Pith. sign in

REVIEW 5 major objections 5 minor 41 references

A Lightweight Multi-Expert Generative Language Model System for Engineering Information and Knowledge Extraction

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

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

Pith's one-line read A graph of small fine-tuned experts, each trained on one isolated subsection of an aircraft repair manual and routed by an orchestrator, reports 3x the exact-match accuracy of a fine-tuned 8B model while training 1.7x faster on a single…

desk verdict A plausible lightweight multi-expert system whose headline 3x EM gain is undercut by a misreported ratio and an unspecified train/test split. read the letter →

arxiv 2505.21109 v1 pith:FCDQ5XUN submitted 2025-05-27 cs.CL cs.AIcs.CEcs.IRcs.LG

Bogdan Bogachov , Yaoyao Fiona Zhao This is my paper
classification cs.CLcs.AIcs.CEcs.IRcs.LG
keywords SmallLanguageGraphmulti-expertmodelLoRAfine-tuningengineeringdocumentquestionansweringhallucinationmitigationknowledgeovershadowinginformationextractionstructuralrepairmanual
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 proposes the Small Language Graph (SLG), an architecture for adapting language models to engineering documents without large compute. Instead of fine-tuning one big model on an entire manual, SLG splits the manual into subsections and fine-tunes a separate 1B-parameter expert (Llama-3.2-1B-Instruct, with LoRA) on each chunk, then uses an orchestrator to route each user query to the correct expert. On a Cessna Structural Repair Manual, the authors report that SLG reaches an Exact Match score of 0.12 versus 0.05 for a fine-tuned Llama-3.1-8B model, a threefold improvement, while total fine-tuning time is 1.7x shorter (3,475 seconds versus 5,891 seconds) and the whole system runs on one RTX 4090. The authors argue that isolating knowledge by subsection prevents 'knowledge overshadowing,' the blending of overlapping training contexts that blurs meanings, and thereby reduces hallucination in exact engineering answers. If the result holds, small and medium engineering firms could deploy accurate, locally run question-answering over their own documents without expensive cloud or cluster computing.

What carries the argument

The central mechanism is the graph of fine-tuned experts, each a Llama-3.2-1B-Instruct model adapted with LoRA (low-rank addition of trainable adapter layers) and trained only on question-answer pairs synthesized from one subsection of the aircraft repair manual, so no expert sees overlapping knowledge from another subsection. An orchestrator node, trained to output expert names as answers, routes each incoming query to the most relevant expert. The design rationale is that eliminating data overlap removes 'knowledge overshadowing,' the blending of similar contexts that the paper identifies as a hallucination source: when two procedures share an opening but diverge later, a single model trained on both blends them, whereas an isolated expert cannot. A graph library wires the nodes and orchestrator together, and the orchestrator's measured routing accuracy is about 70 percent, meaning the reported scores are a lower bound on what perfect routing would give.

What would settle it

Run SLG and the fine-tuned Llama-3.1-8B on a test set of questions written and verified by human engineers rather than synthesized by an LLM, forcing the orchestrator to route every query to the correct expert; if SLG's exact-match advantage over the 8B model does not remain near threefold, or reverses, the reported gain is an artifact of the synthetic test set or of asymmetric routing failures rather than of knowledge isolation.

Watch

Extended reading notes

Core claim

The central claim is that a collection of small, narrowly fine-tuned language models connected as a graph can outperform a single larger fine-tuned model on exact engineering information retrieval. Each expert is a Llama-3.2-1B-Instruct model fine-tuned with LoRA on question-answer pairs generated from a single subsection of the source manual; the orchestrator, also a LoRA-tuned Llama-3.2-1B model, maps a user query to an expert name. On the best hyperparameter configuration, SLG scores Exact Match 0.12, ROUGE-L 0.41, METEOR 0.50, while the standalone fine-tuned Llama-3.1-8B scores Exact Match 0.05, ROUGE-L 0.46, METEOR 0.55, and the standalone Llama-3.2-1B scores Exact Match 0.04, ROUGE-L 0.43, METEOR 0.51. The reported threefold Exact Match advantage is the evidence that the multi-expert graph resists hallucination by keeping each expert's knowledge isolated; the loose-overlap metrics remain comparable because they reward similar wording rather than exact reproduction.

