Pith. sign in

REVIEW 6 minor 1 cited by

Generative AI at the Crossroads: Light Bulb, Dynamo, or Microscope?

T0 review · 0 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper claims that generative AI combines the long-lived productivity characteristics of a general-purpose technology and an invention of a method of invention, and that it will therefore raise the level of labor productivity.

desk verdict A careful, honestly hedged qualitative case that genAI is both GPT and IMI; the classification is plausible but the productivity forecast is a leap of analogy, not a derivation. read the letter →

arxiv 2505.14588 v4 pith:HSEZYR3B submitted 2025-05-20 econ.GN q-fin.EC

Martin Baily , David Byrne , Aidan Kane , Paul Soto This is my paper · ORCID
classification econ.GNq-fin.EC
keywords generativeAIproductivitygeneral-purposetechnologiesinventionsofmethodsinventionGPTIMItechnologydiffusioneconomicgrowth
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

This paper asks whether generative AI will behave, for productivity, like the light bulb, the electric dynamo, or the compound microscope. Its answer: genAI is on track to be both a general-purpose technology (GPT) and an invention of a method of invention (IMI), the two classes that historically give longer-lived productivity effects. On the GPT side, the evidence points to broad diffusion potential, abundant knock-on innovations, and sustained improvement in the core technology. On the IMI side, genAI is already used as an observational, analytical, communication, and organizational tool in research. If the classification holds, the expected outcome is a meaningful rise in the level of labor productivity, with the size of the growth-rate effect depending on how fast adoption and complementary investment arrive.

What carries the argument

The organizing device is a three-way taxonomy of innovations: the light bulb (one-time level gain after diffusion), the general-purpose technology (GPT: repeated waves of adoption and knock-on innovation as the core technology improves), and the invention of a method of invention (IMI: cheaper discovery and R&D). GenAI is located at the intersection of the second and third classes. The argument is carried by checking genAI against the three GPT criteria—diffusion, knock-on innovation, and ongoing core innovation—and against the four IMI channels—observation, analysis, communication, and organization—using survey, field-experiment, patent, earnings-call, and prompt-usage evidence. The classification does the work: once genAI is shown to belong to both historical classes, the historical growth properties of those classes transfer to genAI.

What would settle it

Observe the next several years of enterprise adoption and profitability data: if more than 80 percent of adopting firms continue to report no tangible effect on earnings before interest and taxes, and the share of job postings requiring AI skills stays near its current single-digit level while AI use remains concentrated in a few occupations, the GPT part of the claim would be contradicted. On the IMI side, track research-sector outputs: if patents per researcher and the measured efficiency of R&D do not rise as genAI tools spread through scientific work, the IMI channel would be contradicted.

Watch

Extended reading notes

Core claim

The paper's central claim is that generative artificial intelligence has, simultaneously, the defining characteristics of a general-purpose technology and of an invention of a method of invention. A GPT is widely adopted, generates abundant follow-on innovations, and keeps improving; an IMI raises the efficiency of research by improving observation, analysis, communication, or organization. GenAI qualifies on the GPT criteria through its rapid diffusion into work, the wave of interfaces, copilots, robots, and agentic systems built around it, and falling cost per unit of capability. It qualifies as an IMI because it accelerates the research process itself, from image enhancement and text analysis to drafting, digital twins, and early research agents. The paper therefore expects genAI to raise the level of productivity relative to a counterfactual economy without it, while cautioning that the growth-rate effect will be damped by slow diffusion and the need for complementary investment.

Load-bearing premise

The whole forecast leans on the analogy that genAI's future will resemble the historical track record of earlier GPTs and IMIs; the classification could be right and the productivity effect still small if profitable use at scale fails to materialize or diffusion stalls outside large firms.

Editorial extensions

If this is right

  • If genAI is both a GPT and an IMI, the productivity level should rise over time without waiting for artificial general intelligence.
  • The growth-rate boost will be spread over years, possibly decades, because complementary reorganization and investment are slow.
  • The IMI channel implies cheaper research and faster discovery, so the payoff may show up first in innovation indicators such as patents and R&D efficiency before aggregate labor productivity.
  • Profitability at scale, not technical capability, is the binding constraint; the technology can be a GPT or IMI and still disappoint if profitable applications are slow to emerge.
  • The classification implies that current modest macro productivity data are not evidence against a future effect, since GPT effects historically arrive with long lags.

