{"id":"db6c01a9-7fb2-4e74-8eff-808d3b1ca341","arxiv_id":"2411.16809","paper_version":1,"verdict":"REJECT","confidence":"HIGH","novelty_score":1.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"A literature survey on bidirectional integration of blockchain and large language models, presenting no new experimental or theoretical results.","lead":"This paper surveys current research on using large language models in blockchain systems and using blockchain to support large language models. It offers a taxonomy of six directions, but the manuscript is a rough draft with broken citations, duplicated references, and uneven writing.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Citation failures undermine the central map: a key direction rests on an unresolvable reference and duplicated references inflate the cited literature.","rationale":"The reader's weakest assumption was that the cited papers are real, correctly described, and representative, and that assumption is already violated by the unresolved PDPChain citation and duplicated references. My independent reading confirms this and adds that the duplication is not limited to one pair: [27]/[31]/[35] are the same paper, and [24]/[42] are also the same paper. This means the survey overstates the number of independent contributions in at least one of its six claimed directions, and one direction contains an unverifiable citation. The paper also lacks any stated methodology for literature selection, so representativeness cannot be independently assessed; however, the concrete citation failures are sufficient to reject the current version without relying on methodological absence. I found no machine-checked proofs, reproducible code, or parameter-free derivations that would provide independent support. The verdict should remain REJECT because the central claim of providing a reliable and comprehensive overview is directly undermined by the manuscript's own evidentiary apparatus, not merely by stylistic or organizational weaknesses.","tokens_in":17969,"tokens_out":6505,"duration_ms":59430,"concrete_test":"Construct a citation-integrity audit for all references cited in Section II's six subsections: resolve every reference to a DOI or arXiv identifier, deduplicate identical works, and check each in-text attribution against the source's abstract. Specifically, attempt to locate the PDPChain paper by Liang cited as \"[?]\" in Section II-A2; if it cannot be found, that direction lacks evidentiary support. Count the number of unique papers per subsection before and after deduplication; if the smart-contract-auditing direction drops from three cited works to one after merging [27], [31], and [35], the six-direction map is not supported by the cited literature as presented. Separately, verify whether reference [62]'s retraction is disclosed where it is used as evidence.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central contribution is a six-direction map of blockchain-LLM integration, and that map is only as trustworthy as the underlying citations. This condition fails in concrete, checkable ways. Section II-A2's entire description of PDPChain rests on \"Liang [?]\" with no resolvable reference, leaving one of the six claimed directions without a verifiable primary source. The same paper by Wei et al. on smart contract auditing is cited as [27], [31], and [35], so a single work is counted as three separate developments under the \"smart contract auditing\" direction. Similarly, the vulnerability-constrained decoding work by Storhaug et al. is cited twice as [24] and [42]. These duplicates inflate the apparent volume of distinct research behind the six directions. Additionally, reference [62] is marked as retracted in the bibliography yet is used in Section IV.B.4 as uncontested evidence for blockchain traceability, and several cited items are only loosely related to the claimed integration (e.g., [39] is about generative AI in construction). Because a survey's only product is reliable synthesis, these citation-level defects directly undermine the central claim that the six subareas form a useful and trustworthy map.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript presents itself as a living survey of the bidirectional integration of blockchain and large language models. The abstract announces two directions of technical convergence: applying large language models to blockchain, where six major development directions are claimed, and applying blockchain to large language models. The body reviews a range of topics including privacy-preserving data sharing, decentralized federated learning, decentralized AI, smart contract auditing, DAO governance, and the advantages and constraints of both technologies. The paper's central contribution is intended to be a usable map of current research on blockchain-LLM integration.","tokens_in":18089,"tokens_out":4782,"duration_ms":44343,"significance":"The topic is timely and the intended two-direction taxonomy could serve as a convenient entry point for researchers entering this emerging area. The paper touches on genuinely active lines of work, such as blockchain-based privacy protection for LLM training data, decentralized federated learning, LLM-based smart contract auditing, and LLM support for DAO governance. However, a survey's only product is reliable synthesis, and the manuscript's citation apparatus is not reliable: it contains an