Pith. sign in

REVIEW 5 major objections 5 minor 47 references

An Exploratory Framework for Future SETI Applications: Detecting Generative Reactivity via Language Models

T0 review · 5 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read Whale and bird sounds outscore speech in a language-model SETI probe.

desk verdict The reactivity idea is new and worth discussing, but the whale/bird SIP result is confounded by token statistics without the very controls Section 4.2 says are needed. read the letter →

arxiv 2506.02730 v1 pith:QG6EUVBU submitted 2025-06-03 astro-ph.IM cs.CL

classification astro-ph.IMcs.CL
keywords SETIlanguagemodelsgenerativereactivitysemanticinductionpotentialspectraltokenizationsignaldetectionbioacousticsanomalyscreening
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 proposes that the search for extraterrestrial intelligence need not assume signals carry decodable messages. It argues that a language model's tendency to produce structured, language-like output when fed noise-like input can serve as a measure of latent structure in the data, a property it calls generative reactivity. To test this, the author feeds spectrally tokenized audio (human speech, whale song, birdsong, white noise) to GPT-2 small and scores the responses with a composite metric, Semantic Induction Potential. The reported result is that whale and bird vocalizations trigger substantially higher scores than white noise or human speech, which the paper reads as evidence that the model responds to internal organization rather than semantic content. If the claim holds, it would give SETI a complementary screening tool for archival data that current energy- and periodicity-based filters discard.

What carries the argument

The load-bearing object is the Semantic Induction Potential (SIP), a composite score built from four components: token-level entropy (how confidently the model picks each token), a syntax coherence score (inverse cross-entropy of the generated text), zlib compression gain (a proxy for internal redundancy), and a repetition penalty that discounts degenerate looping. Inputs are converted to prompts by a Semantic Triggering Detection Pipeline (STDP): audio is resampled, converted to log-mel spectrograms, flattened, quantized into discrete symbols by K-means clustering, and the resulting cluster indices are mapped to GPT-2 vocabulary tokens before generation. SIP is the quantity that separates the input classes, and the paper's argument stands or falls on whether this sequence preserves the structure it claims to measure.

What would settle it

Recompute SIP on temporally shuffled versions of the whale and bird clips, keeping the same token vocabulary and distribution but destroying sequential structure; if shuffled clips score as high as intact ones, the effect is an artifact of the token mapping rather than of the acoustic input. A second check is to rerun the pipeline with different cluster counts and random seeds to see whether the whale/bird advantage survives.

Watch

Extended reading notes

Core claim

On its own terms, the paper claims that generative reactivity is measurable and structure-sensitive: a 117M-parameter English-only language model, fed symbolic sequences derived from log-mel spectrograms of whale and bird vocalizations, produces outputs with lower entropy, higher syntactic coherence, and higher compressibility than it produces for white noise, while human speech sits between noise and animal sounds. The central suggestion is that the model acts as a detector of pattern density, responding to latent regularities in inputs that carry no conventional semantics. The paper accordingly frames detection as a reactivity test rather than a decoding task, and proposes that data segments which trigger language-like behavior are candidates for closer scrutiny in SETI pipelines.

Load-bearing premise

The pipeline assumes that flattening log-mel spectrograms, quantizing them with K-means, and mapping the cluster indices to GPT-2 tokens neither destroys nor imposes the regularities that drive the model's response, yet this assumption is never tested.

Editorial extensions

If this is right

  • Archival SETI data currently discarded as noise could be rescored with STDP, and segments with unexpectedly high SIP treated as candidates for follow-up.
  • SIP baselines could be built per target, stellar class, or sky region so that outliers become a new anomaly channel alongside narrowband and transient searches.
  • The effective detection window of a civilization would include the persistence of its structural imprint, not just the active transmission lifetime, supporting archive mining over real-time monitoring.
  • Reactivity-based screening could be distributed: lightweight language models running locally on volunteer computing platforms could compute SIP scores on unlabeled data segments.
  • Active SETI could shift from sending decodable messages to sending inputs designed to trigger structured symbolic behavior in a receiver, a strategy the paper calls cosmic linguistic seeding.

