CAIM: Development and Evaluation of a Cognitive AI Memory Framework for Long-Term Interaction with Intelligent Agents

Frederik Berenz; Rebecca Westh\"au{\ss}er; Sebastian Zepf; Wolfgang Minker

REVIEW 3 major objections 5 minor 56 references

CAIM: Development and Evaluation of a Cognitive AI Memory Framework for Long-Term Interaction with Intelligent Agents

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read CAIM, a prompt-based memory layer, is claimed to raise LLM response correctness on long-term interaction questions without retraining, through ontology-tagged retrieval and a decision controller.

desk verdict Useful framework and honest ablation, but the headline comparison to prior work is uncontrolled; re-run baselines before the 'outperforms' claim is believable. read the letter →

arxiv 2505.13044 v1 pith:AZY3I7UT submitted 2025-05-19 cs.AI cs.HC

Rebecca Westhäu{ss}er , Frederik Berenz , Wolfgang Minker , Sebastian Zepf This is my paper · ORCID

classification cs.AIcs.HC

keywords largelanguagemodelslong-termmemorycognitiveAImemory-augmentedgenerationontology-basedtaggingretrievalaccuracyresponsecorrectness

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 tries to establish that a plug-in memory layer called CAIM can give large language models a durable, context-aware memory without retraining or changing their weights. CAIM routes each user turn through a memory controller that decides whether past information is needed, retrieves only memories tagged with the right ontology labels and timestamps, and then rewrites the session into concise stored thoughts. On 100 fact-checking questions from a public virtual-assistant dataset, the paper reports that CAIM raises response correctness and contextual coherence over two existing memory frameworks across three language models. If the result holds, practical long-term personalization becomes available for API-only models, because the memory system works entirely through prompts.

What carries the argument

The central object is the ontology-based tagging system: a three-level vocabulary of single-word tags (categories, subcategories, attributes) that the LLM is prompted to choose from both when storing memories and when querying them. A Python function retrieves long-term entries only when their stored tags and, when present, timestamps literally match the tags picked for the user input. Around this matching core, the Memory Controller decides whether retrieval is needed at all, the Memory Retrieval module filters by tag and date and then applies a second relevance pass, and Post-Thinking converts session turns into concise inductive thoughts with up to three tags and merges duplicate entries.

What would settle it

An independent replication could change only the ontology's tag vocabulary on the same 100-question benchmark: if response correctness stays high with a different three-level taxonomy, the gains come from the filtering and controller mechanics, whereas if it collapses, the specific tag set is the load-bearing element.

Watch

Extended reading notes

Core claim

The paper's central claim is that a memory framework called CAIM gives an otherwise unchanged language model a long-term memory that is durable and context-aware, and that this improves response correctness in long-term interaction. The framework separates short-term from long-term memory, uses the LLM itself as a decision unit that asks whether history is needed, retrieves stored 'inductive thoughts' by matching tags from a three-level ontology together with optional timestamps, and after each session turns the conversation into concise tagged summaries with duplicates merged. Evaluated on 100 fact-checking questions from a public virtual-assistant dataset, the paper reports that CAIM outperforms two existing memory frameworks on response correctness and contextual coherence for all three models tested, and on retrieval accuracy for the two GPT models. The paper also reports that removing the relevance filter cuts response correctness from 87.5% to 63.5% with GPT-4o, identifying contextual and time-based filtering as the main source of the gain.

Load-bearing premise

CAIM's tag-based retrieval works only if the underlying language model reliably picks the restricted single-word labels from the ontology and returns them in the exact comma-separated form the Python matcher expects; when a model does not, as reported for one of the three tested models, retrieval accuracy drops below both baselines.

