith review arXiv 2020

  3. [3]

    Xu, F.et al.Toward large reasoning models: A survey of reinforced reasoning with large language models.Patterns6, 101370, DOI: https://doi.org/10.1016/j.patter.2025.101370 (2025)

    work page doi:10.1016/j.patter.2025.101370 2025

  4. [4]

    The Roots of Performance Disparity in Multilingual Language Models: Intrinsic Modeling Difficulty or Design Choices?

    Shani, C., Reif, Y ., Roll, N., Jurafsky, D. & Shutova, E. The roots of performance disparity in multilingual language models: Intrinsic modeling difficulty or design choices? (2026). 2601.07220

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  5. [5]

    M¨ uller, L´ aszl´ o N´ emeth, Luis Oala, Lennart Purucker, Sahithya Ravi, 10 Jan N

    Qin, L.et al.A survey of multilingual large language models.Patterns6, 101118, DOI: https://doi.org/10.1016/j.patter. 2024.101118 (2025)

    work page doi:10.1016/j.patter 2024

  6. [6]

    & Pavlopoulos, J

    Bakagianni, J., Pouli, K., Gavriilidou, M. & Pavlopoulos, J. A systematic survey of natural language processing for the greek language.Patterns6, 101313, DOI: https://doi.org/10.1016/j.patter.2025.101313 (2025). 12.Papantoniou, K. & Tzitzikas, Y . Nlp for the greek language: A longer survey (2024). 2408.10962

    work page doi:10.1016/j.patter.2025.101313 2025

  7. [7]

    & Karacapilidis, N

    Giarelis, N., Mastrokostas, C., Siachos, I. & Karacapilidis, N. A review of greek nlp technologies for chatbot development. InProceedings of the 27th Pan-Hellenic Conference on Progress in Computing and Informatics, PCI ’23, 15–20, DOI: 10.1145/3635059.3635062 (Association for Computing Machinery, New York, NY , USA, 2024)

    work page doi:10.1145/3635059.3635062 2024

  8. [8]

    In Christodoulopoulos, C., Chakraborty, T., Rose, C

    Roussis, D.et al.Krikri: Advancing open large language models for Greek. In Christodoulopoulos, C., Chakraborty, T., Rose, C. & Peng, V . (eds.)Findings of the Association for Computational Linguistics: EMNLP 2025, 5012–5033, DOI: 10.18653/v1/2025.findings-emnlp.268 (Association for Computational Linguistics, Suzhou, China, 2025)

    work page doi:10.18653/v1/2025.findings-emnlp.268 2025

  9. [9]

    2602.05150

    Zhang, Y .et al.Greekmmlu: A native-sourced multitask benchmark for evaluating language models in greek (2026). 2602.05150

    work page arXiv 2026

  10. [10]

    & Karacapilidis, N

    Mastrokostas, C., Giarelis, N. & Karacapilidis, N. Evaluating monolingual and multilingual large language models for greek question answering: The demosqa benchmark (2026). 2602.16811. 17.Liu, A. H.et al.Ministral 3 (2026). 2601.08584

    work page arXiv 2026

  11. [11]

    Zhang, T., Kishore, V ., Wu, F., Weinberger, K. Q. & Artzi, Y . Bertscore: Evaluating text generation with bert. In International Conference on Learning Representations(2020). 19.Grattafiori, A.et al.The llama 3 herd of models (2024). 2407.21783. 20.Kamath, A.et al.Gemma 3 technical report (2025). 2503.19786. 21.Yang, A.et al.Qwen3 technical report (2025)...

    work page internal anchor Pith review arXiv 2020

  12. [12]

    In Christodoulopoulos, C., Chakraborty, T., Rose, C

    Peng, X.et al.Plutus: Benchmarking large language models in low-resource Greek finance. In Christodoulopoulos, C., Chakraborty, T., Rose, C. & Peng, V . (eds.)Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, 30176–30202, DOI: 10.18653/v1/2025.emnlp-main.1535 (Association for Computational Linguistics, Suzhou, China, ...

    work page doi:10.18653/v1/2025.emnlp-main.1535 2025

  13. [13]

    2401.16182

    Gesnouin, J.et al.Llamandement: Large language models for summarization of french legislative proposals (2024). 2401.16182

    work page arXiv 2024

  14. [14]

    & Limkonchotiwat, P

    Ong, D. & Limkonchotiwat, P. SEA-LION (Southeast Asian languages in one network): A family of Southeast Asian language models. In Tan, L., Milajevs, D., Chauhan, G., Gwinnup, J. & Rippeth, E. (eds.)Proceedings of the 3rd Workshop for Natural Language Processing Open Source Software (NLP-OSS 2023), 245–245, DOI: 10.18653/v1/2023.nlposs-1.26 (Association fo...