Reading between the lines

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

  • I would read the reported difference cautiously until the tokenization step is controlled: the same experiment run with random cluster labels, or with a randomly initialized GPT-2, would separate structure-sensitivity from pipeline artifacts.
  • SIP's components overlap with generic compressibility measures, so the metric may be detecting statistical regularity rather than anything language-specific; comparing it against a pure compression baseline on the raw spectrograms would clarify what the language model adds.
  • The framework could be tested immediately on known radio-frequency interference: if instrumental signatures score as high as whale song, false positives in real SETI data will be a serious obstacle.
  • A natural extension would be to apply STDP to non-audio modalities, such as time-series filterbank data or optical light curves, to see whether generative reactivity generalizes beyond sound.
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

5 major / 5 minor

Summary. The paper proposes a new exploratory framework, the Semantic Triggering Detection Pipeline (STDP), and a composite metric, Semantic Induction Potential (SIP), intended to detect latent structure in noise-like signals by prompting a language model (GPT-2 small) with tokenized spectral data. Four input categories are compared: human speech, humpback whale vocalizations, birdsong, and white noise. The reported mean SIP scores (Table 1) place whale and bird vocalizations above human speech and white noise, and the authors interpret this as evidence that language models can respond to structural regularities in non-semantic data. The paper frames this as a potential complement to conventional SETI screening and discusses speculative extensions such as 'cosmic linguistic seeding.'

Significance. If the central claim were established, the approach would be a genuinely novel contribution to SETI methodology: using generative reactivity rather than decodable content as a structure filter, with potential application to archival broadband data that is normally discarded. The paper is commendably explicit about its exploratory scope and about known artifacts in discretization (§4.2), and the component scores in Table 1 (entropy and syntax) show the same qualitative ordering as the composite SIP, which partially mitigates concern that the hand-set weights alone create the ranking. However, as presented, the central claim is not yet supported: the tokenization step is critically underspecified, no statistical significance tests are provided, the whale category is dominated by one recording, and no control separates genuine sequential structure from artifacts of token statistics. These gaps are load-bearing because the abstract's conclusion ('language models may detect latent structure') depends on the SIP ranking reflecting input structure rather than pipeline artifacts.