Editorial extensions

If this is right

API-only language models can gain long-term memory and personalization by adding CAIM as a prompt-driven layer, because no model weights are changed.
The contextual and time-based relevance filter is a main source of the gain: removing it drops response correctness from 87.5% to 63.5% in the GPT-4o ablation.
The Memory Controller also matters: always retrieving history instead of letting the controller decide drops correctness from 87.5% to 78.2%.
Storing concise inductive thoughts rather than detailed sentences is associated with higher performance, so memory representation choices affect quality independently of retrieval.
CAIM's benefit is bounded by the underlying model's instruction-following ability; the paper reports the weakest retrieval accuracy for a model that does not return tags in the required format.

Reading between the lines

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

Replacing literal tag matching with fuzzy or embedding-based matching would likely make CAIM robust to models that refuse to follow the comma-separated output format, since the paper's own results trace low retrieval accuracy to exactly that failure.
Adding explicit links between related inductive thoughts (for example, shared entities or conversation ids) is a natural extension pointed at by the paper's 'split information' limitation, and could be tested by measuring recall on questions that require combining two separately stored facts.
The storage statistics hint at a granularity trade-off that could be tested directly: a memory controller that adapts the level of detail stored per query type would probably improve the recipe/list and summarization failures the paper reports.
A separate generator-level test—forcing the model to ground its answer in retrieved memories before answering—could isolate the 'know but don't tell' failures from the retrieval failures and would be a clean follow-up experiment.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Colleague, quick take on CAIM. The new thing here is not any single mechanism but a specific combination: an ontology-based tag selection for storage and retrieval, a memory controller that decides when to fetch history, and a post-thinking merge step. That combination is new, the ablations are informative, and the authors are unusually candid about failure modes—especially that ChatGLM refuses to emit tags in the required format, which breaks exact-match retrieval. Credit where due: the appendix includes full prompts, the GVD dataset is public, and the ablation on GPT-4o shows that removing the relevance filter drags response correctness down from 87.5% to 63.5%. That is evidence the mechanism does something.

The soft spot is the one the stress-test flagged, and it is load-bearing. Table 3 compares CAIM against MemoryBank and TiM numbers reported in the original papers, not re-run in this study. Those baselines were evaluated with specific LLMs and annotation procedures; we don't know whether the comparison holds if you fix the LLM context. The paper's own data reveal the problem: for ChatGLM, CAIM's retrieval accuracy is 67.6% versus 80.9% for MemoryBank and 82% for TiM, yet CAIM's response correctness is 60% versus their 43.8% and 45%. That inversion is suspicious enough that you cannot read 'outperforms' as established. It could still be true—the relevance filter may compensate for poor retrieval—but the uncontrolled baseline comparison plus single runs and no error bars means the central claim is not yet demonstrated. The ChatGLM tag-format issue is genuinely a limitation of exact-match retrieval, and the authors acknowledge it in 4.2.1, but it is secondary next to the baseline problem.

Who is this for? People building memory layers for LLM assistants will find the design and the ablation useful. Researchers doing head-to-head benchmarks should treat the numbers with caution until the baselines are re-run. I would not cite the performance comparisons in my own work yet, but I would send this to a serious referee: the framework is coherent, the evaluation question is important, and the flaws are addressable with controlled re-runs and variance reporting.

Referee Report

3 major / 5 minor

Summary. The paper proposes CAIM, a cognitive-AI-inspired memory framework for large language models (LLMs) that combines three modules: a Memory Controller (a self-asking decision unit), a Memory Retrieval module (tag- and timestamp-based filtering with relevance scoring), and a Post-Thinking module (memory extension and review via inductive thoughts). The framework is evaluated on the public Generated Virtual Dataset (GVD) using three underlying LLMs (ChatGLM, GPT-3.5 turbo, GPT-4o) against MemoryBank and Think-in-Memory on metrics of retrieval accuracy, response correctness, contextual coherence, and a newly introduced memory-storage metric. The paper reports that CAIM outperforms the baselines on most metric-model combinations, and an ablation study with GPT-4o indicates that both the Memory Controller and the relevance-filtering mechanism contribute to response correctness.