    work page doi:10.18653/v1/2023.nlposs-1.26 2023

  15. [15]

    & Semeraro, G

    Polignano, M., Basile, P. & Semeraro, G. Advanced natural-based interaction for the ITAlian language: LLaMAntino-3- ANITA.Sci. Reports16, 5375, DOI: 10.1038/s41598-025-31319-0 (2026). 27.Vreš, D.et al.Building a strong instruction language model for a less-resourced language (2026). 2603.01691

    work page doi:10.1038/s41598-025-31319-0 2026

  16. [16]

    Translationese in Machine Translation Evaluation

    Graham, Y ., Haddow, B. & Koehn, P. Translationese in Machine Translation Evaluation, DOI: 10.48550/arXiv.1906.09833 (2019). ArXiv:1906.09833. 29.V oukoutis, L.et al.Meltemi: The first open large language model for greek (2024). 2407.20743

    work page Pith review doi:10.48550/arxiv.1906.09833 1906

  17. [17]

    InThe Thirteenth International Conference on Learning Representations(2025)

    Romanou, A.et al.INCLUDE: Evaluating multilingual language understanding with regional knowledge. InThe Thirteenth International Conference on Learning Representations(2025)

    work page 2025

  18. [18]

    & Prokopidis, P

    Kyriazi, P. & Prokopidis, P. Multiple choice qa greek asep.Hugging Face. https://huggingface.co/datasets/ilsp/mcqa_ greek_asep/ (2025)

    work page 2025

  19. [19]

    & Prokopidis, P

    Vacalopoulou, A., Sofianopoulos, S. & Prokopidis, P. Greek physical commonsense reasoning dataset.Hugging Face. https://huggingface.co/datasets/ilsp/greek_pcr/ (2025)

    work page 2025

  20. [20]

    Chang, Catherine Arnett, Abdelrahman Eldesokey, Abdelrahman Boda Sadallah, Abeer Kashar, Aitazaz Daud, Abosede Grace Olanihun, et al

    Chang, T. A.et al.Global piqa: Evaluating physical commonsense reasoning across 100+ languages and cultures (2025). 2510.24081

    work page internal anchor Pith review arXiv 2025

  21. [21]

    J.et al.LoRA: Low-rank adaptation of large language models

    Hu, E. J.et al.LoRA: Low-rank adaptation of large language models. InInternational Conference on Learning Representations(2022)

    work page 2022

  22. [22]

    Mao, Y .et al.A survey on LoRA of large language models.Front. Comput. Sci.19, 197605, DOI: 10.1007/ s11704-024-40663-9 (2024)

    work page 2024

  23. [23]

    Biderman, D.et al.LoRA learns less and forgets less.Transactions on Mach. Learn. Res.(2024). Featured Certification

    work page 2024

  24. [24]

    S., Andreas, J., Torralba, A

    Shuttleworth, R. S., Andreas, J., Torralba, A. & Sharma, P. LoRA vs full fine-tuning: An illusion of equivalence. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems(2025). 38.Chowdhery, A.et al.Palm: Scaling language modeling with pathways.J. Mach. Learn. Res.24, 1–113 (2023)

    work page 2025

  25. [25]

    Transformers: State-of-the-Art Natural Language Processing

    Wolf, T.et al.Transformers: State-of-the-art natural language processing. In Liu, Q. & Schlangen, D. (eds.)Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, 38–45, DOI: 10.18653/v1/2020.emnlp-demos.6 (Association for Computational Linguistics, Online, 2020)

    work page doi:10.18653/v1/2020.emnlp-demos.6 2020

  26. [26]

    The effect of sampling temperature on problem solving in large language models

    Renze, M. The effect of sampling temperature on problem solving in large language models. In Al-Onaizan, Y ., Bansal, M. & Chen, Y . N. (eds.)Findings of the Association for Computational Linguistics: EMNLP 2024, 7346–7356, DOI: 10.18653/v1/2024.findings-emnlp.432 (Association for Computational Linguistics, Miami, Florida, USA, 2024)

    work page doi:10.18653/v1/2024.findings-emnlp.432 2024

  27. [27]

    & Zettlemoyer, L

    Dettmers, T., Pagnoni, A., Holtzman, A. & Zettlemoyer, L. Qlora: efficient finetuning of quantized llms. InProceedings of the 37th International Conference on Neural Information Processing Systems, NIPS ’23 (Curran Associates Inc., Red Hook, NY , USA, 2023)

    work page 2023

  28. [28]

    L., Kindermans, P

    Smith, S. L., Kindermans, P. J. & Le, Q. V . Don’t decay the learning rate, increase the batch size. InInternational Conference on Learning Representations(2018)

    work page 2018

  29. [29]