major comments (5)
  1. [§2.2.2 (Tokenization)] The tokenization step is not reproducible and its neutrality is unvalidated. The manuscript does not state the number of K-means clusters, the random seed, the initialization, the mapping rule from cluster indices to GPT-2 token IDs, or the token-to-vocabulary assignment, and no code is released. Because the central claim requires that the token sequence is a neutral projection of time-frequency structure, this underspecification is load-bearing: without validation, the higher SIP for whale and bird could be an artifact of the clustering or mapping rather than of the input signal.
  2. [Table 1 and §3.1] No statistical significance tests are reported. With only n=10 clips per class and large within-class scatter (whale SIP = 1.51±0.88, bird = 1.43±0.43), the claim that whale and bird vocalizations have 'higher SIP scores than white noise' is not statistically supported. The authors should report per-clip values and provide permutation tests, bootstrap confidence intervals, or an ANOVA/effect-size analysis to establish the ordering.
  3. [Table 2] The whale category result is dominated by a single recording. CI02 chunks 6–10 have mean SIP 2.34±0.36, while CI05 chunks 1–5 have mean SIP 0.68±0.16; pooling these into a single 'whale' mean of 1.51 overstates the category-level finding. The paper should treat recording identity as a factor, report per-source results separately, and temper the category-level interpretation accordingly.
  4. [§4.2 and Results] The manuscript acknowledges that discretization 'risks introducing structural artifacts' but does not provide the control needed to address the chief alternative explanation: low-diversity token prompts producing more repetitive, lower-entropy GPT-2 output. If whale and bird spectrograms map to few repeated cluster IDs, the resulting token sequence will be more predictable, mechanically raising (1−H_token), the syntax score, and the compression gain under the hand-set weights α=2.0, β=1.5, γ=1.0, δ=0.5. A token-permutation control (preserving the token histogram while destroying sequential order) and a phase-randomized audio control are necessary to distinguish response to sequential acoustic structure from response to token statistics alone.
  5. [§2.2.2 (SIP equation)] The SIP weights α, β, γ, and δ are set by hand with no sensitivity analysis or stated criterion for their selection. Since the composite ranking depends on these weights, the authors should demonstrate that the qualitative SIP ordering is robust to a weight sweep. The component scores in Table 1 already suggest that entropy and syntax carry the signal independently, but this robustness should be made explicit rather than left implicit.
minor comments (5)
  1. [§3.2] Typo: 'without replying on semantic content' should be 'without relying on semantic content.'
  2. [§2.2.2 (Data projection)] Preprocessing parameters are incomplete: the number of mel bands, window length, hop length, and any normalization details for the log-mel spectrograms are not given, and the GPT-2 generation parameters (maximum number of generated tokens, decoding settings beyond 'sampling disabled') are unspecified. Adding these details would substantially improve reproducibility.
  3. [Figure 2 and Figure 3] Figure 2 shows only a human-speech spectrogram; example spectrograms or token sequences for the whale, bird, and noise categories would help readers assess the structural differences that the pipeline is supposed to capture.
  4. [Table 1 note] The note 'Related indicators across species and data types' is vague; it should state explicitly that values are mean ± standard deviation across the ten 12-second clips per category.
  5. [§4.1] The terms 'semantic evocativity' and 'semantic neural evocativity' are introduced without operational definitions beyond the surrounding text. If these terms are meant to name the research direction, they should be defined concisely and used consistently to avoid terminological overload.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: SIP is an output-side metric with hand-set weights; the whale/bird ordering is an empirical result, not an identity, and the acknowledged tokenization risks are validity gaps rather than circular reductions.

full rationale

The paper's only formal construction is the SIP composite: SIP = alpha*(1-H_token) + beta*Syntax + gamma*Compression - delta*Repetition (Section 2.2.2). The weights alpha=2.0, beta=1.5, gamma=1.0, delta=0.5 are stated as hand-set priorities, not fitted to the observed SIP ordering; Table 1 also reports the four components separately, and each component shows the same whale/bird > human > noise ordering. Hence the composite result is not forced by construction: white noise could in principle have produced lower entropy, higher-compressibility output, but did not. The tokenization stage (K-means on flattened log-mel frames mapped to GPT-2 tokens) is under-specified, and Section 4.2 explicitly concedes that discretization 'risks introducing structural artifacts' and that instrumental features 'can lead to false-positive SIP activations.' Those are genuine validity threats (missing permutation or phase-randomized baselines), but they are experimental confounds, not circular reductions: the input-to-output map is not an identity, and no parameter of SIP is fitted to the target ranking. The discussion's definition of 'semantic evocativity' as the tendency to produce structured language is a naming convention that makes the claim that reactivity is measurable mildly tautological, but that definition is not load-bearing; the central empirical comparison is independent and could have contradicted the hypothesis. There are no load-bearing self-citations; the cited prior work (GPT-2, HuBERT, etc.) is external. Within the categories defined by this review, I find no step where a predicted quantity reduces to an input by construction.

