Lost in the Middle: How Language Models Use Long Contexts
Open accessThe DOI matches its Crossref record; 48 of the 59 references it cites were found in a registry or an index; 15 of the 23 sources its claims cite were checked for support.
Abstract
While recent language models have the ability to take long contexts as input, relatively little is known about how well they use longer context. We analyze the performance of language models on two tasks that require identifying relevant information in their input contexts: multi-document question answering and key-value retrieval. We find that performance can degrade significantly when changing the position of relevant information, indicating that current language models do not robustly make use of information in long input contexts. In particular, we observe that performance is often highest when relevant information occurs at the beginning or end of the input context, and significantly degrades when models must access relevant information in the middle of long contexts, even for explicitly long-context models. Our analysis provides a better understanding of how language models use their input context and provides new evaluation protocols for future long-context language models.
Overview
CoChat Summary
TL;DR
Language models’ performance on tasks requiring access to long input contexts is strongly affected by where relevant information appears, with a U-shaped curve: best at the beginning or end and worst in the middle. Extended-context models are not necessarily better at using context, and reader performance saturates before retriever recall in open-domain QA.
Key findings
- In multi-document QA, performance is highest when the answer-containing document is at the start or end of the context and degrades in the middle; GPT-3.5-Turbo can drop by more than 20%, and in 20- and 30-document settings middle-position performance is lower than its closed-book performance of 56.1% (oracle: 88.3%).
- Extended-context models are not necessarily better at using input context: when the input fits both windows, GPT-3.5-Turbo and GPT-3.5-Turbo(16K) perform nearly identically as a function of relevant-information position.
- In synthetic key-value retrieval, Claude-1.3 and Claude-1.3(100K) do nearly perfectly, while other models struggle especially with 140 or 300 key-value pairs; query-aware contextualization gives GPT-3.5-Turbo(16K) perfect performance at 300 pairs, versus worst-case 45.6% without it.
- In open-domain QA, reader performance saturates long before retriever recall: using 50 instead of 20 retrieved documents only marginally improves performance (∼1.5% for GPT-3.5-Turbo and ∼1% for Claude-1.3).
Methods
- Controlled multi-document QA on NaturalQuestions-Open: 2,655 queries whose annotated long answer is a paragraph; each context has exactly one answer-containing Wikipedia passage plus k−1 Contriever/MS-MARCO distractor passages; the paper varies answer position and context length (10, 20, 30 documents) and measures accuracy.
- Synthetic key-value retrieval: inputs are serialized JSON objects with unique random UUID key-value pairs; models must return the value for a specified key; experiments use 75, 140, and 300 key-value pairs (500 examples each), varying key position and number of pairs, with accuracy as the metric.
- Models include open MPT-30B-Instruct and LongChat-13B(16K), closed GPT-3.5-Turbo, GPT-3.5-Turbo(16K), Claude-1.3, and Claude-1.3(100K), plus additional architecture/instruction experiments with Flan-T5-XXL, Flan-UL2, MPT-30B, Llama-2, and GPT-4; an open-domain QA case study uses a Contriever retriever-reader setup.
Limitations
- Stated: the paper uses greedy decoding and leaves exploration of other decoding methods to future work; the investigations into why models are not robust are described as preliminary; GPT-4 is evaluated only on a subset of 500 multi-document QA examples because full evaluation would cost upwards of $6000.
- Stated: NaturalQuestions-Open has ambiguity, so the authors run an unambiguous subset and also test random distractors and randomized distractor order, finding similar trends.
- Evident: the controlled multi-document QA setting always contains exactly one document that answers the question, unlike open-domain QA; the synthetic key-value task removes natural language semantics; results may not generalize to all tasks, models, or decoding settings.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
7 Valid · 0 Not valid · 5 Not checked · 13 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Language models have become an important and flexible building block in a variety of user-facing language technologies, including conversational interfaces, search and summarization, and collaborative writing (Shuster et al., 2022;Thoppilan et al., 2022;Lee et al., 2022, inter alia).”
§ Lost in the Middle: How Language Models Use Long Contexts > 1 Introduction
What the sources say
- Partly supportedcites Shuster et al., 2022 sourceExperimental — accuracy not yet measured
Section 3.1.1 (p. 4) describes a conversational agent built on a language model that generates internet search queries, supporting conversational interfaces and search as part of dialogue. Section 2 (p. 3) mentions summarization only as related work, and the passages do not establish collaborative writing as an application of this paper’s model. (LLM, conf=0.85)
- Supportedcites Thoppilan et al., 2022 source
The Abstract (p. 1) identifies LaMDA as a family of language models specialized for dialog. In this sentence's list of applications, Thoppilan et al. supports the conversational-interface example; the other cited works need not each cover every example. (LLM, conf=0.92)
- Supportedcites Lee et al., 2022 source
Lee et al. is cited alongside other works, and its contribution to the sentence is the collaborative-writing example. Section 4.2.2 describes writers receiving and selecting model-generated suggestions in a writing interface; Section 2.2.1 also describes language models as applicable to dialogue and summarization. The paper need not establish every example in the jointly cited list. (LLM, conf=0.89)
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“These observations complement prior work, which found that non-instruction fine-tuned language models are biased towards recent tokens (i.e., the end of the input context; Khandelwal et al., 2018;Press et al., 2021).”
§ Lost in the Middle: How Language Models Use Long Contexts > 4.3 Effect of Instruction Fine-Tuning
What the sources say
- Partly supportedcites Khandelwal et al., 2018 sourceExperimental — accuracy not yet measured
The Introduction and Section 5 report that the studied LSTM is much more sensitive to recent word order than to distant word order, supporting a recency effect. The paper studies a standard LSTM on two datasets, however, rather than establishing this tendency across non-instruction-fine-tuned language models generally; it also finds that distant words remain relevant. (LLM, conf=0.82)
- Partly supportedcites Press et al., 2021 sourceExperimental — accuracy not yet measured
Section 3.1 (page not available in our copy) finds that very short contexts hurt performance, while increasing subsequence length from 1,024 to 3,072 tokens gives negligible improvement. That is consistent with limited benefit from more distant context, but the paper does not establish that the models are biased toward recent tokens within an input context. (LLM, conf=0.78)
- Partly supported
The sources with a verdict support this claim; the rest weren’t checked yet, have no full text we could read, or their verdicts aren’t rated yet.