Significance. If the empirical claims are upheld, CAIM would be a practical, training-free memory layer that improves long-term interaction quality across multiple LLMs, and the detailed prompt design in the appendix would support replication. The paper also provides a useful ablation study showing that its components matter, and it reports inter-rater agreement, which strengthens the annotation methodology. However, the central claim of outperforming baselines is currently not well supported because the baseline numbers are taken from prior publications rather than re-run under identical conditions, and because the evaluation consists of single runs without variance or significance measures. The framework itself is well-motivated and the design is clearly presented, so the issues are addressable in revision.

major comments (3)

[Section 4.1.1, Table 3] This is the central issue. Without re-run baselines, the empirical comparison is uncontrolled.
[Section 4.2.1, Table 3, ChatGLM row] This is load-bearing because it directly contradicts the paper's summary claim.
[Section 4.1.4, Tables 2 and 3] This is load-bearing for the quantitative comparative claims.

minor comments (5)

[Section 4.2.2, Table 5] The text states that removing the Memory Controller gives a 'similar Retrieval Accuracy of 85.5%' with the full system at '87.5%', but Table 5 reports 85.7% and 88.7%, respectively. Please reconcile these numbers.
[Table 2] The ICC values for Contextual Coherence are 5.8% (GPT-3.5 turbo) and -0.6% (GPT-4o). The explanation that low rating variability leads to misleadingly low ICC is plausible, but the paper should report the actual rating distributions or use an alternative agreement measure (e.g., weighted kappa) to support the claim of 'strong level of inter-rater agreement'.
[Section 4.1.1] The model description says 'GPT [6]' which is ambiguous; please specify the exact versions used (GPT-3.5 turbo and GPT-4o) and note the access dates.
[Abstract and Section 4.2.1] The claim that CAIM 'outperforms baseline frameworks across different metrics' is too broad given the ChatGLM retrieval-accuracy result. Please add a qualifier such as 'on most metric-model combinations' or explicitly discuss the exception in the abstract.
[Section 4.2.3] The qualitative findings are informative, but they would be strengthened by a quantitative breakdown of failure types per model (e.g., counts of rhetorical-question errors, list-recall errors) so that the reader can connect these observations to the aggregate metrics.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: CAIM is an empirical system evaluation with no fitted parameter renamed as a prediction.

full rationale

CAIM is an empirical systems paper, not a derivation; there is no chain in which an output is constructed from its own premise. The ontology, tagging, retrieval, and post-thinking modules are described procedurally and then evaluated on the public GVD with human annotation against external baselines. No parameter is fitted to the test data and later reported as a prediction: the new Memory Storage metric is a measured outcome, not a tuning target, and the retrieval/response metrics are independently annotated. The baseline comparison against published MemoryBank and TiM numbers may raise uncontrolled-comparison or soundness concerns, but that is an evaluation-validity issue rather than circularity. There is also no load-bearing self-citation chain and no ansatz smuggled in via citation: CAIM explicitly builds on prior frameworks but its contribution is a concrete implementation and benchmark, not a result that reduces by definition to those prior works. No circular step can be exhibited, so the circularity score is 0.

Assumptions & free parameters 3 free parameters · 4 assumptions · 1 invented entities

The framework relies on engineered design choices rather than fitted numerical parameters. The listed free parameters are discrete hand-set choices (tag count, ontology depth, exact-match policy) that are not tuned or ablated. The axioms are domain assumptions about the validity of ontology-based retrieval, the GVD benchmark, human annotation, and LLM instruction-following.

free parameters (3)

Tag count limit = 3
Prompts instruct the LLM to select 'two or three' tags per memory/query; this bound shapes retrieval precision and was not tuned.
Ontology depth = 3 levels
The ontology is limited to categories, subcategories, attributes; a hand-chosen complexity limit without an ablation.
Tag matching policy = exact string match
Python functions match tags/timestamps literally; no fuzzy matching or similarity threshold is used.