Assumptions & free parameters 2 free parameters · 5 assumptions · 3 invented entities

The central empirical result is a measurement of a hand-defined composite metric. The SIP weights are free parameters, and the tokenization step has an unstated number of clusters. The framework rests on several domain assumptions about GPT-2 small's behavior, window length, and the faithfulness of K-means tokenization. The paper introduces conceptual labels, not physical entities. Overall, the contribution is more a proposal than a validated method.

free parameters (2)
  • SIP weights alpha, beta, gamma, delta = 2.0, 1.5, 1.0, 0.5
    Chosen by hand in Section 2.2.2 to prioritize entropy and syntax. No sensitivity analysis is performed, and the exact SIP values and relative rankings depend on these weights, though the component tables suggest the ordering is robust to some weight changes.
  • Number of K-means clusters (k) = not reported
    The tokenization step in Section 2.2.2 specifies K-means clustering but does not state the number of clusters, feature dimension, or random seed. This parameter directly determines the token vocabulary and hence the language model input.
assumptions (5)
  • domain assumption GPT-2 small has a higher threshold for producing structured language than larger models, making it a conservative testbed for detecting input structure.
    Invoked in Section 2.2 to justify the choice of GPT-2 small. No supporting measurement or citation is given for this specific behavioral claim.
  • domain assumption A 12-second window is sufficient to trigger structured responses from the language model.
    Stated in Section 2.2.1 with references to general LLM input-length studies, but not validated for acoustic token sequences.
  • domain assumption Log-mel spectrograms flattened and quantized by K-means preserve the structural regularities that drive language-like output.
    The entire pipeline depends on this projection. No control experiment with alternative projections (e.g., random linear projections or different spectral features) is reported.
  • ad hoc to paper Advanced extraterrestrial civilizations may favor low-power, phase-based signals that resemble background noise.
    This speculative hypothesis in Section 2.1 motivates the framework but is untestable in this paper and does not follow from the empirical results.
  • standard math K-means clustering is a valid and meaningful way to create discrete units from log-mel features.
    K-means is a standard algorithm, but its suitability for preserving temporal structure in acoustic signals is assumed without comparison to learned discrete units (e.g., HuBERT, wav2vec 2.0).
invented entities (3)
  • Semantic evocativity
    purpose: A term introduced in Section 4.1 to name the tendency of generative models to produce structured language from nonlinguistic input. It is used to frame the measured SIP scores as a property of the model-input pair.
    This is a label for the observed reactivity, not a new physical entity. It has no falsifiable handle outside the paper's own metric.
  • Cosmic linguistic seeding
    purpose: A proposed interstellar communication strategy that aims to trigger symbolic behavior in receivers rather than transmit decodable messages. Described in Sections 4.3 and 5.
    A speculative proposal with no evidence for feasibility or for any real extraterrestrial application.
  • Low-power phase-based extraterrestrial signals
    purpose: A hypothesized class of ETI signals that resemble noise but contain structured temporal features, used to justify the usefulness of the reactivity framework.
    Introduced in Section 2.1 as a motivational scenario. It is not detected or tested in this paper and has no empirical support.

how reviews work

0 comments
Cite this review

Pith. "Pith review of An Exploratory Framework for Future SETI Applications: Detecting Generative Reactivity via Language Models." pith.science (2026). https://pith.science/paper/QG6EUVBU

@misc{pith2026250602730,
  author       = {Pith},
  title        = {Pith review of: An Exploratory Framework for Future SETI Applications: Detecting Generative Reactivity via Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QG6EUVBU}},
  note         = {Machine review of arXiv:2506.02730}
}
read the original abstract

We present an exploratory framework to test whether noise-like input can induce structured responses in language models. Instead of assuming that extraterrestrial signals must be decoded, we evaluate whether inputs can trigger linguistic behavior in generative systems. This shifts the focus from decoding to viewing structured output as a sign of underlying regularity in the input. We tested GPT-2 small, a 117M-parameter model trained on English text, using four types of acoustic input: human speech, humpback whale vocalizations, Phylloscopus trochilus birdsong, and algorithmically generated white noise. All inputs were treated as noise-like, without any assumed symbolic encoding. To assess reactivity, we defined a composite score called Semantic Induction Potential (SIP), combining entropy, syntax coherence, compression gain, and repetition penalty. Results showed that whale and bird vocalizations had higher SIP scores than white noise, while human speech triggered only moderate responses. This suggests that language models may detect latent structure even in data without conventional semantics. We propose that this approach could complement traditional SETI methods, especially in cases where communicative intent is unknown. Generative reactivity may offer a different way to identify data worth closer attention.