unresolved placeholder citation, duplicate references that inflate the apparent volume of distinct contributions, use of a retracted source as uncontested evidence, and a structural mismatch between the abstract's claimed six directions and the actual section organization. The paper provides no search protocol, inclusion criteria, or other methodological scaffolding that would support the representativeness of its chosen sources. Because the central map is only as trustworthy as its citations, these are load-bearing defects rather than cosmetic ones.","major_comments":[{"comment":"The abstract states that the paper identifies six major development directions under the application of large language models to blockchain, but Section II organizes six labeled subsections (A-F) of which A-C are blockchain-for-LLM topics and D-F are LLM-for-blockchain topics. The six directions are never explicitly enumerated, and subsections D and E both address smart contract auditing and vulnerability detection, so the claimed six-direction map is not reproducible from the text.","section":"Abstract; Section II"},{"comment":"The description of the PDPChain system rests entirely on the citation \"Liang [?]\", with no resolvable reference in the bibliography. This leaves a key example under the data management and privacy protection direction without a verifiable primary source, directly undermining the survey's reliability.","section":"II-A2"},{"comment":"Two works are each cited multiple times as if they were distinct contributions: the vulnerability-constrained decoding paper by Storhaug et al. appears as both [24] and [42], and the fine-tuned smart contract auditing paper by Wei et al. appears as [27], [31], and [35]. These duplicates inflate the apparent number of independent research results behind the smart contract auditing directions and create inconsistencies in the narrative.","section":"II-D, II-E, references [24], [27], [31], [35], [42]"},{"comment":"Section IV.B.4 quotes reference [62] as uncontested evidence for blockchain traceability in food supply chains, but the bibliography itself marks this source as \"[retracted]\". Using a retracted paper as a supporting citation without any caveat is a serious reliability flaw in a survey.","section":"IV-B4, reference [62]"},{"comment":"The passage on the blockchain trilemma states that \"the original trilemma of scalability, decentralization, security, and trust has expanded into a four-fold challenge,\" yet it lists four properties as the original trilemma and then lists the same four properties as the expansion. This is internally inconsistent and confuses the paper's framing of blockchain constraints.","section":"I"},{"comment":"The future-directions discussion relies heavily on references [69]-[80], most of which are authored or co-authored by the same author and concern federated learning, cloud robotics, smart cities, and network slicing rather than blockchain-LLM integration. The section asserts the relevance of these works to blockchain-LLM convergence without argument, so the prospects discussion does not provide independent support for the survey's map and raises concerns about the breadth of the literature considered.","section":"VI, references [69]-[80]"}],"minor_comments":[{"comment":"The text uses reference [5], which is a paper on extracting training data from diffusion models, to support a claim about LLMs disclosing personal identity information; the mismatch should be corrected.","section":"I, reference [5]"},{"comment":"The heading \"Vulnerablilities detection\" contains a spelling error; it should read \"Vulnerability detection\".","section":"II-E heading"},{"comment":"The affiliation text contains the typo \"Blockchian Group\"; it should read \"Blockchain Group\".","section":"Author affiliation"},{"comment":"The text refers to a figure \"blockchain.png\" at the end of the manuscript, but no figure is included or discussed in the body of the paper.","section":"Figure caption"},{"comment":"Section III largely repeats material already covered in Section II, often with vaguer summaries (for example, III.A.1 and III.B.2 do not add new technical content), and could be merged or referenced to Section II to avoid redundancy.","section":"III"}],"recommendation":"reject","confidential_remarks":"In my view the manuscript is not yet at the standard of a publishable survey. The citation failures (placeholder, duplicates, retracted source) and the abstract/body structural mismatch are central to the paper's purpose, and the heavy concentration of Section VI references on one author's related work raises a concern about the independence of the literature base. I would not encourage a revision within the current scope; a from-scratch rewrite with a documented survey methodology, a clearly enumerated taxonomy, and a carefully vetted reference list would be needed before the paper could be reconsidered."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague—this is a survey of the blockchain/LLM intersection, organized into two directions: LLMs applied to blockchain, and blockchain applied to LLMs, with six subareas under the first. The topic is real and the organization is sensible for a newcomer. Some of the cited papers are relevant and useful. But the reference apparatus is in bad shape, and since a survey's only product is a reliable map, that is fatal.