In the paper
“We instantiate this task with data from NaturalQuestions-Open (Lee et al., 2019;Kwiatkowski et al., 2019), which contains historical queries issued to the Google search engine, coupled with human-annotated answers extracted from Wikipedia.”
§ Lost in the Middle: How Language Models Use Long Contexts > 2.1 Experimental Setup
What the sources say
- Supportedcites Lee et al., 2019 source
Section 7.1 says the paper creates an open version of Natural Questions and uses it for training and evaluation. The Introduction describes the task as answering questions by retrieving evidence from an open corpus, supporting the cited use of NaturalQuestions-Open for this task. (LLM, conf=0.91)
- Not checked — no full text of the sourcecites Kwiatkowski et al., 2019 source
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“Following Kandpal et al. (2022) and Mallen et al. (2023), we use accuracy as our primary evaluation metric, judging whether any of the cor- rect answers (as taken from the NaturalQuestions annotations) appear in the predicted output.”
§ Lost in the Middle: How Language Models Use Long Contexts > 2.1 Experimental Setup
What the sources say
- Supportedcites Kandpal et al. (2022) source
Section 3 (p. 4) says the paper evaluates answers using exact match against a ground-truth answer set, and Section 3.1 (p. 4) reports QA accuracies as its evaluation results. This supports citing Kandpal et al. for using accuracy to evaluate question answering. (LLM, conf=0.90)
- Supportedcites Mallen et al. (2023) source
Section 3.1 (‘Focus and Task’) explicitly names accuracy as the metric and says a prediction is correct when it contains an exact match to any gold answer. This supports the evaluation approach attributed to Mallen et al.; the NaturalQuestions annotations describe the citing authors’ answer source. (LLM, conf=0.94)
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“or when language models are augmented with external information (relevant documents from a search engine, database query results, etc.; Petroni et al., 2020;Ram et al., 2023;Shi et al., 2023;Mallen et al., 2023;Schick et al., 2023, inter alia).”
§ Lost in the Middle: How Language Models Use Long Contexts > 1 Introduction
What the sources say
- Partly supportedcites Petroni et al., 2020 sourceExperimental — accuracy not yet measured
The Abstract (p. 1) says the paper integrates information from a retrieval system with a pre-trained language model, and §3.4.2 (p. 5) specifies a retrieved paragraph as context. This supports augmentation with retrieved documents, but our copy does not show language models being augmented with database query results. (LLM, conf=0.84)
- Partly supportedcites Ram et al., 2023 sourceExperimental — accuracy not yet measured
The Introduction describes augmenting language models with relevant documents retrieved from an external knowledge source, and Section 5 describes using a BM25 retriever to find those documents. The passages do not establish database query results as another type of external information used by this paper. (LLM, conf=0.82)
- Partly supportedcites Shi et al., 2023 sourceExperimental — accuracy not yet measured
Section 3.1 (page unavailable) describes retrieving relevant documents from a corpus, and the Conclusion says REPLUG uses those documents to augment a language model. The paper supports the document-retrieval example, but our copy does not describe augmenting models with database query results or documents from a search engine specifically. (LLM, conf=0.82)
- Partly supported
20 more claims · 2 with their cited sources checked
Create a free account to see all 25 claims and every source verdict.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“In vanilla multi-document or NIAH-based benchmarks (Kamradt, 2023; Liu et al., 2024), models perform consistently well when the needle (or gold document) appears at the very beginning or end of the context window, with minimal impact from context length.”
From NoLiMa: Long-Context Evaluation Beyond Literal Matching (2025) · cites this paper as Liu et al., 2024
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Even within a single long sequence, attention effectiveness degrades with distance due to attention dilution , positional encoding limitations, and token interference, leading to the well-known 'lost-in-the-middle' and context-decay phenomena (Liu et al., 2024; Press et al., 2021a).”
From MAGMA: A Multi-Graph based Agentic Memory Architecture for AI Agents (2026) · cites this paper as Liu et al., 2024
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Current large language models (LLMs) exhibit significant deficiencies in episodic memory tasks including encoding, storing, and retrieving specific information from temporally dependent events over a long period of time.”
From ARTEM: Enhancing Large Language Model Agents with Spatial-Temporal Episodic Memory (2026) · cites this paper as Liu et al. 2024
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Recent approaches to handle memory tasks in LLMs, such as in-context learning, retrieval-augmented generation (RAG), and fine-tuning, may resolve the long-term retention issues, but are still inadequate to handle tasks requiring chronological awareness of the stored information.”
From ARTEM: Enhancing Large Language Model Agents with Spatial-Temporal Episodic Memory (2026) · cites this paper as Liu et al. 2024
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“If not ranked at the top, the key information may be neglected by the generative module, resulting in incomplete or inaccurate responses to the user queries24.”
From Leveraging long context in retrieval augmented language models for medical question answering (2025) · cites this paper as [24]
Experimental — accuracy not yet measured
86 more claims about this paper
Create a free account to see every claim other papers make about this one.