Load-bearing premise

The entire comparison assumes that the question-answer pairs generated automatically by Llama-3.3-70B-Instruct are correct ground truth, that every question maps to exactly one subsection, and that subsection boundaries really isolate knowledge; if the synthetic pairs are noisy, the questions are misplaced, or the overlapping contexts still leak, the threefold exact-match gap may reflect test-set artifacts or routing errors rather than genuine hallucination resistance.

Editorial extensions

If this is right

  • If the reported metrics hold, a graph of 1B-parameter experts can replace a single fine-tuned 8B model for exact-answer engineering question answering while training 1.7 times faster and running on one consumer GPU.
  • The threefold Exact Match advantage with comparable ROUGE-L and METEOR indicates the benefit is specific to exact, hallucination-free reproduction of procedures, not to general fluency or wording overlap.
  • Because the orchestrator routes correctly only about 70 percent of the time, the paper's numbers are conservative; improving the router would raise all three metrics.
  • The same subsection-chunk and per-chunk expert recipe should transfer to other structured engineering documentation, since the authors note that such manuals are usually divided into logical sections.
  • Total SLG fine-tuning takes 3,475 seconds on a single RTX 4090, making the approach feasible for small companies that cannot afford high-end servers or cloud compute.

Reading between the lines

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

  • The synthetic test set is the paper's explicit soft spot: because both training and test QA pairs come from Llama-3.3-70B-Instruct without manual verification, a human-annotated test set could reorder the models; the paper itself concedes that human evaluation would be a more exhaustive hallucination check.
  • Forcing correct routing with an oracle experiment would isolate expert quality from routing quality; if SLG's Exact Match jumps well above 0.12 under oracle routing, the bottleneck is the orchestrator rather than knowledge isolation.
  • The knowledge-overshadowing mechanism yields a direct test: deliberately merge two overlapping subsections into one expert's training data and watch Exact Match fall while ROUGE-L stays roughly constant, which would confirm the mechanism rather than just the architecture.
  • The authors' distributed-inference vision is plausible only after the orchestrator improves; at 70 percent routing accuracy, a peer-to-peer fleet of device-level experts would amplify routing mistakes across the network.
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

5 major / 5 minor

Summary. SLG is a graph-based system in which an orchestrator model, built on Llama-3.2-1B-Instruct and fine-tuned with LoRA, routes a user query to one of several expert models, each fine-tuned on an isolated subsection chunk of an aircraft structural repair manual. Training questions and answers are synthesized by Llama-3.3-70B-Instruct from the chunks. The paper reports that SLG achieves Exact Match 0.12 versus 0.05 for a fine-tuned Llama-3.1-8B-Instruct and 0.04 for a size-matched standalone Llama-3.2-1B-Instruct, that it fine-tunes about 1.7 times faster than the 8B model on one RTX 4090, and that the knowledge-isolation design helps resist hallucinations. The evaluation uses ROUGE-L, EM, and METEOR, with 13 hyperparameter-tuning runs summarized in Table 3 and Figure 4. The paper also reports an orchestrator routing success rate of approximately 70% and acknowledges limitations including the absence of human evaluation and the lack of conversational memory.

Significance. If the empirical claims were fully supported, SLG would be a practically valuable demonstration that a collection of small LoRA-tuned experts can outperform a larger standalone model on exact engineering answer retrieval at substantially lower compute cost, with potential distributed-deployment implications. The authors make a concrete, falsifiable quantitative claim, point to public code for the fine-tuning pipeline, and ground the work in a real maintenance manual, which are strengths. However, the current evaluation has unresolved data-leakage and selection-bias risks, the headline 3x figure is not internally consistent with Table 6, and the hallucination-resistance claim is supported only indirectly. The significance is therefore conditional on a substantially stronger experimental protocol.