assumptions (4)

domain assumption Ontology-based tagging improves retrieval consistency and accuracy compared with semantic similarity.
Section 3.2 states the tagging system aims to reduce tag variability, but no direct comparison against semantic similarity is provided; GLM results in Section 4.2.1 show the assumption can fail.
domain assumption The Generated Virtual Dataset is a valid benchmark for long-term memory in LLMs.
Section 4.1.1 adopts GVD because MemoryBank and TiM used it, assuming the 100 questions and 15 users capture long-term interaction needs.
domain assumption Human annotation of retrieval, correctness, coherence, and storage is reliable despite low ICC values.
Section 4.1.4 argues low ICC is due to low rating variability; the metric remains a subjective judgment.
domain assumption The underlying LLM will follow structured prompts and output tags in the required format.
The framework's retrieval depends on literal tag matching; Section 4.2.1 documents ChatGLM generating tags in Chinese and failing to follow the format.

invented entities (1)

LLM-generated hierarchical ontology of single-word tags
purpose: Provides a controlled vocabulary for tagging memories and queries so that retrieval can use exact tag matching.
The ontology is not released or validated independently; it is generated per run by the LLM and the paper provides no external handle to check its quality.

how reviews work

0 comments

Cite this review

Pith. "Pith review of CAIM: Development and Evaluation of a Cognitive AI Memory Framework for Long-Term Interaction with Intelligent Agents." pith.science (2026). https://pith.science/paper/AZY3I7UT

@misc{pith2026250513044,
  author       = {Pith},
  title        = {Pith review of: CAIM: Development and Evaluation of a Cognitive AI Memory Framework for Long-Term Interaction with Intelligent Agents},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AZY3I7UT}},
  note         = {Machine review of arXiv:2505.13044}
}

read the original abstract

Large language models (LLMs) have advanced the field of artificial intelligence (AI) and are a powerful enabler for interactive systems. However, they still face challenges in long-term interactions that require adaptation towards the user as well as contextual knowledge and understanding of the ever-changing environment. To overcome these challenges, holistic memory modeling is required to efficiently retrieve and store relevant information across interaction sessions for suitable responses. Cognitive AI, which aims to simulate the human thought process in a computerized model, highlights interesting aspects, such as thoughts, memory mechanisms, and decision-making, that can contribute towards improved memory modeling for LLMs. Inspired by these cognitive AI principles, we propose our memory framework CAIM. CAIM consists of three modules: 1.) The Memory Controller as the central decision unit; 2.) the Memory Retrieval, which filters relevant data for interaction upon request; and 3.) the Post-Thinking, which maintains the memory storage. We compare CAIM against existing approaches, focusing on metrics such as retrieval accuracy, response correctness, contextual coherence, and memory storage. The results demonstrate that CAIM outperforms baseline frameworks across different metrics, highlighting its context-awareness and potential to improve long-term human-AI interactions.

Figures

Figures reproduced from arXiv: 2505.13044 by the authors.

**Figure 1.** Example of CAIM’s ontology structure For example, the prompt might ask whether the response requires recalling historical information from earlier in the conversation, or if the current context alone is sufficient. The agent always selects between two options provided in the prompt. Based on these evaluations, the agent selects one of the following options: (1) Is additional information necessary? (2) Are both hist… view at source ↗

**Figure 2.** Workflow of CAIM - Illustration of the three modules, with colors highlighting the different possible paths within [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

56 extracted references · 22 canonical work pages

[1]

Richard C Atkinson and Richard M Shiffrin. 1968. Human memory: A proposed system and its control processes. In Psychology of learning and motivation . Vol. 2. Elsevier, 89–195

1968
[2]

Alan Baddeley. 2020. Working memory. Memory (2020), 71–111

work page 2020
[3]

Sanghwan Bae, Donghyun Kwak, Soyoung Kang, Min Young Lee, Sungdong Kim, Yuin Jeong, Hyeri Kim, Sang-Woo Lee, Woomyoung Park, and Nako Sung
[4]