Citations
Cites
Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., & Polosukhin, I. (2017) · arXiv · doi:10.48550/arxiv.1706.03762 · Cited by 120,404
Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer
Raffel, C., Shazeer, N., Roberts, A., Lee, K., Narang, S., Matena, M., Zhou, Y., Li, W., & Liu, P. J. (2019) · arXiv · doi:10.48550/arxiv.1910.10683 · Cited by 10,929
LLaMA: Open and Efficient Foundation Language Models
Touvron, H., Lavril, T., Izacard, G., Martinet, X., Lachaux, M.-A., Lacroix, T., Rozière, B., Goyal, N., Hambro, E., Azhar, F., et al. (2023) · arXiv · doi:10.48550/arxiv.2302.13971 · Cited by 4,602
Transformer-XL: Attentive language models beyond a fixed-length context
Dai, Z., Yang, Z., Yang, Y., Carbonell, J., Le, Q., & Salakhutdinov, R. (2019) · Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/p19-1285 · Cited by 3,169 · cited 2×
“Recent improvements in hardware (e.g., faster GPUs with more memory) and algorithms (Dai et al., 2019; Dao et al., 2022; Poli et al., 2023; Rubin and Berant, 2023, inter alia ) have resulted in language models with larger context windows (e.g., 4096, 32K, and even 100K tokens), but it remains…”
BackgroundLlama 2: Open Foundation and Fine-Tuned Chat Models
Touvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y., Bashlykov, N., Batra, S., Bhargava, P., Bhosale, S., et al. (2023) · arXiv · doi:10.48550/arxiv.2307.09288 · Cited by 2,604
Longformer: The Long-Document Transformer
Beltagy, I., Peters, M. E., & Cohan, A. (2020) · arXiv · doi:10.48550/arxiv.2004.05150 · Cited by 2,281
Leveraging Passage Retrieval with Generative Models for Open Domain Question Answering
Izacard, G., & Grave, E. (2020) · Conference of the European Chapter of the Association for Computational Linguistics · doi:10.48550/arxiv.2007.01282 · Cited by 2,263
Natural Questions: A benchmark for question answering research
Kwiatkowski, T., Palomaki, J., Redfield, O., Collins, M., Parikh, A., Alberti, C., Epstein, D., Polosukhin, I., Devlin, J., Lee, K., et al. (2019) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl_a_00276 · Cited by 2,069 · cited 1×
“We instantiate this task with data from NaturalQuestions-Open (Lee et al., 2019; Kwiatkowski et al., 2019), which contains historical queries issued to the Google search engine, coupled with human-annotated answers extracted from Wikipedia.”
MethodThe serial position effect of free recall.
Murdock, B. B. (1962) · Journal of Experimental Psychology · doi:10.1037/h0045106 · Cited by 1,717 · cited 1×
“The U-shaped curve we observe in this work has a connection in psychology known as the serial-position effect (Ebbinghaus, 1913; Murdock Jr, 1962), that states that in free-association recall of elements from a list, humans tend to best re-member the first and last elements of the list.”
MethodMemory: A contribution to experimental psychology.
Ebbinghaus, H. (1913) · doi:10.1037/10011-000 · Cited by 1,645 · cited 1×
“The U-shaped curve we observe in this work has a connection in psychology known as the serial-position effect (Ebbinghaus, 1913; Murdock Jr, 1962), that states that in free-association recall of elements from a list, humans tend to best re-member the first and last elements of the list.”
Method
Research works printed but not linked: 1
- Sharan, 2018, Prediction with a short memory, 10.18653/v1/K18-3013 · DOI
Cited by
Items 501–600 of 1,592
Onishi, S., Shibata, D., Tsujikawa, M., Kubo, M., Nakagawa, A., Ishii, R., Furukawa, H., Tominaga, J., & Ota, H. (2025) · Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) · doi:10.1109/embc58623.2025.11252841 · Cited by 2
“Numerous studies have reported that LLMs suffer from decreased accuracy when the prompt length becomes too long [17], [18], [19].”
Farooq, A., Anuragi, D., Li, Z., Ziat, M., Cooperstock, J. R., & Raisamo, R. (2025) · Human-Intelligent Systems Integration · doi:10.1007/s42454-025-00085-9 · Cited by 3
Advancing adversarial and LLM robustness in trustworthy AI: a comprehensive survey
Tang, Q., Qian, J., Du, X., Wang, S., & Liu, H. (2026) · Artificial Intelligence Review · doi:10.1007/s10462-026-11558-x · Cited by 4
Gómez, F. J. D., & Galende, J. N. (2024) · Revista Tecnología Ciencia y Educación · doi:10.51302/tce.2024.19081 · Cited by 6
“…(Onoe et al. , 2022), incluyendo la utilización deficiente de tales datos, como ocurre en ocasiones ante el empleo de atajos (Kandpal et al. , 2023; Kang y Choi, 2023; Li et al. , 2022) y fallos en la recu-peración de conocimientos (Zheng, 2023; Liu, Lin et al. , 2023; Mallen et al. , 2023).”
BackgroundHoltzman, A., West, P., & Zettlemoyer, L. (2025) · Journal of Social Computing · doi:10.23919/jsc.2025.0009 · Cited by 38
Yu, H., Xu, X., Guo, Z., Tianming, W. E. I., & Xu, L. (2026) · Journal of Marine Science and Engineering · doi:10.3390/jmse14050448 · Cited by 3
Hilkenmeier, F., Stoltenberg, M., & Stierle, C. M. G. (2026) · Systematic Reviews · doi:10.1186/s13643-026-03228-4 · Cited by 8
“In this sense, the task is less affected by performance degradation associated with longer or more complex input contexts [49] and less reliant on the kind of open-ended generation in which hallucinated content is more likely to arise [16, 35, 37].”
Remind: Recall Enhanced Memory Integration for Natural Language Dialogue Systems
Huang, B.-S., Kuo, T. T., Wang, L.-J., & Lin, C.-Y. (2025) · IEEE Conference on Artificial Intelligence (CAI) · doi:10.1109/cai64502.2025.00125 · Cited by 2
“Existing solutions, such as expanded context windows [3] and compressive memory techniques [4], partially address memory limitations but remain computationally intensive and susceptible to phenomena like lost-in-the-middle [5].”
