Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

RemoteRAG: A Privacy-Preserving LLM Cloud RAG Service

T0 review · 3 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read RemoteRAG claims that a cloud RAG service can answer a user's query with bounded privacy leakage and no retrieval loss by sending a perturbed embedding and searching a small, provably sufficient candidate set.

desk verdict Real problem, clever system, but the Direct branch's privacy proof does not survive contact with the index-set leak. read the letter →

arxiv 2412.12775 v1 pith:V7FKHI6I submitted 2024-12-17 cs.IR cs.CR

Yihang Cheng , Lan Zhang , Junyang Wang , Mu Yuan , Yunhao Yao This is my paper · ORCID
classification cs.IRcs.CR
keywords privacy-preservingRAG(nepsilon)-DistanceDPembeddingperturbationdifferentialprivacyretrieval-augmentedgenerationoblivioustransferhomomorphicencryptionsemanticleakage
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 user can query a cloud retrieval-augmented generation service without exposing the semantic content of the query, and without sacrificing retrieval accuracy. The proposed protocol, RemoteRAG, sends the cloud a perturbed version of the query embedding, guided by a new privacy definition called $(n,\epsilon)$-DistanceDP, and restricts the search to a small set of $k'$ documents whose size is derived from spherical geometry. The paper proves that under its uniform-sphere model this candidate set always contains the true top-$k$ documents, and reports 100% recall in experiments spanning $10^4$ to $10^6$ documents, several values of $k$, several perturbation magnitudes, and five embedding models. It also reports resistance to a known embedding-inversion attack and a per-query cost of 0.67 seconds and 46.66 KB for a $10^6$-document collection, compared with 2.72 hours and 1.43 GB for the fully cryptographic baseline. If correct, this gives a practical middle path between sending queries in plaintext and encrypting the whole document store.

What carries the argument

The load-bearing object is $(n,\epsilon)$-DistanceDP together with the spherical-cap calculus of Lemma 1 and Theorem 1. The mechanism adds radial noise $r \sim \mathrm{Gamma}(n, 1/\epsilon)$ with a uniformly random direction to the query embedding, giving an average displacement $n/\epsilon$, and the cloud searches only the $k'$ nearest neighbors of the resulting perturbed point. Lemma 1 relates the count $k$ to the polar angle $\alpha_k$ under the assumption that $N$ embeddings are uniformly distributed on the unit $n$-sphere, and Theorem 1 computes the extra count $\Delta k = k' - k$ needed to absorb the perturbation angle $\Delta\alpha_k$; this identity is what carries the lossless-retrieval guarantee. Encrypted cosine-distance computation (partially homomorphic encryption) and $k$-out-of-$k'$ oblivious transfer carry the final selection of the true top-$k$ documents.

What would settle it

Build a document collection whose embeddings form tight, well-separated topical clusters, compute $k'$ from Theorem 1 as if the embeddings were uniform on the sphere, then check whether the $k'$ nearest neighbors of a perturbed query embedding still contain the true top-$k$ for queries drawn from one cluster; one missing document refutes the guarantee as stated. A second, simpler check is to run the paper's recall experiment on such a clustered collection and look for recall below 100%.

Watch

Extended reading notes

Core claim

RemoteRAG's central claim is that privacy and lossless retrieval are compatible in cloud RAG: perturb the query embedding under $(n,\epsilon)$-DistanceDP, expand the search radius according to Theorem 1, and the $k'$ documents nearest the perturbed embedding will contain the top-$k$ documents nearest the original query. The perturbation draws a radial distance from a Gamma$(n, 1/\epsilon)$ distribution and a uniformly random direction, so on average it moves $n/\epsilon$ in embedding space. Lemma 1 converts a desired number $k$ of documents into a polar angle $\alpha_k$ on the unit $n$-sphere, and Theorem 1 sets $k' - k$ equal to the surface area between $\alpha_k$ and $\alpha_k + \Delta\alpha_k$; this is the mechanism that makes the guarantee quantitative. The final top-$k$ selection happens without the cloud seeing the query: cosine distances are computed on encrypted values under partially homomorphic encryption, and the document indices are sent directly only when the mean embedding of the top-$k$ cluster is within the privacy budget, otherwise a $k$-out-of-$k'$ oblivious transfer is used. The paper reports 100% recall in all tested settings and a sharp drop in reconstruction-attack success as the perturbation grows.