Reading between the lines

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

  • If the paper is right, the first detectable macro signal should come from research-sector indicators rather than aggregate labor productivity: faster discovery per dollar of R&D, rising patent counts in AI-adjacent fields, and productivity gains inside scientific workflows.
  • The framework suggests a natural experiment: compare sectors with high genAI exposure in research tasks, such as computing and life sciences, against low-exposure sectors; the IMI hypothesis predicts an earlier and sharper productivity response in the high-exposure group.
  • A testable extension would map falling compute costs onto the timing of diffusion; if the ongoing-core-innovation criterion is correct, price declines per unit of AI capability should continue and adoption should follow the historical GPT pattern.
  • The classification also implies that policy attention should focus on the speed of complementary investment in data, training, reorganization, and energy, rather than on capability milestones such as AGI.
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

0 major / 6 minor

Summary. The paper asks whether generative AI (genAI) is best understood as a conventional \"light bulb\" invention, a general-purpose technology (GPT), or an invention of a method of invention (IMI). Drawing on a broad set of public indicators—Census and McKinsey adoption surveys, job postings, patent data, the Anthropic Economic Index, and earnings calls—the authors argue that genAI has features of both a GPT (wide potential diffusion, abundant knock-on innovation, ongoing core improvement) and an IMI (efficiency gains in observation, analysis, communication, and organization of research). They conclude that this combination is an encouraging sign that genAI will raise the level of labor productivity, while repeatedly acknowledging that adoption is still modest, profitability at scale is unproven, and the timeline may be long.

Significance. If the qualitative classification holds, the paper offers a useful framework for situating early evidence on genAI within the long-run productivity literature. Its main strengths are the breadth of indicators, the explicit handling of conflicting data (for example, the 9% Census versus 72% McKinsey adoption figures), and its transparency about important limitations: the paper notes the retraction of Toner-Rodgers (2024), calls net writing efficiency \"an open question,\" and concedes that profitability at scale is \"the ultimate test.\" The paper is not a formal model or a quantitative forecast, but as a considered qualitative assessment it is a valuable contribution to the policy and research discussion. The forward-looking conclusion is explicitly hedged, which mitigates the concern that the historical GPT/IMI taxonomy, populated ex post by successful technologies, is being used as an uncalibrated predictor.

minor comments (6)
  1. [Abstract and Section 5] The sentence \"it is reasonable to expect genAI will have a noteworthy impact on productivity\" is a forward-looking extrapolation. Because the paper's own evidence shows weak current diffusion (roughly 9% BTOS firm adoption, about 4% AI-related job postings) and limited profitability (McKinsey 2025b: over 80% of genAI-using firms see no tangible EBIT impact), please add an explicit conditional: the expectation holds if genAI follows the historical adoption and profitability trajectory of successful GPTs and IMIs, and would be much weaker if diffusion stalls or profit rates remain low. This small addition would align the conclusion more precisely with the paper's extensive caveats.
  2. [Section 4.2] The claim that AI patents \"surged when the use of genAI became practical\" is not supported by Figure 11, which shows the rise beginning around 2018, before practical genAI applications such as ChatGPT. The surge coincides with the introduction of the Transformer architecture and the broader deep-learning wave. Please rephrase to avoid the implication that the patent increase reflects post-2022 genAI deployment.
  3. [Table 3] The annualized rates of change in Table 3 appear arithmetically inconsistent. For price per TFLOP, the ratio of 349/0.3 to 299/15.1 is about 58.7, implying an annual decline of roughly 21%, not 24%. Similarly, TFLOP growth over 17 years (15.1/0.3 = 50.3) implies about 26% annual growth, not 23%. Please verify the calculations or clarify the method used.
  4. [Section 1 and Section 4.2] The phrase \"substantial evidence\" is stronger than the evidence presented. The authors themselves document major headwinds: 9% BTOS adoption, over 80% of firms with no EBIT impact, and only 0.9% of Anthropic prompts involving scientific-discovery tasks. Consider using \"suggestive evidence\" or \"indicative evidence\" in the abstract and conclusion to match the paper's careful internal hedging.
  5. [Section 3.1] The case-study subsection relies heavily on the authors' own Brookings publications (Baily and Kane 2025a,b; Kane and Baily 2025a,b). Please add one or two sentences describing the data sources and methods used in those case studies, so that readers can assess whether they are independent of the indicators already analyzed in this paper.
  6. [Various] Minor editorial issues: \"Solow-Swann\" in Section 4 should be \"Solow-Swan\"; there is inconsistent capitalization of \"genAI\" and \"GenAI\" across the manuscript; and the reference list includes a few formatting inconsistencies (e.g., \"Akcigit and Van Reenan\" versus \"Akcigit and Van Reenen\"). These do not affect the substance.