Narrative and dialogue generation for video-games: A systematic mapping
Salmaze, P. H., Pereira, G. D. S., de Góes, L. M. G., Ranieri, C. M., Toledo, C. F. M., Silvestrini, M. M., & Pereira, L. P. D. B. (2026) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2026.115041 · Cited by 5
Rank It, Then Ask It: Input Reranking for Maximizing the Performance of LLMs on Symmetric Tasks
Dehghankar, M., & Asudeh, A. (2025) · ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) · doi:10.1145/3711896.3737099 · Cited by 8
Luo, S., Yao, Y., He, B., Shao, W., Xu, J., Huang, Y., Zhou, A., Zhang, X., Xiao, Y., Hou, H., et al. (2026) · IEEE Journal of Selected Topics in Signal Processing · doi:10.1109/jstsp.2026.3653160 · Cited by 71
“From Equation (7), we can observe that for user 𝑢 , the further they are positioned in the long tail ( i.e. , the fewer items they have interacted with), the lower is the value of ℓ 𝑢 and the higher is the value of 𝛼 𝑢 .”
Lost but Not Only in the Middle: Positional Bias in Retrieval Augmented Generation
Hutter, J., Rau, D., Marx, M., & Kamps, J. (2025) · Lecture notes in computer science · doi:10.1007/978-3-031-88708-6_16 · Cited by 9
A method for semantic textual similarity on long texts
Zatarain, O. A., González-Castolo, J. C., & Ramos-Cabral, S. (2025) · PeerJ Computer Science · doi:10.7717/peerj-cs.3202 · Cited by 2
ChatMPI: LLM-Driven MPI Code Generation for HPC Workloads
Valero‐Lara, P., Young, A., Naughton, T. J., Engelmann, C., Geist, A., Vetter, J. S., Teranishi, K., & Godoy, W. F. (2026) · International Conference on High Performance Computing in Asia-Pacific Region (HPC Asia) · doi:10.1145/3773656.3773659 · Cited by 4
Teaching Model-Driven Low-Code Development Platforms
Charles, J., Michael, J., Netz, L., & Rumpe⋆, B. (2024) · ACM/IEEE International Conference on Model Driven Engineering Languages and Systems (MODELS) · doi:10.1145/3652620.3687778 · Cited by 3
InsightAI: Root Cause Analysis in Large Log Files with Private Data Using Large Language Model
Ekhlasi, M., Prakash, A., Lamothe, M., & Dagenais, M. (2025) · IEEE/ACM International Conference on AI Engineering – Software Engineering for AI (CAIN) · doi:10.1109/cain66642.2025.00012 · Cited by 5
“This is because they require advanced GPUs with significant memory capacity to function efficiently [15]–[19].”
Generative AI for autonomous data analytics
Fikardos, M., Lepenioti, K., Bousdekis, A., Apostolou, D., & Mentzas, G. N. (2026) · Intelligent Systems with Applications · doi:10.1016/j.iswa.2026.200626 · Cited by 5
Trustworthy Information Retrieval in the LLM Era: Bias, Unfairness, and Hallucination
Dai, S., Xu, C., Xu, S., Sun, Z., Pang, L., Dong, Z., & Xu, J. (2025) · Annual International ACM SIGIR Conference on Research and Development in Information Retrieval in the Asia Pacific Region (SIGIR-AP) · doi:10.1145/3767695.3769670 · Cited by 5
Üniversite Öğrencilerinin Yazılı Üretimlerinin OpenAI GPT ile Değerlendirilmesi
SAYIN, A., & Melanlıoğlu, D. (2024) · Yuksekogretim Dergisi · doi:10.53478/yuksekogretim.1418870 · Cited by 3
Streaming Videollms for Real-Time Procedural Video Understanding
Chatterjee, D., Remelli, E., Song, Y., Tekin, B., Mittal, A., Bhatnagar, B., Camgöz, N. C., Hampali, S., Sauser, E. L., Ma, S., et al. (2025) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv51701.2025.02097 · Cited by 19
“1) are re-ported on a single A6000 GPU, excluding flash attention [2] for fair comparison.”
Sac-lstm: optimal resource allocation based on user intent in computility networks
Zheng, Y., Chen, N., Yin, Y., et al. (2025) · The Journal of Supercomputing · doi:10.1007/s11227-025-07718-5 · Cited by 2
“LSTM capture long-range dependencies, suitable for tasks like machine translation [27] and language modeling [28].”
Augmenting, Not Replacing: The Role of LLMs in Human-Centric Formal RE
Halili, S., Spoletini, P., & Grubb, A. M. (2025) · IEEE International Requirements Engineering Conference (RE) · doi:10.1109/re63999.2025.00029 · Cited by 2
“We strategically placed this example at the end of the prompt rather than the middle, as LLMs have been shown to exhibit U-shaped attention bias and the order in which samples are provided can make the difference between near state-of-the-art and random guess performance [30].”
Tutorial: guidance on the use of large language models for medical research
Jin, Q., Wan, N., Leaman, R., Tian, S., Wang, Z., Yang, Y., Wang, Z., Xiong, G., Lai, P.-T., Zhu, Q. L., et al. (2026) · Nature Protocols · doi:10.1038/s41596-026-01408-z · Cited by 16
Chain-of-Detection enables robust and efficient jailbreak defense
Wu, T., & Zhang, H. (2025) · Neural Networks · doi:10.1016/j.neunet.2025.108217 · Cited by 4
Alba, C., Wang, X., Wang, C., et al. (2025) · ACM Technical Symposium on Computer Science Education (SIGCSE) · doi:10.1145/3641555.3705141 · Cited by 2
Avula, S., Lee, C., Zhang, R., & Murdock, V. (2025) · International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR) · doi:10.1145/3726302.3730230 · Cited by 5
“Recent work has shown that the Generator can place significant attention on passages appearing later in the context [15, 17, 27], with attention distributions that deviate substantially from the predictable decay patterns observed in traditional ranking systems.”
Structure-Grounded Document Agents for Faithful Long-Context Reasoning
Zhou, J., Zhu, Z., Zhu, A., Wang, S., & Zheng, Y. (2026) · International Conference on Intelligent Communications and Computing (ICICC) · doi:10.1109/icicc71012.2026.11637767 · Cited by 7
“Liu et al. [1] demonstrated that LLMs exhibit a characteristic U-shaped performance pattern, performing well on information near the beginning and end of the input but degrading substantially when relevant evidence resides in the middle.”