Load-bearing premise

The lossless-retrieval guarantee rests on the assumption that document embeddings are spread uniformly over the surface of a high-dimensional unit sphere; real embeddings cluster by topic, so Theorem 1's candidate size may be too small when that assumption fails.

Editorial extensions

If this is right

  • A user can delegate RAG to a semi-honest cloud while keeping the query's semantic content within a formally bounded leakage budget.
  • The cloud searches only $k'$ documents instead of all $N$, so per-query computation drops to sub-second levels for a million-document store.
  • The accuracy of the protocol becomes a theorem rather than an empirical hope, as long as the uniform-sphere premise holds.
  • The fully cryptographic baseline, encrypting and scanning the whole collection, is unnecessary in the regimes the paper tests.
  • The same perturb-then-expand recipe could be reused for any similarity search task where privacy of the query vector matters.

Reading between the lines

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

  • Because real embeddings are topic-clustered, a practical deployment should add a safety margin to Theorem 1's $k'$ or estimate the local embedding density empirically instead of relying on the uniform-sphere formula.
  • The protocol assumes an embedding model the user can run locally and that stays frozen on both sides; a cloud that switches to a proprietary or updated encoder would break the user's ability to generate the perturbation.
  • The privacy guarantee is framed against a semi-honest cloud; a malicious cloud that stores perturbed queries across many requests could correlate them over time, a leakage channel the paper does not quantify.
  • When the oblivious-transfer path is used, communication grows with $k'$ because $k'$ encrypted documents are transferred, so choosing the direct path whenever Theorem 3's angle condition holds is what keeps the reported 46.66 KB transmission small.
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

3 major / 5 minor

Summary. RemoteRAG proposes a privacy-preserving protocol for cloud-based retrieval-augmented generation. A user perturbs their query embedding using a Laplace-like mechanism that is claimed to satisfy (n,ε)-DistanceDP, sends the perturbed embedding to the cloud, and requests the top-k' documents for this noisy vector. The user then uses partially homomorphic encryption (PHE) to have the cloud compute cosine distances over this candidate set, decrypts them, and obtains the true top-k indices. To return the documents, either the indices are sent directly to the cloud (when a mean-angle condition is satisfied) or a k-out-of-k' oblivious transfer (OT) is used. The paper claims lossless retrieval (the top-k are always contained in the k' candidates) with a theoretical guarantee based on a uniform-sphere model, and reports high efficiency (0.67 s and 46.66 KB for 10^6 documents) plus resistance to the Vec2Text inversion attack.

Significance. If the stated claims held, RemoteRAG would be a practical contribution to private RAG: it formalizes the problem, introduces a metric-DP notion for embeddings, and combines perturbation, PHE, and OT in one protocol. The derivation of k' from a geometric model is a parameter-free formula that is then verified across a grid of settings (N, k, r, five embedding models), which is a strength relative to purely empirical heuristics. The efficiency results and the 100% recall in the reported experiments are also encouraging. However, the central privacy claim is not established for the efficient 'Direct' branch, and the accuracy guarantee rests on a uniform-distribution assumption that the paper itself acknowledges can fail. These issues are load-bearing for the paper's main contributions, so the work requires substantial revision before the claims are supportable.