Circularity Check

0 steps flagged · score 2.0 of 10

No material circularity; the GPT/IMI classification is assessed against external indicators and the productivity conclusion is an explicit analogy, not a tautology.

full rationale

The paper's central claim is a qualitative classification of genAI into two externally defined categories, GPT and IMI, evaluated with a broad set of independent indicators: BTOS and McKinsey adoption surveys, Lightcast job postings, field experiments on writing, coding, and customer service, USPTO patent data, Anthropic Economic Index prompt shares, and earnings-call mentions. The definitions are taken from prior literature (Lipsey, Carlaw, and Bekar 2005; Cockburn, Henderson, and Stern 2019) rather than being fitted to the productivity outcome being predicted. The main predictive step—'Because both GPTs and IMIs promote productivity growth for extended periods, it is reasonable to expect genAI will have a noteworthy impact on productivity'—is an analogical forecast, not a derivation: the paper does not define genAI's GPT/IMI status in terms of the productivity outcome, and it repeatedly concedes the key uncertainties. Section 3.4 states that 'The ultimate test of whether genAI is a GPT will be the profitability of genAI use at scale,' Section 4 calls the net efficiency of genAI writing support 'an open question,' and footnote 48 explicitly retracts the Toner-Rodgers (2024) RCT evidence after its veracity was questioned. These disclosures weigh against circularity. The only in-house references are four Brookings case studies (Baily and Kane 2025a,b; Kane and Baily 2025a,b) used to illustrate sector-level adoption; these are peripheral to the classification, which is supported mainly by external sources, so they are not load-bearing. The self-citations are minor and non-essential, and the central claim retains independent evidentiary content.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The central claim rests on no fitted parameters; the listed thresholds are hand-chosen data-processing choices used in figures. The main assumptions are the validity of the GPT/IMI taxonomy and the analogical inference from historical technologies to genAI, plus several data representativeness assumptions that the authors partly acknowledge.

free parameters (2)
  • AI patent probability threshold = 93% probability
    Chosen from Pairolero et al. 2025 as the 'most conservative' threshold to identify AI-related patents in Figure 11. It is a hand-selected threshold, though not fitted to the paper's conclusion.
  • Research-context word window = 10 words
    Hand-chosen window in the earnings-call analysis for Figure 13 to classify a mention as research-related. No robustness check is shown.
assumptions (5)
  • domain assumption The GPT taxonomy (widespread adoption, knock-on innovation, ongoing core innovation) and the IMI taxonomy (observation, analysis, communication, organization) are valid and jointly useful categories for classifying technologies.
    Adopted from Lipsey et al. (2005), David (1990), and Whitehead (1925); the paper applies these to genAI without testing whether the categories are predictive of productivity outcomes. Sections 3 and 4.
  • domain assumption Historical GPT and IMI examples are a reliable guide to the future productivity effects of genAI.
    The conclusion in Section 5 infers a likely noteworthy effect on the productivity level from the fact that past GPTs and IMIs had such effects. This analogical inference is not independently validated.
  • domain assumption Field studies of genAI productivity effects are internally valid and generalize beyond their samples.
    Table 1 cites field experiments in call centers, coding, and writing as evidence of genAI productivity gains. The paper itself notes only 'green shoots' and that over 80% of firms report no EBIT impact.
  • domain assumption The Anthropic Economic Index prompt data are representative enough of genAI use in research to support the IMI conclusion.
    Section 4.1 relies on Claude conversations mapped to O*NET tasks; the paper notes in the Table 7 note that patterns of Anthropic use may not be representative of all genAI.
  • ad hoc to paper Generative models can form genuine world models of phenomena, enabling scientific discovery.
    Section 4, in the discussion of genAI as an analytical tool, invokes the OthelloGPT 'emergent world model' debate. If LLMs are only 'bags of heuristics', the IMI contribution to discovering fundamental laws is weakened. The paper leaves this unresolved.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Generative AI at the Crossroads: Light Bulb, Dynamo, or Microscope?." pith.science (2026). https://pith.science/paper/HSEZYR3B