ReLOpt: A Retriever-Augmented Framework for Optimizing Code with Long-Range Dependencies
Qiu, L., Lyu, F., Ming, Z., Wang, L., & Feng, X. (2025) · Lecture notes in computer science · doi:10.1007/978-981-95-4367-0_4 · Cited by 2
Zabihi, P., Nawara, D., Ibrahim, A., & Kashef, R. F. (2026) · Applied Sciences · doi:10.3390/app16073410 · Cited by 5
Zhao, Z., Qu, C., Dou, Z., Chen, H., & Jin, J. (2025) · ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) · doi:10.1145/3711896.3737104 · Cited by 3
“Considering that LLMs may respond differently depending on the order of demonstrations presented in the prompt [12, 24, 41], we also conduct a full permutation of the order of demonstrations within the prompt.”
GEMMA-SQL: A Novel Text-to-SQL Model Based on Large Language Models
Pandey, H. M., Gupta, A., Sarkar, S., Tomer, M., Schneider, J., & Gong, Y. (2025) · Applied Artificial Intelligence · doi:10.1080/08839514.2025.2587371 · Cited by 4
“Once the schemas are identified, they are passed to an LLM to generate the final SQL queries.”
A collaborative reasoning framework for large language models in long-context Q&A
Yao, J., He, G., & Xu, X. (2025) · Expert Systems with Applications · doi:10.1016/j.eswa.2025.129960 · Cited by 4
Experience Retrieval Compression for Continual Adaptation of LLM Agents
Zhu, Z., Fang, R., & Hu, R. (2026) · International Conference on Artificial Intelligence and Electromechanical Automation (AIEA) · doi:10.1109/aiea70743.2026.11632622 · Cited by 6
“First, agent trajectories may span 10–30 turns, translating to hundreds or thousands of tokens per retrieved example [4]; injecting several such trajectories can exhaust the context window.”
Selvaraj, S. P. P., Yen, R. W., Forcino, R. C., & Elwyn, G. J. (2025) · Patient Education and Counseling · doi:10.1016/j.pec.2025.109362 · Cited by 6
Hydra-Reviewer: A Holistic Multi-Agent System for Automatic Code Review Comment Generation
Ren, X., Dai, C., Huang, Q., Wang, Y., Liu, C., & Jiang, B. (2025) · IEEE Transactions on Software Engineering · doi:10.1109/tse.2025.3621462 · Cited by 14
“As the length of the input context increases, the LLM exhibits performance degradation and tends to overlook critical information in long inputs [17].”
Guo, Z., & Duan, Y. (2026) · Electronics · doi:10.3390/electronics15050963 · Cited by 2
AuditAgent: Expert-Guided Multi-Agent Reasoning for Cross-Document Fraudulent Evidence Discovery
Bai, S., Wu, B., Zhang, Y., Chengke, W., Zheng, X., Yuan, Y., Wu, K., & Li, J. (2025) · ACM International Conference on AI in Finance (ICAIF) · doi:10.1145/3768292.3770383 · Cited by 8
Robust Pronoun Fidelity with English LLMs: Are they Reasoning, Repeating, or Just Biased?
Gautam, V., Bingert, E., Zhu, D., Lauscher, A., & Klakow, D. (2024) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl_a_00719 · Cited by 22
“Your task is to determine the appropriate pronoun to fill in the blank, and enter it in the “Pronoun” column.”
Unifying Corroborative and Contributive Attributions in Large Language Models
Worledge, T., Shen, J. H., Meister, N., Winston, C., & Guestrin, C. (2024) · IEEE Conference on Secure and Trustworthy Machine Learning (SaTML) · doi:10.1109/satml59370.2024.00039 · Cited by 25
Liu, B., Li, W., Zhao, R., Klauer, R. R., Hansen, A., Miller, N., Chen, Y., Blenner, M., & Tang, Y. (2025) · Environmental Science & Technology · doi:10.1021/acs.est.5c14258 · Cited by 3
Language Models and Food–Health Evidence: Challenges, Opportunities, and Implications
Jackson, D., Gousiopoulos, A., & Soldatos, T. G. (2026) · BioMedInformatics · doi:10.3390/biomedinformatics6020013 · Cited by 2
“…with own shortcomings (including retrieval of irrelevant passages, context-window truncation and chunking constraints, “lost-in-the-middle” phenomenon in long contexts ([95,96]), and conflicts between retrieved and parametric knowledge), which can themselves generate or perpetuate errors [97].”
The Dual Imperative in AI for OCD: Bridging Ethical Frameworks and Explainable Diagnostics
Zaboski, B. A., & Muller, G. N. (2026) · AI in Medicine · doi:10.3390/aimed1030017 · Cited by 2
“While current AI memory models have limitations (e.g., the tendency to lose context buried in the middle of long conversations [43]), this structural constraint provides a stronger safety net than out-of-the-box commercial chatbots.”