\n\nThe concrete failures are easy to verify. Section II-A2's description of PDPChain rests on 'Liang [?]'—no resolvable reference. The same arXiv paper on fine-tuned smart contract auditing appears as [27], [31], and [35], so one work counts as three separate developments. The vulnerability-constrained decoding paper is [24] and [42]. Reference [62] is marked retracted in the bibliography but is used without caveat in Section IV.B.4 as evidence for blockchain traceability. The prospects section leans on a run of co-author self-citations [69]–[80] with no external support. These are not cosmetic issues; they inflate the apparent volume and diversity of the literature.\n\nWhat the paper does well: the two-direction framing is clear, and the challenges sections (e.g., scalability, privacy) are competently summarized. It could be a useful orientation piece if the citations were fixed and the prose cleaned up. But the proposal is not new—the paper's own reference [49] already surveys 'blockchain large language models'—and the paper does not build on or acknowledge that overlap.\n\nBottom line: I would not send this to a serious referee in its current form. The citation trust is the load-bearing wall, and it's cracked. Desk reject, tell the authors to rework the bibliography, acknowledge [49], and resubmit. If they do, the field could use a clean survey; right now this isn't it.","headline":"A sensible two-direction taxonomy, but the load-bearing references are broken—the survey can't be trusted until the bibliography is rebuilt.","tokens_in":18686,"tokens_out":3395,"would_cite":false,"duration_ms":31203,"reading_group":"no","serious_thinker":"yes","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A living survey of blockchain and large language models argues that the two technologies are converging in two directions and six development areas, with each side supplying what the other lacks.","keywords":["blockchain","large language models","smart contract auditing","decentralized federated learning","decentralized AI","DAO governance","privacy protection","living survey"],"falsifier":"A concrete check is to resolve every citation in the survey against arXiv and publisher databases. If the PDPChain system named in Section II-A2 cannot be located because its '[?]' citation is unresolvable, and if the duplicated paper [27]/[31]/[35] is counted once, the survey's census of the field shrinks; enough failed references would mean the claimed six-direction map cannot be confirmed as a faithful representation of the literature.","tokens_in":17699,"feed_emoji":"🔗","tokens_out":7865,"duration_ms":66227,"temperature":0.7,"pith_summary":"The paper is a living survey of the technical convergence between blockchain and large language models. It organizes current research into six development directions under two headings: using blockchain to strengthen LLMs, and using LLMs to strengthen blockchain. The directions are data management and privacy protection, decentralized model training, decentralized AI, smart contract auditing, vulnerability detection, and DAO language interaction and decision support. The survey's stated purpose is to give researchers a shared map of where integration already has evidence and where it remains prospective, pairing each technology's core weakness with the other's core strength.","feed_headline":"Survey maps six ways blockchain and LLMs can fix each other","feed_subtitle":"The survey sorts the integration literature into two directions and six subareas, pairing each technology's weakness with the other's…","key_machinery":"The key machinery is the survey's organizing taxonomy: two directions, six development areas. The blockchain-to-LLM side covers privacy-preserving data sharing and training, decentralized federated learning and incentive mechanisms, and decentralized AI systems such as GradientCoin and opML. The LLM-to-blockchain side covers smart contract auditing frameworks such as iAudit and FTSmartAudit, vulnerability-constrained decoding during code completion, and LLM-based DAO proposal classification. The taxonomy does the work of turning a scattered set of systems into a map: each area is defined by the blockchain feature it exploits or the LLM capability it applies, and the map becomes a checklist for where integration has evidence and where it is still prospective.","core_discovery":"The paper's central claim is that blockchain and large language models are already bidirectionally integrating, and that this integration is best understood through six development directions: data management and privacy protection, decentralized model training, decentralized AI, smart contract auditing, vulnerability detection, and DAO language interaction and decision support. In the blockchain-to-LLM direction, the survey argues that blockchain's decentralization, immutability, traceability, and tamper-proof storage can mitigate LLM weaknesses in data privacy, single-point-of-failure training, and trust in generated content. In the LLM-to-blockchain direction, it argues that LLMs' natural-language and code-generation abilities can strengthen smart contract security auditing, vulnerability detection, and decentralized governance. The survey supports each direction with concrete named systems, including federated-learning platforms, audit frameworks, and a DAO proposal classifier reported at 95 percent accuracy.","pith_inferences":["One testable extension is to measure the six directions