@misc{pith2026250514588,
  author       = {Pith},
  title        = {Pith review of: Generative AI at the Crossroads: Light Bulb, Dynamo, or Microscope?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HSEZYR3B}},
  note         = {Machine review of arXiv:2505.14588}
}
read the original abstract

With the advent of generative AI (genAI), the potential scope of artificial intelligence has increased dramatically, but the future effect of genAI on productivity remains uncertain. The effect of the technology on the innovation process is a crucial open question. Some inventions, such as the light bulb, temporarily raise productivity growth as adoption spreads, but the effect fades when the market is saturated; that is, the level of output per hour is permanently higher but the growth rate is not. In contrast, two types of technologies stand out as having longer-lived effects on productivity growth. First, there are technologies known as general-purpose technologies (GPTs). GPTs (1) are widely adopted, (2) spur abundant knock-on innovations (new goods and services, process efficiencies, and business reorganization), and (3) show continual improvement, refreshing this innovation cycle; the electric dynamo is an example. Second, there are inventions of methods of invention (IMIs). IMIs increase the efficiency of the research and development process via improvements to observation, analysis, communication, or organization; the compound microscope is an example. We show that GenAI has the characteristics of both a GPT and an IMI -- an encouraging sign that genAI will raise the \textit{level} of productivity. Even so, genAI's contribution to productivity \textit{growth} will depend on the speed with which that level is attained and, historically, integrating revolutionary technologies into the economy is a protracted process.

Figures

Figures reproduced from arXiv: 2505.14588 by the authors.