MLLM-Search: A Zero-Shot Approach to Finding People Using Multimodal Large Language Models
Fung, A., Tan, A. H., Wang, H., Benhabib, B., & Nejat, G. (2025) · Robotics · doi:10.3390/robotics14080102 · Cited by 13
Leveraging Open Big Data from R&D Projects with Large Language Models
Ruiz, D., Cardinale, Y., Casas, A., & Moscardó, V. (2025) · Big Data and Cognitive Computing · doi:10.3390/bdcc9020026 · Cited by 3
Analyzing Logs of Large-Scale Software Systems Using Time Curves Visualization
Borysenkov, D., Vogel, A., Henning, S., & Pérez-Wohlfeil, E. (2025) · IEEE International Conference on Software Analysis, Evolution and Reengineering (SANER) · doi:10.1109/saner64311.2025.00038 · Cited by 4
Rethinking Chart Understanding Using Multimodal Large Language Models
Tanasa, A.-M., & Oprea, S. (2025) · Computers, materials & continua/Computers, materials & continua (Print) · doi:10.32604/cmc.2025.065421 · Cited by 4
Roeschl, T., Hoffmann, M., Hashemi, D., Rarreck, F., Hinrichs, N., Trippel, T. D., Gröschel, M. I., Unbehaun, A., Klein, C., Kempfert, J., et al. (2024) · medRxiv · doi:10.1101/2024.11.20.24313385 · Cited by 2
Context-Aware Code Review Automation: A Retrieval-Augmented Approach
İçöz, B., & Biricik, G. (2026) · Applied Sciences · doi:10.3390/app16041875 · Cited by 7
Breaking Size Barrier: Enhancing Reasoning for Large-Size Table Question Answering
Wu, X., Liang, D., Yang, J., Cheng, X., Chai, L., Li, T., Yang, L., & Li, Z. (2026) · Lecture notes in computer science · doi:10.1007/978-981-95-3830-0_16 · Cited by 12
Multi-agent Chatbot for Efficient Interaction with Blockchain APIs
Nguyen, S.-H.-D., Trinh, T.-D., & Tran, Q.-V.-Q. (2025) · Communications in computer and information science · doi:10.1007/978-981-96-4288-5_33 · Cited by 2
Medical multimodal large language models: A survey
Xiao, H., Hui, N., Xu, Y. D., Li, Z., & Peng, J. (2026) · Information Fusion · doi:10.1016/j.inffus.2026.104386 · Cited by 5
Mind the semantic gap: semantic efficiency in human computer interfaces
Horsley, J. (2025) · Frontiers in Artificial Intelligence · doi:10.3389/frai.2025.1451865 · Cited by 3
“…window is the only way to provide context that cannot be learned during training, (2) the semantic context itself is encoded through verbose forms such general human language, and (3) overproviding context comes with penalties to computational cost, latency, and output quality (Liu et al., 2024).”
Widiiswa, R. A. N., Henry, M. M., Imani, A., & Pardamean, B. (2026) · International Conference on Information Technology · doi:10.1109/it67293.2026.11435730 · Cited by 2
“Preserving semantic integrity during chunking is critical [23], [26].”
Quantifying Uncertainty: A New Era of Measurement through Large Language ModelsPreprint
Audrino, F., Gentner, J., & Stalder, S. (2024) · SSRN Electronic Journal · doi:10.2139/ssrn.4903414 · Cited by 4
LURE-RAG: Lightweight Utility-Driven Reranking for Efficient RAG
Chandra, M., Ganguly, D., & Ounis, I. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-21289-4_36 · Cited by 7
Gatt, A. (2025) · Oxford University Press eBooks · doi:10.1093/acrefore/9780199384655.013.896 · Cited by 2
Improving Knowledge Base Question Answering via Retrieval Enhancement and Stepwise Reasoning
Huang, D. K., Gao, J., Luo, X., & Wu, H. (2025) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp49660.2025.10890582 · Cited by 4
“However, achieving comprehensive knowledge coverage requires a substantial amount of context, which can overwhelm LLMs and hinder their understanding [12].”
Write, Rank, or Rate: Comparing Methods for Studying Visualization Affordances
Stokes, C., Lin, K., & Xiong, C. (2025) · IEEE Transactions on Visualization and Computer Graphics · doi:10.1109/tvcg.2025.3633872 · Cited by 4
Self-citationToward trustworthy chatbots: a protocol for red teaming for health related conversations
Hussain, S.-A., Jackson, D. I., Lewis, A., Fosler‐Lussier, E., & Sezgın, E. (2026) · Scientific Reports · doi:10.1038/s41598-026-45719-3 · Cited by 6
LVMScissor: Split and Schedule Large Vision Model Inference on Mobile Edges via Salp Swarm Algorithm
Liu, Y., Lu, R., & Wang, D. (2025) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2025.3550519 · Cited by 6
CacheFormer: High-Attention-Based Segment Caching
Singh, S. K., & Mahmood, A. (2025) · AI · doi:10.3390/ai6040085 · Cited by 2
Yan, S., Yang, X., Huang, Z., Nie, E., Ding, Z., Li, Z., Ma, X., Bi, J., Kersting, K., Pan, J. Z., et al. (2025) · arXiv · doi:10.48550/arxiv.2508.19828 · Cited by 297
“While this extends access to past information, it also creates a fundamental retrieval challenge: heuristics may return too few entries, omitting crucial context, or too many, flooding the model with irrelevant information and degrading performance (Liu et al., 2023).”
Maragheh, R. Y., & Deldjoo, Y. (2026) · ACM Transactions on Recommender Systems · doi:10.1145/3845988 · Cited by 47
Knoll: Creating a Knowledge Ecosystem for Large Language Models
Zhao, D., Yang, D., & Bernstein, M. S. (2025) · ACM Symposium on User Interface Software and Technology (UIST) · doi:10.1145/3746059.3747711 · Cited by 11
“In cases where we retrieve irrelevant modules or the model ignores the inserted information (e.g., providing too much context leading to the “lost-in-the-middle” phenomenon [64]), Knoll’s answer relevance and quality will suffer compared to that of the baseline.”
MemOS: An Operating System for Memory-Augmented Generation (MAG) in Large Language Models
Li, Z., Song, S., Wang, H., Niu, S., Chen, D., Yang, J., Xi, C., Lai, H., Zhao, J., Wang, Y., et al. (2025) · arXiv · doi:10.48550/arxiv.2505.22101 · Cited by 2
“Explicit short-term memory typically involves prompt concatenation within the context window, but remains limited by context length constraints [18, 21].”
Self-Taught Optimizer (STOP): Recursively Self-Improving Code Generation
Zelikman, E., Lorch, E., Mackey, L., & Kalai, A. T. (2023) · arXiv · doi:10.48550/arxiv.2310.02304 · Cited by 4
“1 .”