major comments (3)
  1. [§3.3.2, §4.1, Theorem 3, Definition 1] The claim that module 2(b) (Direct mode) achieves the stated privacy is not established. The cloud receives the exact top-k index set, which is a deterministic function of the true query embedding with unbounded sensitivity: there exist pairs of queries with arbitrarily small L2 distance whose top-k sets are disjoint. For such a pair, the joint output (perturbed embedding plus index set) has probability zero under one query and nonzero under the other, so the privacy loss is infinite for every finite ε, violating Definition 1. Theorem 3 bounds only the mean angle between the query embedding and the average of the top-k document embeddings; it does not constrain the specific indices, which can identify the query even when the mean is far. The privacy experiments in Section 5.2 test Vec2Text only on the perturbed embedding and do not test index disclosure. The OT branch (module 2(c)) does hide indices, but the headline efficiency numbers in the Abstract and Table 4 are for Direct mode. Please either remove the privacy claim for Direct mode, always use OT, or redesign the index release with a calibrated randomized mechanism and prove its privacy.
  2. [§3.2.2, Lemma 1, Theorem 1, Appendix B.5.2] The theoretical accuracy guarantee is conditional on Lemma 1's assumption that document embeddings are uniformly distributed on the unit n-sphere, but the Abstract and Section 3.2.2 state the guarantee as unconditional. Real text embeddings cluster by topic and are not uniformly distributed. Appendix B.5.2 explicitly acknowledges that "in some rare exceptions, there might be a chance of RemoteRAG failing to preserve the top-k documents" and provides a 2D counterexample. The paper offers no quantitative bound on the failure probability or on how k' must be enlarged for non-uniform distributions, so the "no loss in retrieval" claim in Section 5.3 and Table 3 is supported only empirically, not by the stated theorems. This is load-bearing for the accuracy contribution. Please either provide a distribution-robust bound (for example, using local intrinsic dimension or covering numbers) or clearly state the guarantee as conditional and qualify the Abstract and conclusion accordingly.
  3. [Appendix C, Theorem 6 proof] The proof of Theorem 3 relies on the approximation that when n is large, the top-k embeddings lie exactly at angle α_k from the query embedding ("we approximately believe") and are then treated as uniformly distributed on a lower-dimensional sphere of radius sin α_k. This is not a rigorous bound: the top-k embeddings lie in a spherical cap, not on a full sphere, and the mean of points from a cap can deviate from the stated formula. Because Theorem 3 is used in Algorithm 2 (line 7) to decide whether the Direct branch is safe, this approximation affects an operational security decision. Please provide a quantified error bound or a rigorous derivation, or state the condition as heuristic.
minor comments (5)
  1. [Definition 1] The definition of L(K(x),K(x')) as ln(Pr(K(x)=y)/Pr(K(x')=y)) is ambiguous for continuous mechanisms, where the probability of a single point is zero; it should be stated as a supremum over output sets or using density ratios.
  2. [Section 5.3] Please report the number of runs and the worst-case recall per setting, not only the average; the default of 50 experiments appears only in Appendix B.4, and with N=10^6 and k=5, 50 samples may not reveal rare failures.
  3. [Appendix B.5.2] The "rare exceptions" limitation is important enough to be stated in the main text or in the Limitations section; its current placement under "Extra Experimental Results" understates its significance for the accuracy claim.
  4. [Table 5] Typo in the caption: "Patermeters" should be "Parameters".
  5. [Section 3.3.1, Appendix A.1] For reproducibility, please specify the PHE scheme used in the experiments (e.g., Paillier) and its security parameters, as the communication and computation costs depend on the ciphertext size.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: k' is computed from a stated geometric model and validated empirically; privacy definition is imported from external DP/geo-indistinguishability work rather than reduced to its own conclusions.

full rationale

The claimed derivation chain is self-contained. (n,epsilon)-DistanceDP is defined from first principles with external citations (Andres et al. 2013; Dwork et al. 2006), and the Laplace/gamma perturbation mechanism is standard. The accuracy result (Theorem 1 and Lemma 1) computes k' from N, k, and the perturbation angle via a uniform-sphere model; the reported 100% recall is an empirical check on that computed k' across a grid of settings, not a fitted parameter used to produce the k' formula. The privacy analysis for Module 2(b) bounds the mean angle omega using Theorem 3, and the condition omega >= Delta alpha_k is a direct comparison to the perturbation budget, not a circular definition. The paper explicitly acknowledges the uniform-sphere assumption's fragility in Appendix B.5.2 ('rare exceptions'), which is a correctness/robustness concern, not circularity. There are no author self-citations invoked as load-bearing evidence; all cited building blocks (geo-indistinguishability, PHE, OT, Vec2Text) are external and independently established. The Direct-mode index-set sensitivity issue raised by the reviewer is a potential gap between the claimed privacy property and what is proven, but it does not make the derivation circular: the paper simply does not establish the required hiding property for the exact index set. Accordingly, the circularity score is 0.