    Dettmers, M

    Zhang, J., He, T., Sra, S. & Jadbabaie, A. Why gradient clipping accelerates training: A theoretical justification for adaptivity. InInternational Conference on Learning Representations(2020). 44.Dettmers, T., Lewis, M., Shleifer, S. & Zettlemoyer, L. 8-bit optimizers via block-wise quantization (2022). 2110.02861

    work page arXiv 2020

  30. [30]

    The Hitchhiker ' s Guide to Testing Statistical Significance in Natural Language Processing

    Dror, R., Baumer, G., Shlomov, S. & Reichart, R. The hitchhiker’s guide to testing statistical significance in natural language processing. In Gurevych, I. & Miyao, Y . (eds.)Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 1383–1392, DOI: 10.18653/v1/P18-1128 (Association for Computational L...

    work page doi:10.18653/v1/p18-1128 2018

  31. [31]

    & Kritsis, K

    Kaddas, P., Gatos, B., Palaiologos, K., Christopoulou, K. & Kritsis, K. Text line detection and recognition of greek polytonic documents. In Coustaty, M. & Fornés, A. (eds.)Document Analysis and Recognition – ICDAR 2023 Workshops, 213–225 (Springer Nature Switzerland, Cham, 2023). 12/15

    work page 2023

  32. [32]

    Chatzikyriakidis, S., Papadakis, D., Papaioannou, S. I. & Psaltaki, E. Grdd+: An extended greek dialectal dataset with cross-architecture fine-tuning evaluation (2026). 2511.03772

    work page arXiv 2026

  33. [33]

    Chlapanis, Dimitris Galanis, Nikolaos Aletras, and Ion Androut- sopoulos

    Chlapanis, O. S., Galanis, D., Aletras, N. & Androutsopoulos, I. GreekBarBench: A challenging benchmark for free-text legal reasoning and citations. In Christodoulopoulos, C., Chakraborty, T., Rose, C. & Peng, V . (eds.)Findings of the Association for Computational Linguistics: EMNLP 2025, 25099–25119, DOI: 10.18653/v1/2025.findings-emnlp.1368 (Associatio...

    work page doi:10.18653/v1/2025.findings-emnlp.1368 2025

  34. [34]

    & Tzovaras, D

    Tsourma, M., Michail, D., Varlamis, I., Drosou, A. & Tzovaras, D. Legal assistance in low-resource languages: Evaluating rag and fine-tuned llms for greek e-governance. In2025 3rd International Conference on Foundation and Large Language Models (FLLM), 366–373, DOI: 10.1109/FLLM67465.2025.11391043 (2025)

    work page doi:10.1109/fllm67465.2025.11391043 2025

  35. [35]

    & Tsanakas, P

    Koniaris, M., Galanis, D., Giannini, E. & Tsanakas, P. Evaluation of automatic legal text summarization techniques for greek case law.Information14, DOI: 10.3390/info14040250 (2023)

    work page doi:10.3390/info14040250 2023

  36. [36]

    & Karacapilidis, N

    Giarelis, N., Mastrokostas, C. & Karacapilidis, N. Greekt5: Sequence-to-sequence models for greek news summarization. In Maglogiannis, I., Iliadis, L., Macintyre, J., Avlonitis, M. & Papaleonidas, A. (eds.)Artificial Intelligence Applications and Innovations, 60–73 (Springer Nature Switzerland, Cham, 2024)

    work page 2024

  37. [37]

    & Karacapilidis, N

    Giarelis, N., Mastrokostas, C. & Karacapilidis, N. Greek wikipedia: A study on abstractive summarization. InProceedings of the 13th Hellenic Conference on Artificial Intelligence, SETN ’24, DOI: 10.1145/3688671.3688769 (Association for Computing Machinery, New York, NY , USA, 2024)

    work page doi:10.1145/3688671.3688769 2024

  38. [38]

    Loukas, L.et al.GR-NLP-TOOLKIT: An open-source NLP toolkit for Modern Greek. In Rambow, O.et al.(eds.) Proceedings of the 31st International Conference on Computational Linguistics: System Demonstrations, 174–182 (Association for Computational Linguistics, Abu Dhabi, UAE, 2025)

    work page 2025

  39. [39]

    M., Kyritsis, K., Perikos, I

    Liapis, C. M., Kyritsis, K., Perikos, I. & Paraskevas, M. Transformer-based embeddings for greek language categorization. In2024 IEEE/ACIS 24th International Conference on Computer and Information Science (ICIS), 176–181, DOI: 10.1109/ ICIS61260.2024.10778332 (2024)

    work page arXiv 2024

  40. [40]

    & Kompatsiaris, I

    Stylianou, N., Tsikrika, T., Vrochidis, S. & Kompatsiaris, I. Cross-domain hate speech detection for content moderation in greek social networks. In2024 IEEE/WIC International Conference on Web Intelligence and Intelligent Agent Technology (WI-IAT), 373–379, DOI: 10.1109/WI-IAT62293.2024.00059 (2024)

    work page doi:10.1109/wi-iat62293.2024.00059 2024

  41. [41]