Shweta Bajpai, Ram Bajpai, and Himanshu Chaturvedi. 2015. Evaluation of inter- rater agreement and inter-rater reliability for observational data: an overview of concepts and methods. Journal of the Indian Academy of Applied Psychology 41, 3 (2015), 20–27

work page 2015
[5]

Matthias Baldauf, Schahram Dustdar, and Florian Rosenberg. 2007. A survey on context-aware systems. International Journal of ad Hoc and ubiquitous Computing 2, 4 (2007), 263–277

work page 2007
[6]

Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. 2020. Language models are few-shot learners. Advances in neural information processing systems 33 (2020), 1877–1901

2020
[7]

Jin Chen, Zheng Liu, Xu Huang, Chenwang Wu, Qi Liu, Gangwei Jiang, Yuanhao Pu, Yuxuan Lei, Xiaolong Chen, Xingmei Wang, et al. 2024. When large language models meet personalization: Perspectives of challenges and opportunities.World Wide Web 27, 4 (2024), 42

2024
[8]

Min Chen, Francisco Herrera, and Kai Hwang. 2018. Cognitive computing: architecture, technologies and intelligent applications.Ieee Access 6 (2018), 19774– 19783

work page 2018

Show all 56 references

[9]

Weihao Chen, Chun Yu, Huadong Wang, Zheng Wang, Lichen Yang, Yukun Wang, Weinan Shi, and Yuanchun Shi. 2023. From Gap to Synergy: Enhancing Contextual Understanding through Human-Machine Collaboration in Personalized Systems. In Proceedings of the 36th Annual ACM Symposium on ...

2023
[10]

Can Cui, Yunsheng Ma, Xu Cao, Wenqian Ye, Yang Zhou, Kaizhao Liang, Jintai Chen, Juanwu Lu, Zichong Yang, Kuei-Da Liao, et al . 2024. A survey on mul- timodal large language models for autonomous driving. In Proceedings of the IEEE/CVF Winter Conference on Applications of Comp...

2024
[11]

Hung Du, Srikanth Thudumu, Rajesh Vasa, and Kon Mouzakis. 2024. A Survey on Context-Aware Multi-Agent Systems: Techniques, Challenges and Future Directions. arXiv preprint arXiv:2402.01968 (2024)

2024 arXiv
[12]

Joel Eapen and VS Adhithyan. 2023. Personalization and customization of llm responses. International Journal of Research Publication and Reviews 4, 12 (2023), 2617–2627

2023
[13]

Natasa Gisev, J Simon Bell, and Timothy F Chen. 2013. Interrater agreement and interrater reliability: key concepts, approaches, and applications. Research in Social and Administrative Pharmacy 9, 3 (2013), 330–338

2013
[14]

Team GLM, Aohan Zeng, Bin Xu, Bowen Wang, Chenhui Zhang, Da Yin, Dan Zhang, Diego Rojas, Guanyu Feng, Hanlin Zhao, et al. 2024. Chatglm: A fam- ily of large language models from glm-130b to glm-4 all tools. arXiv preprint arXiv:2406.12793 (2024)

2024 arXiv
[15]

Nicola Guarino, Daniel Oberle, and Steffen Staab. 2009. What is an ontology? Handbook on ontologies (2009), 1–17

2009
[16]

Kostas Hatalis, Despina Christou, Joshua Myers, Steven Jones, Keith Lambert, Adam Amos-Binks, Zohreh Dannenhauer, and Dustin Dannenhauer. 2023. Mem- ory Matters: The Need to Improve Long-Term Memory in LLM-Agents. InPro- ceedings of the AAAI Symposium Series , Vol. 2. 277–280

2023
[17]

Aaron Hurst, Adam Lerer, Adam P Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, et al. 2024. Gpt-4o system card. arXiv preprint arXiv:2410.21276 (2024)

2024 arXiv
[18]