major comments (5)
  1. [Section 3.1, Section 4.1] The manuscript never defines a train/test split or a held-out set of subsections. Section 3.1 states that synthesized QA pairs are used for model fine-tuning and testing, and Section 4.1 refers to a test dataset described in Subsection 3.1, but no partition is specified. Since the same QA pairs feed the expert models and the orchestrator, the reported EM values may reflect memorization of training questions rather than generalization; this is the principal load-bearing issue for the claimed 3x EM advantage. The authors must document the split, ensure test questions and subsections are disjoint from training, and preferably add manual verification of the synthesized ground truth.
  2. [Section 4.2, Table 3, Table 6] The hyperparameters are selected from 13 runs by choosing the values exhibiting the best performance, and Table 6 then reports these best results without error bars, repeated seeds, or a separate validation set. With no holdout, part of the observed EM gap can be attributed to selection bias, especially because the tuning grid is small and only EM drives the reported advantage. Please report validation-based selection, multiple runs, and variance or confidence intervals.
  3. [Section 4.3, Section 6, abstract] The headline claim of a 3x EM improvement is not supported by the paper's own Table 6: SLG EM 0.12 versus Llama-3.1-8B EM 0.05 is a 2.4x ratio, not 3x; the 3x ratio holds only against the 1B baseline (0.12/0.04). The abstract and Section 6 state '3 times better' without specifying the baseline, while Section 4.3 makes the 3x claim immediately after presenting Table 6. The text and abstract must be corrected to match the data or must explicitly state the baseline.
  4. [Section 4.3, Section 5] EM is not by itself a hallucination metric, and Section 5 explicitly concedes that human evaluation and fact-checking could be a more exhaustive way to estimate hallucination avoidance. Since the paper's motivation is reducing hallucinations, the conclusion that SLG has the potential to better resist hallucinations needs either direct hallucination evaluation, for example adversarial or out-of-distribution queries, or a substantially softened claim.
  5. [Section 4.3] The orchestrator has a reported routing success rate of only about 70%, but Table 6 reports only end-to-end metrics. Because the orchestrator is part of the system, the 0.12 EM could be depressed by routing errors; conversely, if test questions overlap training data, routing could be artificially easy. Reporting oracle-routed EM, per-expert accuracy, and the number of experts would clarify whether the advantage comes from expert specialization rather than from routing artifacts.
minor comments (5)
  1. [Table 3] Rows 7 and 10 have identical hyperparameter values (learning rate 1e-3, LoRA rank 16, gradient accumulation 2, LoRA alpha 8); one of them appears to be a duplicate and should be removed or replaced.
  2. [Section 3.1] The dataset description omits the number of subsections, the number of experts, and the number of synthesized QA pairs per expert, which are needed to assess the system's scale and data isolation.
  3. [Section 4.2, Table 4] Table 4 refers LoRA alpha, LoRA r, learning rate, and gradient accumulation back to Table 3; the final best configuration should be stated explicitly so that the reported results are reproducible.
  4. [Section 4.1] The cross-reference to a test dataset described in Subsection 3.1 is inaccurate because Subsection 3.1 does not define any test split; please add a dedicated dataset section.
  5. [Section 4.1] The metric descriptions should mention the specific implementations or packages used for ROUGE-L and METEOR, since score values can differ across implementations.

Circularity Check

2 steps flagged · score 6.0 of 10

SLG's headline EM comparison is evaluated on the same QA pairs used for fine-tuning, with no documented train/test split; the reported 3x gain reduces to a fit on the training data.

  1. fitted input called prediction [Section 3.1 (Dataset) and Section 4.1 (Experimentation setup)]
    "Subsequently, each chunk is fed into Llama-3.3-70B-Instruct LLM [34], asking it to generate questions for the text. Thus, question-answer pairs are created, which are used for model fine-tuning and testing. ... All models are tested using a test dataset described in Subsection 3.1 by comparing generated answers to ground truth answers."

    The test dataset is defined by the same subsection-chunk QA pipeline used for fine-tuning, and no train/test split, held-out subsections, or manual QA verification is documented. The experts and orchestrator are trained on these QA pairs and then evaluated against the same QA pairs, so the reported EM values (SLG 0.12 vs 0.05/0.04) measure fit to the training distribution rather than generalization. The claimed 3x EM advantage is therefore forced by the evaluation design, not by the SLG architecture.

  2. fitted input called prediction [Section 4.2 (Fine-tuning strategy), Tables 3 and 6]
    "The values in bold indicate which hyperparameter is tuned at each specific row. After tuning, the values exhibiting the best performance are fixed and highlighted in green. ... Table 6 showcases the best experiment results."