bibliometrically: count peer-reviewed systems per direction and track growth rates over time, which would reveal which directions rest on mature results and which rest on isolated preprints.","The survey's implicit pairing principle, where each technology's weakness maps onto the other's strength, suggests a design template for future systems: identify one LLM limitation, then select the blockchain property of immutability, traceability, decentralization, or incentives that addresses it, and vice versa.","Because the paper's own examples concentrate on audits and decentralized training, the map can be read as predicting that the most immediately practical integrations are LLM-assisted code security and blockchain-mediated federated learning."],"forward_implications":["Blockchain can serve as a decentralized coordination and incentive layer for federated LLM training, removing the central server as a single point of failure.","LLM-based tools can substantially automate smart contract auditing, from vulnerability detection to repair suggestion, complementing traditional static and dynamic analyzers.","DAO governance can scale proposal review with LLM classification, but the survey's cited work shows that transparency and consistency gaps must be resolved first.","The convergence remains in an exploratory stage: most of the systems cited are research proposals or early frameworks rather than production deployments.","The two technologies' constraint sets line up, so the direction of integration is not symmetric: blockchain supplies verification and provenance, while LLMs supply interpretation and generation."],"supporting_citations":[{"why":"Presents blockchain-managed on-device federated learning, load-bearing for the decentralized training direction.","marker":"[11]"},{"why":"Describes a peer-to-peer decentralized LLM framework with a gradient-coin incentive mechanism, anchoring the decentralized AI direction.","marker":"[21]"},{"why":"Introduces optimistic machine learning with fraud proofs for on-chain AI inference, anchoring the decentralized AI direction.","marker":"[22]"},{"why":"Presents vulnerability-constrained decoding to stop LLM code completion from generating vulnerable smart contracts, load-bearing for the vulnerability-detection direction.","marker":"[24]"},{"why":"Combines fine-tuning and LLM agents to mimic human auditors, load-bearing for the smart contract auditing direction.","marker":"[26]"},{"why":"Evaluates whether manual smart contract audits are still needed, used by the survey to position LLM audit tools.","marker":"[30]"},{"why":"Reports that LLMs classify DAO proposals effectively at 95 percent accuracy, load-bearing for the DAO governance direction.","marker":"[37]"},{"why":"Constructs a dataset of 16,427 DAOs to expose transparency and consistency gaps, load-bearing for the DAO governance analysis.","marker":"[38]"},{"why":"Introduces the Transformer attention architecture, which the survey credits as the technical basis of LLM processing power.","marker":"[58]"},{"why":"Defines the peer-to-peer electronic cash system whose decentralized ledger properties underwrite the blockchain side of the survey.","marker":"[59]"}],"fun_headline_variants":["Survey: Blockchain and LLMs can fix each other in six ways","Six bidirectional fixes for blockchain and large language models","Survey maps six synergies between blockchain and LLMs","Bidirectional survey: six integration points for blockchain and LLMs","LLMs and blockchain: six ways they strengthen each other"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The survey's map of the field depends on every cited system being real, correctly attributed, and described faithfully; that premise is not guaranteed by the paper itself, since PDPChain is cited with an unresolved '[?]' and one arXiv paper appears three times as [27], [31], and [35].","fun_headline_variants_meta":{"raw":{"variants":["Survey: Blockchain and LLMs can fix each other in six ways","Six bidirectional fixes for blockchain and large language models","Survey maps six synergies between blockchain and LLMs","Bidirectional survey: six integration points for blockchain and LLMs","LLMs and blockchain: six ways they strengthen each other"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001211,"raw_usage":{"total_tokens":4983,"prompt_tokens":938,"completion_tokens":4045,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":554,"completion_tokens_details":{"reasoning_tokens":3978}},"tokens_in":554,"tokens_out":4045,"duration_ms":31500,"temperature":1.0,"reasoning_tokens":3978,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T13:05:10.200714+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A concrete check is to resolve every citation in the survey against arXiv and publisher databases. If the PDPChain system named in Section II-A2 cannot be located because its '[?]' citation is unresolvable, and if the duplicated paper [27]/[31]/[35] is counted once, the survey's census of the field shrinks; enough failed references would mean the claimed six-direction map cannot be confirmed as a faithful representation of the literature.","supporting_citations":[{"cited_title":"Bitcoin: A peer-to-peer electronic cash sys- tem,","cited_arxiv_id":null,"evidence_quote":"Defines the peer-to-peer electronic cash system whose decentralized ledger properties underwrite the blockchain side of the survey."}],"review_version":1}