Kai Hwang and Min Chen. 2017. Big-data analytics for cloud, IoT and cognitive computing. John Wiley & Sons

2017
[19]

Johannes Kirmayr, Lukas Stappen, Phillip Schneider, Florian Matthes, and Elis- abeth André. 2025. CarMem: Enhancing Long-Term Memory in LLM Voice Assistants through Category-Bounding. arXiv preprint arXiv:2501.09645 (2025)

2025 arXiv
[20]

Gibbeum Lee, Volker Hartmann, Jongho Park, Dimitris Papailiopoulos, and Kang- wook Lee. 2023. Prompted llms as chatbot modules for long open-domain con- versation. arXiv preprint arXiv:2305.04533 (2023)

2023 arXiv
[21]

Min Kyung Lee, Jodi Forlizzi, Sara Kiesler, Paul Rybski, John Antanitis, and Sarun Savetsila. 2012. Personalization in HRI: A longitudinal field experiment. In Proceedings of the seventh annual ACM/IEEE international conference on Human- Robot Interaction. 319–326

2012
[22]

Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, et al. 2020. Retrieval-augmented generation for knowledge-intensive nlp tasks. Advances in Neural Information Processing S...

2020
[23]

Hao Li, Chenghao Yang, An Zhang, Yang Deng, Xiang Wang, and Tat-Seng Chua
[24]

Yinheng Li. 2023. A practical survey on zero-shot prompt design for in-context learning. arXiv preprint arXiv:2309.13205 (2023)

2023 arXiv
[25]

Mike EU Ligthart, Mark A Neerincx, and Koen V Hindriks. 2022. Memory- based personalization for fostering a long-term child-robot relationship. In 2022 17th ACM/IEEE International Conference on Human-Robot Interaction (HRI) . IEEE, 80–89

2022
[26]

Lei Liu, Xiaoyan Yang, Yue Shen, Binbin Hu, Zhiqiang Zhang, Jinjie Gu, and Guannan Zhang. 2023. Think-in-memory: Recalling and post-thinking enable llms with long-term memory. arXiv preprint arXiv:2311.08719 (2023)

2023 arXiv
[27]

Michael Xieyang Liu, Aniket Kittur, and Brad A Myers. 2021. To reuse or not to reuse? A framework and system for evaluating summarized knowledge. Proceed- ings of the ACM on Human-Computer Interaction 5, CSCW1 (2021), 1–35

2021
[28]

Taiming Lu, Muhan Gao, Kuai Yu, Adam Byerly, and Daniel Khashabi. 2024. Insights into LLM Long-Context Failures: When Transformers Know but Don’t Tell. arXiv preprint arXiv:2406.14673 (2024)

2024 arXiv
[29]

Kenneth O McGraw and Seok P Wong. 1996. Forming inferences about some intraclass correlation coefficients. Psychological methods 1, 1 (1996), 30

1996
[30]

Lin Ning, Luyang Liu, Jiaxing Wu, Neo Wu, Devora Berlowitz, Sushant Prakash, Bradley Green, Shawn O’Banion, and Jun Xie. 2024. User-LLM: Efficient LLM Contextualization with User Embeddings. arXiv preprint arXiv:2402.13598 (2024)

2024 arXiv
[31]

Qian Niu, Junyu Liu, Ziqian Bi, Pohsun Feng, Benji Peng, Keyu Chen, Ming Li, Lawrence KQ Yan, Yichao Zhang, Caitlyn Heqi Yin, et al. 2024. Large language models and cognitive science: A comprehensive review of similarities, differences, and challenges. arXiv preprint arXiv:240...