    Thirteen hyperparameter configurations were evaluated on the same QA dataset and the best-performing run was then reported as the system's final 'best experiment results.' Because the evaluation set is not held out from model selection, the reported EM 0.12 is the maximum of the tuning grid rather than an unbiased prediction; part of the claimed gain over the baselines is a selection artifact.

full rationale

The paper contains no mathematical derivation to audit; its central claim is an empirical comparison. The main circularity is in the evaluation protocol: Section 3.1 states that synthesized QA pairs are 'used for model fine-tuning and testing,' and Section 4.1 says the test dataset is the one 'described in Subsection 3.1.' No train/test split or held-out subsections are reported, so the same QA content appears to feed both training and evaluation. Under this design, the reported EM values (SLG 0.12 vs Llama-3.1-8B 0.05 and Llama-3.2-1B 0.04) measure the models' fit to the training distribution rather than generalization, making the headline '3 times better' claim an artifact of the evaluation design. A second, related issue is that the 13 hyperparameter configurations in Table 3 were evaluated on the same data and the best was then reported as the final result in Table 6; without a separate holdout for model selection, part of the gain is the maximum of 13 trials, not an unbiased estimate. The paper is otherwise self-contained: it benchmarks against two external base models with a fixed LoRA pipeline, and there is no load-bearing self-citation chain (the only self-citation, [37], is a code link). The arithmetic claim '3 times' is also inconsistent with Table 6 (0.12/0.05 = 2.4x), but that is a correctness issue, not circularity. Overall, because the central quantitative claim appears to reduce to evaluation on training data, I assign a circularity score of 6.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

No invented entities beyond the SLG architecture itself; the numerical free parameters are the tuned hyperparameters, and the central claims rest on assumptions about the quality of LLM-generated QA pairs, subsection-based chunking, single-expert routing, and EM as a hallucination proxy.

free parameters (5)
  • learning_rate = 1e-3 (best of 1e-5, 1e-4, 1e-3)
    Tuned in experiment rows 1-3; best value selected by evaluation metric and reported in Table 6.
  • LoRA rank = 16 (best of 4, 8, 16, 32)
    Tuned in experiment rows 3-6; best value selected by evaluation metric.
  • LoRA alpha = 64 (best of 8, 16, 32, 64)
    Tuned in experiment rows 10-13; best value selected by evaluation metric.
  • gradient_accumulation_steps = 2 (best of 2, 4, 8)
    Tuned in experiment rows 7-9; best value selected by evaluation metric.
  • number of chunks/experts = not reported
    The document structure determines expert count, but the paper never states how many chunks or experts were used, making the routing and training-cost claims hard to interpret.
assumptions (4)
  • domain assumption Engineering manuals have well-defined subsection boundaries that yield logically isolated chunks.
    Section 3.1: the entire data-isolation strategy assumes subsections are the right unit and that chunking by subsections prevents harmful overlap.
  • domain assumption Llama-3.3-70B-generated question-answer pairs are correct enough to serve as ground truth for fine-tuning and evaluation.
    Section 3.1: QA pairs are created by prompting an LLM, with no manual verification or quality filter reported.
  • ad hoc to paper Exact Match is the most powerful indicator of hallucination resistance.
    Section 4.3: the authors assert EM is the most powerful indication, but provide no external evidence that EM tracks factual correctness better than ROUGE-L or METEOR, which favor the baseline.
  • domain assumption Each user query can be answered by exactly one expert.
    Sections 3.2 and 5: the orchestrator is trained to return a single expert name; multi-part queries are not supported, and the 70% routing success rate shows this assumption is frequently violated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Lightweight Multi-Expert Generative Language Model System for Engineering Information and Knowledge Extraction." pith.science (2026). https://pith.science/paper/FCDQ5XUN

@misc{pith2026250521109,
  author       = {Pith},
  title        = {Pith review of: A Lightweight Multi-Expert Generative Language Model System for Engineering Information and Knowledge Extraction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FCDQ5XUN}},
  note         = {Machine review of arXiv:2505.21109}
}
read the original abstract

Despite recent advancements in domain adaptation techniques for large language models, these methods remain computationally intensive, and the resulting models can still exhibit hallucination issues. Most existing adaptation methods do not prioritize reducing the computational resources required for fine-tuning and inference of language models. Hallucination issues have gradually decreased with each new model release. However, they remain prevalent in engineering contexts, where generating well-structured text with minimal errors and inconsistencies is critical. This work introduces a novel approach called the Small Language Graph (SLG), which is a lightweight adaptation solution designed to address the two key challenges outlined above. The system is structured in the form of a graph, where each node represents a lightweight expert - a small language model fine-tuned on specific and concise texts. The results of this study have shown that SLG was able to surpass conventional fine-tuning methods on the Exact Match metric by 3 times. Additionally, the fine-tuning process was 1.7 times faster compared to that of a larger stand-alone language model. These findings introduce a potential for small to medium-sized engineering companies to confidently use generative AI technologies, such as LLMs, without the necessity to invest in expensive computational resources. Also, the graph architecture and the small size of expert nodes offer a possible opportunity for distributed AI systems, thus potentially diverting the global need for expensive centralized compute clusters.