    & Karacapilidis, N

    Mastrokostas, C., Giarelis, N. & Karacapilidis, N. Social media topic classification on greek reddit.Information15, DOI: 10.3390/info15090521 (2024). Acknowledgements This work has received funding from the European Union’s Horizon Europe research and innovation programme under grant agreement No. 101235708 (BLUEPRINT – Building Living Urban Ecosystems th...

    work page doi:10.3390/info15090521 2024

  42. [43]

    3.Απόφυγε τη χρήση στερεοτύπων

    Λάβε υπόψη τον ελληνικό πολιτισμό και την ελληνική κοινωνική πραγματικότητα όπου είναι σχετικό. 3.Απόφυγε τη χρήση στερεοτύπων

    work page

  43. [44]

    Βάλε πάντα το σύμβολο • πριν από κάθε ερώτηση

    work page

  44. [45]

    Δημιούργησε σημαντικές, συχνά προκύπ- τουσες και χρήσιμες ερωτήσεις για το θέμα

    work page

  45. [46]

    ΄Ολες οι ερωτήσεις πρέπει να μπορούν να απαντηθούν αντικειμενικά

    work page

  46. [47]

    Μη δημιουργείς επαναλαμβανόμενες ερωτήσεις

    work page

  47. [48]

    Κάθε ερώτηση πρέπει να είναι σαφώς ορισμένη

    work page

  48. [49]

    Παρακαλώ δημιούργησε 15 ερώτησεις για το εξής θέμα:{topic} You are an extremely developed Artificial Intelligence model for the Greek Language

    Γράψε μόνο το κείμενο των ερωτήσεων, χωρίς επιπλέον σχόλια. Παρακαλώ δημιούργησε 15 ερώτησεις για το εξής θέμα:{topic} You are an extremely developed Artificial Intelligence model for the Greek Language. Use the following instructions to create a series of questions on the topic mentioned by the user: Instructions:

    work page

  49. [52]

    Avoid the use of stereotypes

    work page

  50. [53]

    Always place the symbol • before each question

    work page

  51. [54]

    Create significant, frequently occurring and useful questions for the topic

    work page

  52. [55]

    All questions must be able to be answered objectively

    work page

  53. [56]

    Do not create repeated questions

    work page

  54. [57]

    Every question must be clearly defined

    work page

  55. [58]

    Write only the text of the questions, without extra comments. Please create 15 questions for the follow- ing topic: {topic} Continued on next page 14/15 Table A1 – continued from previous page Prompt Type Greek English (Translated) Dataset Creation (Answers) Είσαι ένα εξαιρετικά ανεπτυγμένο μοντέλο Τεχνητής Νοημοσύνης για την Ελληνική γλώσσα. Χρησιμοποίησ...

    work page

  56. [59]

    Απάντα αποκλειστικά στα Ελληνικά με άψογη γραμματική, σύνταξη και ορθογραφία

    work page

  57. [60]

    Λάβε υπόψη τον ελληνικό πολιτισμό και την ελληνική κοινωνική πραγματικότητα όπου είναι σχετικό

    work page

  58. [61]

    4.Απόφυγε τη χρήση στερεότυπων

    Απάντησε στις ερωτήσεις του χρήστη με ειλικρίνεια και επιστημονική ακρίβεια. 4.Απόφυγε τη χρήση στερεότυπων

    work page

  59. [62]

    χώρα, περίοδος): –Μην ζητήσεις διευκρίνιση

    Αν η ερώτηση είναι ασαφής ή λείπει πληροφορία (π.χ. χώρα, περίοδος): –Μην ζητήσεις διευκρίνιση. – Δώσε την απάντηση κάνοντας ρητές παραδοχές (π.χ. ¨Ελλείψει άλλης αναφοράς, θεωρούμε ως προεπιλογή την Ελλάδα και το τρέχον έτος¨). Παρακαλώ απάντησε στη παρακάτω ερώτηση: {question} You are an extremely developed Artificial Intelligence model for the Greek La...

    work page

  60. [63]

    Answer exclusively in Greek with impeccable grammar, syntax and spelling

    work page

  61. [64]

    Take into consideration the Greek civilization and the Greek social reality where relevant

    work page

  62. [65]

    Answer the user question with honesty and scientific accuracy

    work page

  63. [66]

    In the absence of other reference, we assume as a default Greece and the current year

    If the question is vague or information is missing (e.g., country, time period): – Do not ask for clarification. – Give the answer by making explicit assump- tions (e.g. "In the absence of other reference, we assume as a default Greece and the current year”). Please answer the following question: {ques- tion} Multiple Choice (Evaluation) Διάβασε προσεκτικ...

    work page