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

The central claim depends on three modeling assumptions (uniform embedding distribution, small-angle approximation, and the angular concentration approximation in Theorem 6) plus standard cryptographic security assumptions. The only hand-chosen parameters are the privacy budget epsilon, the perturbation magnitude r in experiments, and the default candidate count k'. No novel entities are postulated.

free parameters (3)
  • privacy budget epsilon = Typical range 10n to 50n in Section 3.2.1; experiments use values from 3000 to 50000 (Figure 4)
    User-chosen knob that controls the Gamma-distributed perturbation. The operating range is a hand-selected guideline, not derived from first principles or from data.
  • perturbation magnitude r = Default 0.03; varied as 0.03, 0.05, 0.07, 0.1 in Table 3
    In experiments the perturbation is directly fixed to specific values rather than only sampled from the Gamma distribution. These values are hand-picked to balance privacy and efficiency, and the same r is used for all magnitude comparisons.
  • candidate count k' = Default 160; recommended range 100 to 200 in Appendix B.5.1
    The expanded candidate set size is set as an operating point for efficiency in the experiments. It is theoretically computable from Theorem 1 given epsilon and the uniform-sphere assumption, but the default value is a hand-chosen trade-off rather than a measurement.
assumptions (4)
  • domain assumption All N document embeddings are uniformly distributed on the surface of the n-dimensional unit sphere (Lemma 1, Section 3.2.2).
    This is the foundation for the k-to-alpha_k relation and hence for the computation of k' in Theorem 1. Real embeddings are not uniformly distributed; the paper labels violations as 'rare exceptions' in Appendix B.5.2 without a quantitative bound.
  • domain assumption Perturbed angle approximation: Delta alpha_k is approximately equal to the radial perturbation r (Section 3.2.2).
    The theorem alpha_k' = alpha_k + Delta alpha_k relies on small-angle approximation. It is only valid for perturbations much smaller than 1, which matches the operating range but is not a hard guarantee.
  • domain assumption In Theorem 6, top-k embeddings lie at exactly polar angle alpha_k and are uniformly distributed on an (n-1)-sphere of radius sin(alpha_k), with n >> 1 (Appendix C, proof of Theorem 6).
    This approximation is used to derive tan(omega) = tan(alpha_k)/sqrt(k), which determines whether index protection via OT is needed. It is a heuristic that becomes plausible only in high dimensions and is not proven for real embedding distributions.
  • domain assumption Security of the partially homomorphic encryption (PHE) and the k-out-of-k' oblivious transfer protocol (Sections 3.3 and Appendix A.1).
    The privacy guarantees in Module 2 depend on the underlying cryptographic primitives being secure. The paper cites standard schemes but provides no formal security proof or parameter selection details.
invented entities (1)
  • None
    purpose: No new particles, forces, mediators, dimensions, or ledger entries are introduced.
    (n,epsilon)-DistanceDP is a formal privacy definition, not a physical or cryptographic entity. No new primitive is invented beyond a renamed existing notion.

how reviews work

0 comments
Cite this review

Pith. "Pith review of RemoteRAG: A Privacy-Preserving LLM Cloud RAG Service." pith.science (2026). https://pith.science/paper/V7FKHI6I

@misc{pith2026241212775,
  author       = {Pith},
  title        = {Pith review of: RemoteRAG: A Privacy-Preserving LLM Cloud RAG Service},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/V7FKHI6I}},
  note         = {Machine review of arXiv:2412.12775}
}
abstract

Retrieval-augmented generation (RAG) improves the service quality of large language models by retrieving relevant documents from credible literature and integrating them into the context of the user query. Recently, the rise of the cloud RAG service has made it possible for users to query relevant documents conveniently. However, directly sending queries to the cloud brings potential privacy leakage. In this paper, we are the first to formally define the privacy-preserving cloud RAG service to protect the user query and propose RemoteRAG as a solution regarding privacy, efficiency, and accuracy. For privacy, we introduce $(n,\epsilon)$-DistanceDP to characterize privacy leakage of the user query and the leakage inferred from relevant documents. For efficiency, we limit the search range from the total documents to a small number of selected documents related to a perturbed embedding generated from $(n,\epsilon)$-DistanceDP, so that computation and communication costs required for privacy protection significantly decrease. For accuracy, we ensure that the small range includes target documents related to the user query with detailed theoretical analysis. Experimental results also demonstrate that RemoteRAG can resist existing embedding inversion attack methods while achieving no loss in retrieval under various settings. Moreover, RemoteRAG is efficient, incurring only $0.67$ seconds and $46.66$KB of data transmission ($2.72$ hours and $1.43$ GB with the non-optimized privacy-preserving scheme) when retrieving from a total of $10^6$ documents.

Figures

Figures reproduced from arXiv: 2412.12775 by the authors.