Figures

Figures reproduced from arXiv: 2505.21109 by the authors.

Figure 1
Figure 1. FIGURE 1: DATA OVERLAPPING ILLUSTRATION. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. FIGURE 2: SCHEMATIC REPRESENTATION OF ISOLATED TRAIN [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. FIGURE 3: SMALL LANGUAGE GRAPH [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: FIGURE 4: EXPERIMENT CHARTS [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

41 extracted references · 33 canonical work pages

  1. [1]

    “OpenAI.” https://openai.com

    OpenAI. “OpenAI.” https://openai.com. Accessed: Febru- ary 13, 2025

    work page 2025

  2. [2]

    “Gemini.” https://gemini.google.com/app

    Google. “Gemini.” https://gemini.google.com/app. Ac- cessed: February 13, 2025

    work page 2025

  3. [3]

    Llama: Openandefficientfoundationlanguagemod- els. CoRR, abs/2302.13971, 2023. doi: 10.48550

    Touvron,Hugo,Lavril,Thibaut,Izacard,Gautier,Martinet, Xavier, Lachaux, Marie-Anne, Lacroix, Timothée, Roz- ière, Baptiste, Goyal, Naman, Hambro, Eric, Azhar, Faisal etal. “Llama: Openandefficientfoundationlanguagemod- els. CoRR, abs/2302.13971, 2023. doi: 10.48550.”arXiv preprint arXiv.2302.13971(2023)

    arXiv 2023

  4. [4]

    From Language Models to Practical Self-Improving Computer Agents

    Sheng, Alex. “From Language Models to Practical Self- Improving Computer Agents.” arXiv (2024)URL 2404. 11964, URL https://arxiv.org/abs/2404.11964

    work page Pith review arXiv 2024

  5. [5]

    TheRiseandPotentialofLargeLanguage Model Based Agents: A Survey

    Xi, Zhiheng, Chen, Wenxiang, Guo, Xin, He, Wei, Ding, Yiwenandetel. “TheRiseandPotentialofLargeLanguage Model Based Agents: A Survey.”arXiv (2023)URL 2309. 07864, URL https://arxiv.org/abs/2309.07864

    arXiv 2023

  6. [6]

    ReAct — Synergizing Reasoning and Acting in Language Models

    Yao, Shunyu, Zhao, Jeffrey, Yu, Dian, Du, Nan, Shafran, Izhak, Narasimhan, Karthik and Cao, Yuan. “ReAct — Synergizing Reasoning and Acting in Language Models.” Conferencepaper.2023.InternationalConferenceonLearn- ingRepresentations,ICLR. URLhttps://www.scopus.com/ inward/record.uri?eid=2-s2.0-85199863002&partnerID= 40&md5=7cd72e9c58ecd294bec0f951cfda3ef8...

    work page 2023

  7. [7]

    Data-driven innovation: What is it?

    Luo, Jianxi. “Data-driven innovation: What is it?”IEEE Transactions on Engineering ManagementVol. 70 No. 2 (2022): pp. 784–790

    work page 2022

  8. [8]

    A data-driven approach for creative concept gener- ation and evaluation

    Han, Ji, Forbes, Hannah, Shi, Feng, Hao, Jia and Schaefer, Dirk. “A data-driven approach for creative concept gener- ation and evaluation.”Proceedings of the Design Society: DESIGN Conference, Vol. 1: pp. 167–176. 2020. Cam- bridge University Press

    work page 2020

Show all 41 references
  1. [9]

    EW-Tune: A FrameworkforPrivatelyFine-TuningLargeLanguageMod- els with Differential Privacy