LLMLingua-2: Data Distillation for Efficient and Faithful Task-Agnostic Prompt Compression
Pan, Z., Wu, Q., Jiang, H., Xia, M., Luo, X., Zhang, J., Lin, Q., Rühle, V., Yang, Y., Lin, C.-Y., et al. (2024) · arXiv · Cited by 2
PyramidKV: Dynamic KV Cache Compression based on Pyramidal Information Funneling
Cai, Z., Zhang, Y., Gao, B., Liu, Y., Li, Y., Liu, T., Lu, K., Xiong, W., Dong, Y., Hu, J., et al. (2024) · arXiv · doi:10.48550/arxiv.2406.02069 · Cited by 2
MAGMA: A Multi-Graph based Agentic Memory Architecture for AI Agents
Jiang, D., Li, Y., Li, G., & Li, B. (2026) · Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/2026.acl-long.1709 · Cited by 2
The New Shape of Search: How Conversational AI Recomposes Information Seeking
Iannelli, M., & Ai, A. (2026) · arXiv · doi:10.48550/arxiv.2607.04282 · Cited by 2
“The human’s role gains an asking mode alongside the foraging, navigating, and verifying it already did (Figure 1); we quantify each move in §7.”
Chang, Y., Li, Z., Zhang, H., Kong, Y., Wu, Y., So, H. K.-H., Guo, Z., Zhu, L., & Wong, N. (2025) · Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/2025.emnlp-main.790 · Cited by 2
Eliciting In-context Retrieval and Reasoning for Long-context Large Language Models
Qiu, Y., Embar, V., Zhang, Y., Jaitly, N., Cohen, S. B., & Han, B. (2025) · arXiv · doi:10.48550/arxiv.2501.08248 · Cited by 2
“…has significantly expanded their applicability across various domains, including book-level information retrieval (Ding et al., 2024; Jin et al., 2024), summarization (Kim et al., 2024; Saxena and Keller, 2024; Qiu et al., 2023), and question answering (Liu et al., 2024; Wang et al., 2024a).”
Maragheh, R. Y., & Deldjoo, Y. (2025) · arXiv · doi:10.48550/arxiv.2507.02097 · Cited by 2
Medicherla, C., Kulkarni, C., Patel, T., Shivam, S., Nutalapati, V., Parikh, M., Kumari, M., & Elengovan, A. K. (2026) · International Seminar on Intelligent Business and Edge-Computing Research (ISIBER) · doi:10.1109/isiber68248.2026.11470068 · Cited by 2
“…RAG systems hallucinate in 15-35% of cases, with rates varying significantly by domain: 15-20% for factual question answering with high-quality retrieval, 25-30% for conversational systems with diverse topics, and 30-35% for technical or specialized domains where retrieval quality degrades [3,10].”
Retrieving on a topic graph for long document question answering
Chen, S., Liang, B., Yu, Y., Wong, K., Wang, H., & Xu, R. (2025) · Neurocomputing · doi:10.1016/j.neucom.2025.132447 · Cited by 4
Towards a Framework for Hierarchical Text Segmentation using Large Language Models
Flokas, L., Cao, J., Xu, Y., Wu, E., Chu, X., & Yu, C. (2025) · Workshop on Data Management for End-to-End Machine Learning (DEEM) · doi:10.1145/3735654.3735941 · Cited by 1
THINK: Tackling API Hallucinations in LLMs via Injecting Knowledge
Liu, J., Zhang, Y., Wang, D., Li, Y., & Dong, W. (2025) · IEEE International Conference on Software Analysis, Evolution and Reengineering (SANER) · doi:10.1109/saner64311.2025.00029 · Cited by 5
“Lastly, LLMs cannot robustly make use of information in long input contexts [21].”
Bharucha, D., Palanisamy, N., & Reddy, M. (2026) · Biennial International Conference on Nascent Technologies in Engineering (ICNTE) · doi:10.1109/icnte66387.2026.11437450 · Cited by 2
“This technique has led to the improvement of factual correctness in some applications [16], [17], hence it becomes a perfect match for regulatory analysis where accuracy is very important.”
Analyzing Retrieval Scaling in RAG Systems for Complex QA Benchmarks
Mosquera, J. S. R., & Manrique, R. (2025) · SN Computer Science · doi:10.1007/s42979-025-04302-9 · Cited by 1
“Moreover, even when large context windows are supported, model performance often deteriorates as more context is added, particularly due to challenges in attending to relevant information in the middle of long sequences [17].”
Tommasel, A., & Assent, I. (2024) · IEEE International Conference on Big Data (BigData) · doi:10.1109/bigdata62323.2024.10826117 · Cited by 4
Evaluating LLM-Based Detection of Malicious Package Updates in npm
Wyss, E., Tassio, D., Carli, L. D., & Davidson, D. (2025) · International Symposium on Research in Attacks, Intrusions and Defenses (RAID) · doi:10.1109/raid67961.2025.00047 · Cited by 5
“In addition to lower costs, smart-diffs provide significantly smaller input sizes, which are shown to improve LLM classification accuracy in practice [23], as removing large portions of irrelevant and redundant code helps to reduce distractions that the LLM may latch onto [22].”
InfluentialBenchmarking international construction contract knowledge of large language models
Xia, H., Chen, Y., Qi, X., & Wang, Y. (2026) · Expert Systems with Applications · doi:10.1016/j.eswa.2026.131754 · Cited by 3
Empowering LLMs for Long-Text Information Extraction in Chinese Legal Documents
Shen, C., Ji, C., Yue, S., Shen, X., Song, Y. S., Huang, X., & Wei, Z. (2024) · Lecture notes in computer science · doi:10.1007/978-981-97-9431-7_35 · Cited by 1
BRIDGES: Bridging Graph Modality and Large Language Models within EDA Tasks
Li, W., Zou, Y., Ellis, C., Purdy, R., Blanton, R. D., & Moura, J. M. F. (2025) · IEEE International Conference on LLM-Aided Design (ICLAD) · doi:10.1109/iclad65226.2025.00030 · Cited by 1
“Recent studies have found that LLM performance suffers when graph structures are presented as sequential text [15], [16].”