Figure 1
Figure 1. AI Benchmark Performance Note: The “human baseline” concept used varies by task. For more challenging tasks, the baseline tends to reflect expert-level performance. Source: Reproduced with permission from the 2024 AI Index Report, Stanford Institute for Human￾centered Artificial Intelligence. potential for genAI to spur an information technology (IT)-fueled produc￾tivity boom comparable to the late 1990s and early 2… view at source ↗
Figure 2
Figure 2. Indicators of Interest in GenAI (a) GenAI Mobile App Downloads (b) Web Searches for AI Note: Apps are ChatGPT, Claude, DeepSeek, and Perplexity. Includes Android and iOS. Android down￾load information not available for China. Does not account for access via application program interface. Web searches include related terms in Google’s “AI” topic. Source: appfigures; Google Trends. substantial evidence that genAI is b… view at source ↗
Figure 3
Figure 3. Indicators of U.S. AI-Related Investment [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: AI Use Over Time (a) Census BTOS (United States) (b) McKinsey (Global) Note: For BTOS, respondents were asked about AI use in producing goods or services during the past two weeks and anticipated in the next six months. For McKinsey, respondents were asked if they “use…
Figure 5
Figure 5. Figure 5: AI-Related Job Postings Note: Jobs classified using AI-related terms found in job descriptions, as described in Acemoglu et al. (2020). List of AI-related terms updated by Lightcast. Source: Lightcast. Case study evidence x The information sector has adopted genAI rapi…
Figure 6
Figure 6. Figure 6: Number of Parameters and Training Dataset Size [PITH_FULL_IMAGE:figures/full_fig_p021_6.png]
Figure 7
Figure 7. Figure 7: Open-Source AI Models Source: Hugging Face. In addition to training and inference innovations, advances in performance have come from novel model concepts. Mamba, introduced in 2023, achieved subquadratic-time sequence modelling by avoiding the pairwise comparison amon…
Figure 8
Figure 8. Figure 8: Price Indices of GPU Improvements (a) Price per TFLOP Note: Price per TFLOPS (trillion floating-point operations per second). The blue line represents the best fit line for NVIDIA GPUs, and the orange line represents the best fit line for AMD GPUs. Source: TechPowerUp.…
Figure 9
Figure 9. Figure 9: Progress on Moore’s Law: Two Perspectives [PITH_FULL_IMAGE:figures/full_fig_p031_9.png]
Figure 10
Figure 10. Figure 10: Indicators of Power Demand and Supply (a) Data Center Efficiency (b) U.S. Electricity Generation Note: Data center efficiency indicator is gigaflops per watt of supercomputers labeled “industry” or “vendor”. Source: Top500.org for efficiency. U.S. Energy Information A…
Figure 11
Figure 11. Figure 11: AI Mentions in Scientific Patents Source: Artificial Intelligence Patent Dataset (2023), U.S. Patent Office. GenAI Prompts Handa et al. (2025) provide a rich set of information on actual genAI use in their Anthropic Economic Index (AEI), a useful com￾plement to the de…
Figure 12
Figure 12. Figure 12: GenAI Automation vs. Augmentation in Researcher Roles [PITH_FULL_IMAGE:figures/full_fig_p045_12.png]
Figure 13
Figure 13. Figure 13: Mentions of AI Usage for Research in Conference Calls [PITH_FULL_IMAGE:figures/full_fig_p048_13.png]

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. Towards a future space-based, highly scalable AI infrastructure system design

    cs.DC 2025-11 conditional novelty 5.0 of 10

    Space-based AI compute is argued feasible via close-formation laser-linked satellites, radiation-survivable TPUs, and launch costs projected below $200/kg by the mid-2030s.

Reference graph

Works this paper leans on

16 extracted references · 11 canonical work pages · cited by 1 Pith paper

  1. [9]

    Domain Generalization: A Survey

    “Domain Generalization: A Survey.”IEEE Transactions on Pat- tern Analysis and Machine Intelligence45 (4): 4396–4415. 72 A Definitions of AI We illustrate the varied use of the term “artificial intelligence” by dis- cussing four influential definitions. Alan Turing devised a broad, conceptual definition—the “Turing test”—to determine if a system was indist...

    work page 1950

  2. [62]

    Chinese room argu- ment

    This set of definitions is far from exhaustive. See the discussion in Filippucci et al. (2024) for a definition of scope for AI usefully grounded in a production function framework as well as references to other definitions. The OECD, for example, has codified this definition: “An AI system is a machine-based system that, for explicit or implicit objectiv...

    work page 2024

  3. [63]

    Artificial Intelligence

    He chose the term “Artificial Intelligence” to distinguish the field from “automata theory”—a branch of computer science focused on rule-based mathematical models of computation—and “cybernetics”—a field focused on control systems, feedback, and com- munication in machines and living things. 74 The Dartmouth AI definition is far broader than the Turing te...

    work page 2019

  4. [64]

    For more on the conference, see Nilsson (2009), Wooldridge (2021), and Olson (2024)

    work page 2009

  5. [65]

    Logic Theorist

    Indeed, Andrey Markov identified language as a use for his mathematical structures as early as 1906 (Markov 2006). 78 B.1 Early AI Research Following the Dartmouth project, AI research developed models distin- guished along several dimensions (table 9 on the previous page). •Symbolic AIencoded a system of explicit rules in computer programs. For example, ...

    work page 1958

  6. [66]

    expert systems

    Strictly speaking, some AI models, such as the “expert systems” described below, are neither generative or discriminative, so our classification scheme is not exhaustive. 79 reinforcement learning, interacting with the environment to refine the model. Others usepredictive learning, where the system is trained in advance of use. Predictive learning primari...

    work page 1958

  7. [67]