    Behnia, Rouzbeh, Ebrahimi, Mohammadreza Reza, Pacheco, Jason and Padmanabhan, Balaji. “EW-Tune: A FrameworkforPrivatelyFine-TuningLargeLanguageMod- els with Differential Privacy.” 2022 IEEE International ConferenceonDataMiningWorkshops(ICDMW) :pp.560–

    2022

  2. [10]

    Exploring the limits of transfer learning withaunifiedtext-to-texttransformer

    Raffel,Colin,Shazeer,Noam,Roberts,Adam,Lee,Kather- ine,Narang,Sharan,Matena,Michael,Zhou,Yanqi,Li,Wei and Liu, Peter J. “Exploring the limits of transfer learning withaunifiedtext-to-texttransformer.” Journalofmachine learning researchVol. 21 No. 140 (2020): pp. 1–67

    2020

  3. [11]

    Languagemodelsare unsupervisedmultitasklearners

    Radford, Alec, Wu, Jeffrey, Child, Rewon, Luan, David, Amodei,Dario,Sutskever,Ilyaetal. “Languagemodelsare unsupervisedmultitasklearners.” OpenAIblog Vol.1No.8 (2019): p. 9

    2019

  4. [12]

    AttentionisAllyouNeed

    Vaswani, Ashish, Shazeer, Noam, Parmar, Niki, Uszkoreit, Jakob, Jones, Llion, Gomez, Aidan N, Kaiser, Ł ukasz and Polosukhin,Illia. “AttentionisAllyouNeed.” Proceedings ofthe31stInternationalConferenceonNeuralInformation Processing Systems. 2017

    2017

  5. [13]

    meta-llama/Llama-3.2-1B-Instruct

    Meta. “meta-llama/Llama-3.2-1B-Instruct.” https:// huggingface.co/meta-llama/Llama-3.2-1B-Instruct. Ac- cessed: February 26, 2025

    2025

  6. [14]

    The Llama 3 Herd of Models

    Grattafiori, Aaron, Dubey, Abhimanyu, Jauhri, Abhinav, Pandey, Abhinav, Kadian, Abhishek and et al. “The Llama 3 Herd of Models.”arXiv (2024)URL 2407.21783, URL https://arxiv.org/abs/2407.21783

    2024 arXiv

  7. [15]

    Causal Parrots: Large Lan- guage Models May Talk Causality But Are Not Causal

    Zečević, Matej, Willig, Moritz, Dhami, Devendra Singh and Kersting, Kristian. “Causal Parrots: Large Lan- guage Models May Talk Causality But Are Not Causal.” 8 Transactions on Machine Learning Research(2023)URL https://openreview.net/forum?id=tv46tCzs83

    2023

  8. [16]

    Knowl- edge Overshadowing Causes Amalgamated Hallucination in Large Language Models

    Zhang, Yuji, Li, Sha, Liu, Jiateng, Yu, Pengfei, Fung, Yi R., Li, Jing, Li, Manling and Ji, Heng. “Knowl- edge Overshadowing Causes Amalgamated Hallucination in Large Language Models.” (2024). URL 2407.08039, URL https://arxiv.org/abs/2407.08039

    2024 arXiv

  9. [17]

    Conceptual Design Generation Using Large Language Mod- els

    Ma, Kevin, Grandi, Daniele, McComb, Christo- pher and Goucher-Lambert, Kosa. “Conceptual Design Generation Using Large Language Mod- els.” Vol. 6. 2023. URL https://www.scopus.com/ inward/record.uri?eid=2-s2.0-85178515903&doi= 10.1115%2fdetc2023-116838&partnerID=40&md5= 217f1f...

    2023

  10. [18]

    Augmenting human innovation teams with arti- ficial intelligence: Exploring transformer-based language models

    Bouschery, Sebastian G., Blazevic, Vera and Piller, Frank T. “Augmenting human innovation teams with arti- ficial intelligence: Exploring transformer-based language models.” Journal of Product Innovation Management Vol. 40 No. 2 (2023): p. 139 – 153. URL https://www. scopus.co...

    2023

  11. [19]

    Artificial intelligence prompt engineering as a new digital competence: Anal- ysis of generative AI technologies such as ChatGPT

    Korzynski, Pawel, Mazurek, Grzegorz, Krzypkowska, Pamela and Kurasinski, Artur. “Artificial intelligence prompt engineering as a new digital competence: Anal- ysis of generative AI technologies such as ChatGPT.” Entrepreneurial Business and Economics ReviewVol. 11 No. 3 (2023)...