Figure 1
Figure 1. The flowchart of RemoteRAG. Module 1 preserves privacy with [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. The probability density function of different [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 4
Figure 4. The attack performance corresponding to the [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figures from the paper (4 more)
Figure 5
Figure 5. Figure 5: Efficiency study of each module. Communication cost. From the results in Fig￾ure 5(b), module 2(a) has a larger starting point, but the transmission size of module 2(c) soon surpasses module 2(a) as 𝑘 ′ increases. Numerous basic pa￾rameters in PHE cause the former whil…
Figure 6
Figure 6. Figure 6: Relationships among hyperparameters. number of possible embeddings with the same top 𝑘 documents also increases. Therefore, the same value of 𝑘 ′ implies a looser privacy requirement, which is reflected by a larger privacy budget 𝜖. B.5.2 2D Simulations The following t…
Figure 8
Figure 8. Figure 8: Illustration of the proof of Lemma 2. Proof. Define 𝑆𝑛,𝑟 (𝛼) = Ω𝑛(𝛼)𝑟 𝑛−1 as the surface area of the spherical sector with a polar angle 𝛼 ∈ [0, 𝜋] in the 𝑛-sphere with radius 𝑟, where Ω𝑛(𝛼) represents the corresponding surface area in the unit 𝑛-sphere. Then, referrin…
Figure 9
Figure 9. Figure 9: Illustration of the proof of Theorem 6. Proof. Lemma 2 tells us that top 𝑘 embeddings are within the polar angle 𝛼𝑘. When 𝑛 ≫ 1, we approximately believe that the angle they make with 𝑒𝑘 is exactly 𝛼𝑘, which means 𝑘 embeddings are uniformly distributed on the surface o…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Collaborative Inference and Learning between Edge SLMs and Cloud LLMs: A Survey of Algorithms, Execution, and Open Challenges

    cs.DC 2025-07 conditional novelty 4.0 of 10

    A survey that builds a taxonomy of edge-cloud LLM-SLM collaboration for inference and training, claiming to be the first to unify both phases.

Reference graph

Works this paper leans on

43 extracted references · 19 canonical work pages · cited by 1 Pith paper

  1. [1]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block STRING...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Andr \' e s, Nicol \' a s Emilio Bordenabe, Konstantinos Chatzikokolakis, and Catuscia Palamidessi

    Miguel E. Andr \' e s, Nicol \' a s Emilio Bordenabe, Konstantinos Chatzikokolakis, and Catuscia Palamidessi. 2013. https://doi.org/10.1145/2508859.2516735 Geo-indistinguishability: differential privacy for location-based systems . In 2013 ACM SIGSAC Conference on Computer and Communications Security, CCS'13, Berlin, Germany, November 4-8, 2013 , pages 90...

    arXiv 2013

  4. [4]

    Rae, Erich Elsen, and Laurent Sifre

    Sebastian Borgeaud, Arthur Mensch, Jordan Hoffmann, Trevor Cai, Eliza Rutherford, Katie Millican, George van den Driessche, Jean - Baptiste Lespiau, Bogdan Damoc, Aidan Clark, Diego de Las Casas, Aurelia Guy, Jacob Menick, Roman Ring, Tom Hennigan, Saffron Huang, Loren Maggiore, Chris Jones, Albin Cassirer, Andy Brock, Michela Paganini, Geoffrey Irving, O...

    2022

  5. [5]

    Chatchat - Space. 2023. L angchain- C hatchat. https://github.com/chatchat-space/Langchain-Chatchat

    work page 2023

  6. [6]

    Jianlv Chen, Shitao Xiao, Peitian Zhang, Kun Luo, Defu Lian, and Zheng Liu. 2024. https://doi.org/10.48550/ARXIV.2402.03216 BGE m3-embedding: Multi-lingual, multi-functionality, multi-granularity text embeddings through self-knowledge distillation . CoRR, abs/2402.03216

    arXiv doi:10.48550/arxiv.2402.03216 2024

  7. [7]

    Tung Chou and Claudio Orlandi. 2015. https://doi.org/10.1007/978-3-319-22174-8_3 The simplest protocol for oblivious transfer . In Progress in Cryptology - LATINCRYPT 2015 - 4th International Conference on Cryptology and Information Security in Latin America, Guadalajara, Mexico, August 23-26, 2015, Proceedings , volume 9230 of Lecture Notes in Computer S...

    doi:10.1007/978-3-319-22174-8_3 2015

  8. [8]

    Chroma. 2022. C hroma. https://github.com/chroma-core/chroma

    work page 2022

Show all 43 references
  1. [9]