Classification of Distributed Fiber Optic Vibration Signals Using the Diff-ViT Model
Zhang, H., Zhao, Q., Zou, J., et al. (2025) · IEEE Sensors Journal · doi:10.1109/jsen.2025.3599045 · Cited by 1
“For example, in the study by Liu et al. [32], for the same classification model, GAF improved the accuracy by approximately 3.4% compared with MTF and by about 2.2% compared with RP. GAF technology effectively avoids laborious and complex processes such as manual feature extraction and multifeature…”
InfluentialDebnath, T., Siddiky, M. N. A., Rahman, M. E., Das, P., Guha, A. K., Rahman, M. R., & Kabir, H. M. D. (2026) · Computer Science Review · doi:10.1016/j.cosrev.2026.100979 · Cited by 4
Aiyappa, R., Senthilmani, S., An, J., Kwak, H., & Ahn, Y. (2026) · PeerJ Computer Science · doi:10.7717/peerj-cs.3540 · Cited by 15
Jung, S., & Nah, K. (2026) · Information · doi:10.3390/info17030243 · Cited by 2
Evaluating and improving the robustness of security attack detectors generated by LLMs
Pasini, S., Kim, J.-H., Aiello, T., Lozoya, R. C., Sabetta, A., & Tonella, P. (2025) · Empirical Software Engineering · doi:10.1007/s10664-025-10743-w · Cited by 8
LUMEN: A Systems Approach to LLM-Guided Activation of Hidden Behaviors in Malware
Valakuzhy, K., Wong, M. Y., Blough, D. M., Ahamad, M., & Monrose, F. (2026) · International Conference on Dependable Systems & Networks/Proceedings · doi:10.1109/dsn69566.2026.00026 · Cited by 2
“As prior successes using off-the-shelf LLMs for malware-related tasks have shown, selecting the appropriate text to provide the LLM in a prompt, referred to as context, is especially important because LLMs perform poorly when given too little [7] or too much [22], [23] information.”
Principled Context Engineering for RAG: Statistical Guarantees via Conformal Prediction
Chakraborty, D., Yang, E., Khashabi, D., Lawrie, D., & Duh, K. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-21300-6_45 · Cited by 9
“The lost-in-the-middle effect [17] shows that LLMs attend poorly to mid-prompt evidence, limiting effective use of long-context capacity to 10–20% [9, 8].”
Scientific knowledge graph and ontology generation using open large language models
Bran, A. M., Hart, M., Bran, A. C. M. M., Lederbauer, M., & Schwaller, P. (2026) · Digital Discovery · doi:10.1039/d5dd00275c · Cited by 3
“As LCLLMs are known to be sensitive to input order 31 , we generate multiple answers, each with a random shuffle in the order of the papers.”
Pinna, M., Galletta, S., Ash, E., & Elmer, T. (2026) · JMIR Formative Research · doi:10.2196/85427 · Cited by 2
“LLMs are known to exhibit attention issues when handling long contexts, often assigning importance based on text location rather than relevance [26,27].”
Deep-shallow semantic fusion inference based on graph neural network
Wu, Z., xiao, S., & gao, J. (2025) · Information Fusion · doi:10.1016/j.inffus.2025.103849 · Cited by 3
Unsupervised Context-Linking Retriever for Question Answering on Long Narrative Books
Ateeq, M. A., Tiun, S. B., Abdelhaq, H., & Kaur, W. (2025) · IEEE Access · doi:10.1109/access.2025.3578497 · Cited by 1
“A study by [17] evaluated the performance of language models on tasks requiring the identification of relevant information within long input contexts.”
MMM-RAG: A Multi-Agent Multi-Feature Method for Multimodal Retrieval-Augmented Generation
Chang, T., Fang, Y., & Zhang, Y. (2025) · IEEE International Conference on Parallel and Distributed Systems (ICPADS) · doi:10.1109/icpads67057.2025.11323162 · Cited by 1
“Retrieval-Augmented Generation (RAG) has achieved significant breakthroughs in enhancing large language models by integrating external knowledge, with early systems focusing primarily on textual modality [17].”
CHASE: LLM Agents for Dissecting Malicious PyPI Packages
Toda, T., & Mori, T. (2025) · ACM International Conference on AI-Powered Software (AIware) · doi:10.1109/aiware69974.2025.00008 · Cited by 2
“1) State-Based Communication for Context Integrity: Conventional multi-turn conversations that accumulate message history are susceptible to context bloat and the “Lost in the Middle” problem [32].”
Kim, Y., Kim, M., Kim, S., Cho, S., & Kim, J. (2026) · CHI Conference on Human Factors in Computing Systems · doi:10.1145/3772318.3790971
Evaluating the utility of large language models for detecting and simulating language dysfunction
Cong, Y., Lee, J., Rajput, N., Tumacder, E., & LaCroix, A. N. (2026) · Frontiers in Artificial Intelligence · doi:10.3389/frai.2026.1857106 · Cited by 1
“From a computational perspective (Liu et al., 2024; Touvron et al., 2023; Grattafiori et al., 2024), model performance tends to be strongest when relevant information appears at the beginning or end of the input, but declines substantially when key details are embedded in the middle of longer…”
Metadata
- Authors
- Liu, N. F., Lin, K., Hewitt, J., Paranjape, A., Bevilacqua, M., Petroni, F., & Liang, P.
- Affiliations
- Stanford University (US)
- University of California, Berkeley (US)
- Samaya AI, UK
- Corresponding author
- Liu, Nelson F.
- Venue
- Transactions of the Association for Computational Linguistics
- Publisher
- MIT Press
- Published
- 2024
- Volume
- 12
- Pages
- 157-173
- ISSN
- 2307-387X (ISSN-L)
- OpenAlex
- W4391876619
- Semantic Scholar
- 1733eb7792f7a43dd21f51f4d1017a1bffd217b5
- Cited by
- 1,612
- References
- 59
- Citations by year
- 2026: 901 · 2025: 516 · 2024: 172
- FWCI
- 233.95
- Type
- journal article
- Language
- en
- Open access
- diamond
- doi.org
- Licence
- Locations
- Transactions of the Association for Computational Linguistics (publishedVersion, cc-by)
- arXiv (Cornell University) (submittedVersion)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- arXiv (Cornell University)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Multimodal Machine Learning Applications
- Keywords
- long-context language models, multi-document question answering, position bias
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-09-27 (prompt extract-v2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.