    2023

  12. [20]

    A Multi- task, Multilingual, Multimodal Evaluation of ChatGPT on Reasoning, Hallucination, and Interactivity

    Bang, Yejin, Lee, Nayeon, Dai, Wenliang, Su, Dan, Wilie, Bryan, Lovenia, Holy, Ji, Ziwei, Yu, Tiehzheng, Chung, Willy, Do, Quyet, Yan, Xu and Fung, Pascale. “A Multi- task, Multilingual, Multimodal Evaluation of ChatGPT on Reasoning, Hallucination, and Interactivity.” (2023). ...

    2023 doi arXiv

  13. [21]

    Finetuned Language Models Are Zero-Shot Learners

    Wei, Jason, Bosma, Maarten, Zhao, Vincent Y., Guu, Kelvin, Yu, Adams Wei, Lester, Brian, Du, Nan, Dai, Andrew M. and Le, Quoc V. “Finetuned Language Models Are Zero-Shot Learners.” ICLR 2022 - 10th International Conference on Learning Representations (2022)URL https://www.scop...

    2022

  14. [22]

    LaMDA: Language Models for Dialog Applications

    Thoppilan,Romal,Freitas,DanielDe,Hall,Jamie,Shazeer, Noam M., Kulshreshtha, Apoorv and et al. “LaMDA: Language Models for Dialog Applications.” ArXiv Vol. abs/2201.08239 (2022). URL https://api.semanticscholar. org/CorpusID:246063428

    2022 arXiv

  15. [23]

    Efficient Hierarchical Domain Adaptation for Pretrained Language Models

    Chronopoulou, Alexandra, Peters, Matthew E. and Dodge, Jesse. “Efficient Hierarchical Domain Adaptation for Pretrained Language Models.” NAACL 2022 - 2022 Conference of the North American Chapter of the Asso- ciation for Computational Linguistics: Human Language Technologies, ...

    2022

  16. [24]

    LoRA: Low-Rank Adaptation of Large Language Models

    Hu, Edward, Shen, Yelong, Wallis, Phillip, Allen- Zhu, Zeyuan, Li, Yuanzhi, Wang, Shean, Wang, Lu and Chen, Weizhu. “LoRA: Low-Rank Adaptation of Large Language Models.” ICLR 2022 - 10th In- ternational Conference on Learning Representations (2022)URL https://www.scopus.com/in...

    2022

  17. [25]

    Retrieval-augmented generation for knowledge-intensive NLP tasks

    Lewis, Patrick, Perez, Ethan, Piktus, Aleksandra, Petroni, Fabio, Karpukhin, Vladimir, Goyal, Naman, Küttler, Heinrich, Lewis, Mike, Yih, Wen-Tau, Rocktäschel, Tim, Riedel,SebastianandKiela,Douwe. “Retrieval-augmented generation for knowledge-intensive NLP tasks.” Ad- vances i...

    2020

  18. [26]

    Enhancing Retrieval-Augmented Generation: AStudyofBestPractices

    Li, Siran, Stenzel, Linus, Eickhoff, Carsten and Ali Bahrainian, Seyed. “Enhancing Retrieval-Augmented Generation: AStudyofBestPractices.” Vol.PartF206484- 1: p. 6705 – 6717. 2025. URL https://www.scopus.com/ inward/record.uri?eid=2-s2.0-85218493737&partnerID= 40&md5=b24733a11...

    2025

  19. [27]

    Im- proving large language model applications in biomedicine with retrieval-augmented generation: a systematic re- view, meta-analysis, and clinical development guide- lines

    Liu, Siru, McCoy, Allison B and Wright, Adam. “Im- proving large language model applications in biomedicine with retrieval-augmented generation: a systematic re- view, meta-analysis, and clinical development guide- lines.” Journal of the American Medical Informatics As- sociat...

    2025 doi

  20. [28]

    Question-Based Retrieval using Atomic Units for Enterprise RAG

    Raina, Vatsal and Gales, Mark. “Question-Based Retrieval using Atomic Units for Enterprise RAG.”: p. 219 –

  21. [29]

    Benchmarking Large Language Models in Retrieval-Augmented Generation

    Chen, Jiawei, Lin, Hongyu, Han, Xianpei and Sun, Le. “Benchmarking Large Language Models in Retrieval-Augmented Generation.” Vol. 38. 16: p. 17754 – 17762. 2024. URL https://www.scopus. com/inward/record.uri?eid=2-s2.0-85189613527&doi= 10.1609%2faaai.v38i16.29728&partnerID=40&...