    Shehzaad Dhuliawala, Mojtaba Komeili, Jing Xu, Roberta Raileanu, Xian Li, Asli Celikyilmaz, and Jason Weston. 2023. https://doi.org/10.48550/ARXIV.2309.11495 Chain-of-verification reduces hallucination in large language models . CoRR, abs/2309.11495

    2023 doi arXiv

  2. [10]

    Matthijs Douze, Alexandr Guzhva, Chengqi Deng, Jeff Johnson, Gergely Szilvasy, Pierre - Emmanuel Mazar \' e , Maria Lomeli, Lucas Hosseini, and Herv \' e J \' e gou. 2024. https://doi.org/10.48550/ARXIV.2401.08281 The faiss library . CoRR, abs/2401.08281

    2024 doi arXiv

  3. [11]

    Cynthia Dwork, Frank McSherry, Kobbi Nissim, and Adam D. Smith. 2006. https://doi.org/10.1007/11681878_14 Calibrating noise to sensitivity in private data analysis . In Theory of Cryptography, Third Theory of Cryptography Conference, TCC 2006, New York, NY, USA, March 4-7, 200...

    2006 doi

  4. [12]

    E lastic. 2010. E lasticsearch. https://github.com/elastic/elasticsearch

    2010

  5. [13]

    Geniusee. 2024. R A G as a service. https://geniusee.com/retrieval-augmented-generation

    2024

  6. [14]

    Gautier Izacard and Edouard Grave. 2021. https://doi.org/10.18653/V1/2021.EACL-MAIN.74 Leveraging passage retrieval with generative models for open domain question answering . In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Li...

    2021 doi

  7. [15]

    L ang C hain. 2022. L ang C hain. https://github.com/langchain-ai/langchain

    2022

  8. [16]

    L ang C hain. 2024. H ow to recursively split text by characters. https://python.langchain.com/v0.2/docs/how_to/recursive_text_splitter/

    2024

  9. [17]

    L ang G enius. 2023. D ify. https://github.com/langgenius/dify

    2023

  10. [18]

    Florian Leiser, Sven Eckhardt, Valentin Leuthe, Merlin Knaeble, Alexander M \" a dche, Gerhard Schwabe, and Ali Sunyaev. 2024. https://doi.org/10.1145/3613904.3642428 HILL: A hallucination identifier for large language models . In Proceedings of the CHI Conference on Human Fac...

    2024

  11. [19]

    u ttler, Mike Lewis, Wen - tau Yih, Tim Rockt \

    Patrick S. H. Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich K \" u ttler, Mike Lewis, Wen - tau Yih, Tim Rockt \" a schel, Sebastian Riedel, and Douwe Kiela. 2020. https://proceedings.neurips.cc/paper/2020/hash/6b493230205f780e...

    2020

  12. [20]

    Jiatong Li, Yunqing Liu, Wenqi Fan, Xiao - Yong Wei, Hui Liu, Jiliang Tang, and Qing Li. 2023. https://doi.org/10.48550/ARXIV.2306.06615 Empowering molecule discovery for molecule-caption translation with large language models: A chatgpt perspective . CoRR, abs/2306.06615

    2023 doi arXiv

  13. [21]

    Xianming Li and Jing Li. 2023. https://doi.org/10.48550/ARXIV.2309.12871 Angle-optimized text embeddings . CoRR, abs/2309.12871

    2023 doi arXiv

  14. [22]

    L lama I ndex. 2023. E valuating the I deal C hunk S ize for a R A G S ystem using L lama I ndex. https://www.llamaindex.ai/blog/evaluating-the-ideal-chunk-size-for-a-rag-system-using-llamaindex-6207e5d3fec5

    2023

  15. [23]

    Xinbei Ma, Yeyun Gong, Pengcheng He, Hai Zhao, and Nan Duan. 2023. https://doi.org/10.48550/ARXIV.2305.14283 Query rewriting for retrieval-augmented large language models . CoRR, abs/2305.14283

    2023 doi arXiv

  16. [24]

    Morris, Volodymyr Kuleshov, Vitaly Shmatikov, and Alexander M

    John X. Morris, Volodymyr Kuleshov, Vitaly Shmatikov, and Alexander M. Rush. 2023. https://doi.org/10.18653/V1/2023.EMNLP-MAIN.765 Text embeddings reveal (almost) as much as text . In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, EMNLP...