2024
[32]

OpenAI. 2024. Memory and New Controls for ChatGPT. https://openai.com/ index/memory-and-new-controls-for-chatgpt/ Accessed: 2025-04-07

2024
[33]

Charles Packer, Sarah Wooders, Kevin Lin, Vivian Fang, Shishir G Patil, Ion Stoica, and Joseph E Gonzalez. 2023. Memgpt: Towards llms as operating systems. arXiv preprint arXiv:2310.08560 (2023)

2023 arXiv
[34]

Joon Sung Park, Joseph C O’Brien, Carrie J Cai, Meredith Ringel Morris, Percy Liang, Michael S Bernstein, et al. 2023. Generative agents: Interactive simulacra of human behavior. arXiv. Org (2023, April 7) https://arxiv. org/abs/2304.03442 v2 (2023)

2023 arXiv
[35]

Saurav Pawar, SM Tonmoy, SM Zaman, Vinija Jain, Aman Chadha, and Amitava Das. 2024. The What, Why, and How of Context Length Extension Techniques in Large Language Models–A Detailed Survey. arXiv preprint arXiv:2401.07872 (2024)

2024 arXiv
[36]

Ori Ram, Yoav Levine, Itay Dalmedigos, Dor Muhlgay, Amnon Shashua, Kevin Leyton-Brown, and Yoav Shoham. 2023. In-context retrieval-augmented language models. Transactions of the Association for Computational Linguistics 11 (2023), 1316–1331

2023
[37]

Alireza Salemi, Surya Kallumadi, and Hamed Zamani. 2024. Optimization meth- ods for personalizing large language models through retrieval augmentation. In Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval . 752–762

2024
[38]

Seongbum Seo, Sangbong Yoo, and Yun Jang. 2025. A Prompt Chaining Frame- work for Long-Term Recall in LLM-Powered Intelligent Assistant. In Proceedings of the 30th International Conference on Intelligent User Interfaces . 89–105

2025
[39]

Theodore R Sumers, Shunyu Yao, Karthik Narasimhan, and Thomas L Griffiths
[40]

Ron Sun. 2024. Can A Cognitive Architecture Fundamentally Enhance LLMs? Or Vice Versa? arXiv preprint arXiv:2401.10444 (2024)

2024 arXiv
[41]

Gemini Team, Petko Georgiev, Ving Ian Lei, Ryan Burnell, Libin Bai, Anmol Gulati, Garrett Tanzer, Damien Vincent, Zhufeng Pan, Shibo Wang, et al. 2024. Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context. arXiv preprint arXiv:2403.05530 (2024)

2024 arXiv
[42]

Bing Wang, Xinnian Liang, Jian Yang, Hui Huang, Shuangzhi Wu, Peihao Wu, Lu Lu, Zejun Ma, and Zhoujun Li. 2023. Enhancing large language model with self-controlled memory framework. arXiv preprint arXiv:2304.13343 (2023)

2023 arXiv
[43]

Justin D Weisz, Jessica He, Michael Muller, Gabriela Hoefer, Rachel Miles, and Werner Geyer. 2024. Design principles for generative AI applications. In Pro- ceedings of the 2024 CHI Conference on Human Factors in Computing Systems . 1–22

2024
[44]

Stanisław Woźniak, Bartłomiej Koptyra, Arkadiusz Janz, Przemysław Kazienko, and Jan Kocoń. 2024. Personalized large language models. arXiv preprint arXiv:2402.09269 (2024)

2024 arXiv
[45]

Bin Wu, Zhengyan Shi, Hossein A Rahmani, Varsha Ramineni, and Emine Yilmaz
[46]

Yuanzhen Xie, Tao Xie, Mingxiong Lin, WenTao Wei, Chenglin Li, Beibei Kong, Lei Chen, Chengxiang Zhuo, Bo Hu, and Zang Li. 2023. Olagpt: Empowering llms with human-like problem-solving abilities. arXiv preprint arXiv:2305.16334 (2023)

2023 arXiv
[47]

J Xu. 2021. Beyond goldfish memory: Long-term open-domain conversation. arXiv preprint arXiv:2107.07567 (2021)

2021 arXiv
[48]