    2024

  22. [30]

    In- formation Extraction of Aviation Accident Causation Knowledge Graph: An LLM-Based Approach

    Chen, Lu, Xu, Jihui, Wu, Tianyu and Liu, Jie. “In- formation Extraction of Aviation Accident Causation Knowledge Graph: An LLM-Based Approach.” Elec- tronics (Switzerland) Vol. 13 No. 19 (2024). DOI 10.3390/electronics13193936. URL https://www.scopus. com/inward/record.uri?eid...

    2024 doi

  23. [31]

    MechBERT: Language Models for Extracting Chemical 9 and Property Relationships about Mechanical Stress and Strain

    Kumar, Pankaj, Kabra, Saurabh and Cole, Jacqueline M. “MechBERT: Language Models for Extracting Chemical 9 and Property Relationships about Mechanical Stress and Strain.” Journal of Chemical Information and Modeling Vol. 65 No. 4 (2025): p. 1873 – 1888. DOI 10.1021/acs.jcim.4c...

    2025 doi

  24. [32]

    An Approach to Intelligent Information Extraction and Utilization from Diverse Documents

    Joshi, Raghav, Bubna, Yash, Sahana, M. and Shruthiba, A. “An Approach to Intelligent Information Extraction and Utilization from Diverse Documents.” Conference paper

  25. [33]

    Single Engine Models 172, 182, T182, 206 AND T206 1996 And On

    COMPANY, CESSNA AIRCRAFT. “Single Engine Models 172, 182, T182, 206 AND T206 1996 And On.” http://www.aeroelectric.com/Reference_Docs/Cessna/ cessna-maintenance-manuals/CessnaSingle_1996on_ structural_repair_MM_SESR04.pdf. Accessed: August 04, 2024

    1996

  26. [34]

    meta-llama/Llama-3.3-70B-Instruct

    Meta. “meta-llama/Llama-3.3-70B-Instruct.” https:// huggingface.co/meta-llama/Llama-3.3-70B-Instruct. Ac- cessed: February 26, 2025

    2025

  27. [35]

    LangGraph

    LangChain. “LangGraph.” https://www.langchain.com/ langgraph. Accessed: March 07, 2025

    2025

  28. [36]

    meta-llama/Llama-3.1-8B-Instruct

    Meta. “meta-llama/Llama-3.1-8B-Instruct.” https:// huggingface.co/meta-llama/Llama-3.1-8B-Instruct. Ac- cessed: February 26, 2025

    2025

  29. [37]

    fine-tune: full fine-tuning pipeline

    Bogachov, Bogdan. “fine-tune: full fine-tuning pipeline.” (2025). URLhttps://github.com/bogdanbogachov/eng_llm/ blob/main/finetune/finetune.py.Accessed: March20,2025

    2025

  30. [38]

    “Llama.” https://www.llama.com/

    Meta. “Llama.” https://www.llama.com/. Accessed: March 21, 2025

    2025

  31. [39]

    The Bittorrent P2P File-Sharing System: Mea- surements and Analysis

    Pouwelse, J.A., Garbacki, Pawel, Epema, D. and Sips, Henk. “The Bittorrent P2P File-Sharing System: Mea- surements and Analysis.” Vol. 3640: pp. 205–216. 2005. DOI 10.1007/11558989_19. 10

    2005 doi

  32. [233]

    URL https://www.scopus.com/inward/record

    2024. URL https://www.scopus.com/inward/record. uri?eid=2-s2.0-85216930459&partnerID=40&md5= 0fba8949ac025ab64333e07b555615d5

    2024

  33. [2024]

    URL https://www.scopus.com/inward/record

    DOI 10.1109/CSITSS64042.2024.10816908. URL https://www.scopus.com/inward/record. uri?eid=2-s2.0-85216926987&doi=10.1109% 2fCSITSS64042.2024.10816908&partnerID=40&md5= 80eec4445ca534540378b17544e3014d. Cited by: 0

    2024

Pith tools

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