    2023 doi

  17. [25]

    Tri Nguyen, Mir Rosenberg, Xia Song, Jianfeng Gao, Saurabh Tiwary, Rangan Majumder, and Li Deng. 2016. https://ceur-ws.org/Vol-1773/CoCoNIPS_2016_paper9.pdf MS MARCO: A human generated machine reading comprehension dataset . In Proceedings of the Workshop on Cognitive Computat...

    2016

  18. [26]

    Nuclia. 2024. N uclia, the all-in-one R A G as a service platform. https://nuclia.com/rag-as-a-service/

    2024

  19. [27]

    OpenAI. 2022 a . I ntroducing C hat G P T . https://openai.com/index/chatgpt/

    2022

  20. [28]

    OpenAI. 2022 b . N ew and improved embedding model. https://openai.com/index/new-and-improved-embedding-model/

    2022

  21. [29]

    OpenAI. 2024. N ew embedding models and API updates. https://openai.com/index/new-embedding-models-and-api-updates/

    2024

  22. [30]

    Yang, Zachary DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit Steiner, Lu Fang, Junjie Bai, and Soumith Chintala

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas K \" o pf, Edward Z. Yang, Zachary DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit Stei...

    2019

  23. [31]

    Matt Post. 2018. https://doi.org/10.18653/V1/W18-6319 A call for clarity in reporting BLEU scores . In Proceedings of the Third Conference on Machine Translation: Research Papers, WMT 2018, Belgium, Brussels, October 31 - November 1, 2018 , pages 186--191. Association for Comp...

    2018 doi

  24. [32]

    Q uivr H Q . 2023. Q uivr. https://github.com/QuivrHQ/quivr

    2023

  25. [33]

    SentenceTransformers. 2021 a . all- M ini L M - L 12-v2. https://huggingface.co/sentence-transformers/all-MiniLM-L12-v2

    2021

  26. [34]

    SentenceTransformers. 2021 b . all-mpnet-base-v2. https://huggingface.co/sentence-transformers/all-mpnet-base-v2

    2021

  27. [35]

    SentenceTransformers. 2022. gtr-t5-base. https://huggingface.co/sentence-transformers/gtr-t5-base

    2022

  28. [36]

    Sophia Yang

    Ph.D. Sophia Yang. 2023. A dvanced R A G 01: S mall-to- B ig R etrieval. https://towardsdatascience.com/advanced-rag-01-small-to-big-retrieval-172181b396d4

    2023

  29. [37]

    Paul Voigt and Axel von dem Bussche. 2017. https://doi.org/10.1007/978-3-319-57959-7 The EU General Data Protection Regulation (GDPR) . Springer International Publishing

    2017 doi

  30. [38]

    Voyage. 2024. E mbeddings. https://docs.voyageai.com/docs/embeddings

    2024

  31. [39]

    Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, R \' e mi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger,...

    2020 doi

  32. [40]

    Jia - Yu Yao, Kun - Peng Ning, Zhen - Hui Liu, Munan Ning, and Li Yuan. 2023. https://doi.org/10.48550/ARXIV.2310.01469 LLM lies: Hallucinations are not bugs, but features as adversarial examples . CoRR, abs/2310.01469

    2023 doi arXiv

  33. [41]

    Shenglai Zeng, Jiankun Zhang, Pengfei He, Yue Xing, Yiding Liu, Han Xu, Jie Ren, Shuaiqiang Wang, Dawei Yin, Yi Chang, and Jiliang Tang. 2024. https://doi.org/10.48550/ARXIV.2402.16893 The good and the bad: Exploring privacy issues in retrieval-augmented generation (RAG) . CoR...

    2024 doi arXiv

  34. [42]

    Le, and Ed H

    Denny Zhou, Nathanael Sch \" a rli, Le Hou, Jason Wei, Nathan Scales, Xuezhi Wang, Dale Schuurmans, Claire Cui, Olivier Bousquet, Quoc V. Le, and Ed H. Chi. 2023. https://openreview.net/forum?id=WZH7099tgfM Least-to-most prompting enables complex reasoning in large language mo...

    2023

  35. [43]

    Shengyao Zhuang, Bevan Koopman, Xiaoran Chu, and Guido Zuccon. 2024. https://doi.org/10.48550/ARXIV.2402.12784 Understanding and mitigating the threat of vec2text to dense retrieval systems . CoRR, abs/2402.12784

    2024 doi arXiv

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.