Xinchao Xu, Zhibin Gou, Wenquan Wu, Zheng-Yu Niu, Hua Wu, Haifeng Wang, and Shihang Wang. 2022. Long time no see! open-domain conversation with long-term persona memory. arXiv preprint arXiv:2203.05797 (2022)

2022 arXiv
[49]

arXiv preprint arXiv:2406.17803 (2024)

Understanding the Role of User Profile in the Personalization of Large Language Models. arXiv preprint arXiv:2406.17803 (2024)

2024 arXiv
[50]

Kai Zhang, Fubang Zhao, Yangyang Kang, and Xiaozhong Liu. 2023. Memory- augmented llm personalization with short-and long-term memory coordination. arXiv preprint arXiv:2309.11696 (2023)

2023 arXiv
[51]

Guoying Zhao, Yante Li, and Qianru Xu. 2022. From emotion AI to cognitive AI. International Journal of Network Dynamics and Intelligence (2022), 65–72

2022
[52]

Prompt Retrieval and Conversation:

Wanjun Zhong, Lianghong Guo, Qiqi Gao, He Ye, and Yanlin Wang. 2024. Memo- rybank: Enhancing large language models with long-term memory. InProceedings of the AAAI Conference on Artificial Intelligence , Vol. 38. 19724–19731. A PROMPT LIST A.1 Memory Controller Prompt Informat...

2024
[53]

Ryan Yen and Jian Zhao. 2024. Memolet: Reifying the Reuse of User-AI Conver- sational Memories. In Proceedings of the 37th Annual ACM Symposium on User Interface Software and Technology. 1–22

2024
[2022]

arXiv preprint arXiv:2210.08750 (2022)

Keep me updated! memory management in long-term conversations. arXiv preprint arXiv:2210.08750 (2022)

2022 arXiv
[2023]

Cognitive architectures for language agents.arXiv preprint arXiv:2309.02427 (2023)

2023 arXiv
[2024]

arXiv preprint arXiv:2406.05925 (2024)

Hello Again! LLM-powered Personalized Agent for Long-term Dialogue. arXiv preprint arXiv:2406.05925 (2024)

2024 arXiv

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.

[1] [1]

Richard C Atkinson and Richard M Shiffrin. 1968. Human memory: A proposed system and its control processes. In Psychology of learning and motivation . Vol. 2. Elsevier, 89–195

1968

[2] [2]

Alan Baddeley. 2020. Working memory. Memory (2020), 71–111

work page 2020

[3] [3]

Sanghwan Bae, Donghyun Kwak, Soyoung Kang, Min Young Lee, Sungdong Kim, Yuin Jeong, Hyeri Kim, Sang-Woo Lee, Woomyoung Park, and Nako Sung

[4] [4]

Shweta Bajpai, Ram Bajpai, and Himanshu Chaturvedi. 2015. Evaluation of inter- rater agreement and inter-rater reliability for observational data: an overview of concepts and methods. Journal of the Indian Academy of Applied Psychology 41, 3 (2015), 20–27

work page 2015

[5] [5]

Matthias Baldauf, Schahram Dustdar, and Florian Rosenberg. 2007. A survey on context-aware systems. International Journal of ad Hoc and ubiquitous Computing 2, 4 (2007), 263–277

work page 2007

[6] [6]

Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. 2020. Language models are few-shot learners. Advances in neural information processing systems 33 (2020), 1877–1901

2020

[7] [7]

Jin Chen, Zheng Liu, Xu Huang, Chenwang Wu, Qi Liu, Gangwei Jiang, Yuanhao Pu, Yuxuan Lei, Xiaolong Chen, Xingmei Wang, et al. 2024. When large language models meet personalization: Perspectives of challenges and opportunities.World Wide Web 27, 4 (2024), 42

2024

[8] [8]

Min Chen, Francisco Herrera, and Kai Hwang. 2018. Cognitive computing: architecture, technologies and intelligent applications.Ieee Access 6 (2018), 19774– 19783

work page 2018