    Landmark AI Models: The Transformer,

    Computer scientists have wrestled with this word sense disambiguation problem since the 1950s. Bar-Hillel (1960) in discussing the prospects for fully automatic high-quality translation, offered this assessment: “What such a suggestion amounts to, if taken se- 82 A major breakthrough in addressing this shortcoming came with the in- troduction of the Trans...

    work page 1960

  8. [68]

    The final ‘T’ in BERT stands for ‘Transformer’ (bidirectional encoder representations from transformers) 83

    Particularly important was the introduction of the BERT model the following year (Devlin 2018). The final ‘T’ in BERT stands for ‘Transformer’ (bidirectional encoder representations from transformers) 83

    work page 2018

Show all 16 references
  1. [1991]

    Adaptive Mixtures of Local Experts

    “Adaptive Mixtures of Local Experts.”Neural Computation3 (1): 79–87. James, Conrad D, James B Aimone, Nadine E Miner, Craig M Vineyard, Fredrick H Rothganger, Kristofor D Carlson, Samuel A Mulder, et al

  2. [2017]

    A Historical Survey of Algorithms and Hardware Architectures for Neural-Inspired and Neuromorphic Computing Applications

    “A Historical Survey of Algorithms and Hardware Architectures for Neural-Inspired and Neuromorphic Computing Applications.”Bio- logically Inspired Cognitive Architectures19:49–64. 61 Jiang, Albert Q, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot,...

    2023 arXiv

  3. [2019]

    Deep Active Learning for Efficient Training of a Lidar 3d Object Detector

    “Deep Active Learning for Efficient Training of a Lidar 3d Object Detector.” In2019 IEEE Intelligent Vehicles Symposium (IV),667–674. IEEE. Ferrucci, David A. 2012. “Introduction to “this is watson”.”IBM Journal of Research and Development56 (3.4): 1–1. 58 Filippucci, Francesc...

    2012 arXiv

  4. [2020]

    Baricitinib as Potential Treatment for 2019-nCoV Acute Respi- ratory Disease

    “Baricitinib as Potential Treatment for 2019-nCoV Acute Respi- ratory Disease.”The Lancet395 (10223): e30–e31. Romer, Paul M. 1994. “The Origins of Endogenous Growth.”Journal of Economic Perspectives8 (1): 3–22. Rosenblatt, Frank. 1958. “The Perceptron: A Probabilistic Model f...

    2019

  5. [2021]

    Are We Learning Yet? A Meta Review of Evaluation Failures across Machine Learning

    “Are We Learning Yet? A Meta Review of Evaluation Failures across Machine Learning.” InThirty-fifth Conference on Neural Infor- mation Processing Systems Datasets and Benchmarks Track (Round 2). Lino, Giro. 2024. “Nvidia GPU Evolution: From GeForce to AI Powerhouse.” girolino....

    2024 arXiv

  6. [2022]

    Digital Twins for Materials

    “Digital Twins for Materials.”Frontiers in Materials9:818535. Kamiya, George, and Vlad C. Coroam˘ a. 2025. “Data Centre Energy Use: Critical Review of Models and Results.”IEA 4E TCP Efficient, Demand Flexible Networked Appliances (EDNA). Kane, Aidan, and Martin Baily. 2025a.AI...

    2025 arXiv

  7. [2023]

    Efficiently Scaling Transformer Inference

    “Efficiently Scaling Transformer Inference.”Proceedings of Ma- chine Learning and Systems5:606–624. Porter, Michael E, and Scott Stern. 2001. “Innovation: location matters.” MIT Sloan Management Review. Radford, Alec, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, and Ilya...

    2001 arXiv

  8. [2025]

    Using AI-based Coding Assistants in practice: State of affairs, perceptions, and ways forward

    “Using AI-based Coding Assistants in practice: State of affairs, perceptions, and ways forward.”Information and Software Technology 178:107610. Serradilla, Oscar, Ekhi Zugasti, Jon Rodriguez, and Urko Zurutuza. 2022. “Deep Learning Models for Predictive Maintenance: A Survey, ...

    2022 arXiv

Pith tools

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