Figures

Figures reproduced from arXiv: 2506.02730 by the authors.

Figure 1
Figure 1. Semantic Triggering Detection Pipeline (STDP). Audio inputs are preprocessed [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Example of spoken English exhibiting dense spectral energy across a wide frequency [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Mean scores for SIP and its components (entropy, syntax, compression, and repe [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

47 extracted references · 25 canonical work pages

  1. [1]

    & Werthimer, D

    Anderson, D.P., Cobb, J., Korpela, E., Lebofsky, M. & Werthimer, D. (2002) SETI@home: an experiment in public-resource computing. Communications of the ACM 45(11), 56--61. https://doi.org/10.1145/581571.581573

  2. [2]

    & Elliott, J

    Atwell, E. & Elliott, J. (2001) A corpus for interstellar communication. In: Rayson P, Wilson A, McEnery T, Hardie A, & Khoja S (eds.), Proceedings of CL2001: International Conference on Corpus Linguistics, 29 Mar--2 Apr 2001, Lancaster University, UK. UCREL, Lancaster University, pp. 31--39. ISBN: 1862201072. Available at: https://eprints.whiterose.ac.uk...

  3. [3]

    (1973) The Zoo Hypothesis

    Ball, J.A. (1973) The Zoo Hypothesis. Icarus 19, 347--349. https://doi.org/10.1016/0019-1035(73)90111-5

  4. [4]

    (2011) Dysonian approach to SETI: A fruitful middle ground? JBIS - Journal of the British Interplanetary Society 64, 156--161

    Bradbury, R.J., \' C irkovi\' c , M.M., & Dvorsky, G. (2011) Dysonian approach to SETI: A fruitful middle ground? JBIS - Journal of the British Interplanetary Society 64, 156--161

  5. [5]

    (2014) The search for extraterrestrial intelligence: Should we message ET? Journal of the British Interplanetary Society 67(1), 8--16

    Brin, D. (2014) The search for extraterrestrial intelligence: Should we message ET? Journal of the British Interplanetary Society 67(1), 8--16

  6. [6]

    & Askell, A

    Brown, T.B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G. & Askell, A. et al. (2020) Language models are few-shot learners. Advances in Neural Information Processing Systems 33, 1877--1901. https://doi.org/10.48550/arXiv.2005.14165

  7. [7]

    (2016) Alien mindscapes---A perspective on the search for extraterrestrial intelligence

    Cabrol, N.A. (2016) Alien mindscapes---A perspective on the search for extraterrestrial intelligence. Astrobiology 16, 661--676. https://doi.org/10.1089/ast.2016.1536

  8. [8]

    Edge Detection and Deep Learning Based SETI Signal Classification Method

    Chen, Z. & Haider, S.A. (2022) Edge detection and deep learning based SETI signal classification method. arXiv preprint arXiv:2203.15229. Available at: https://arxiv.org/abs/2203.15229

Show all 47 references
  1. [9]

    & Morrison, P

    Cocconi, G. & Morrison, P. (1959) Searching for interstellar communications. Nature 184, 844--846. https://doi.org/10.1038/184844a0

  2. [10]

    & Voien, J

    Cox, G.A., Egly, S., Harp, G.R., Richards, J., Vinodababu, S. & Voien, J. (2018) Classification of simulated radio signals using wide residual networks for use in the search for extra-terrestrial intelligence. arXiv preprint arXiv:1803.08624. Available at: https://arxiv.org/ab...

  3. [11]

    & Schulze-Makuch, D

    Crawford, I.A. & Schulze-Makuch, D. (2024) Is the apparent absence of extraterrestrial technological civilizations down to the zoo hypothesis or nothing? Nature Astronomy 8, 44--47. https://doi.org/10.1038/s41550-023-02134-2

  4. [12]

    & Veness, J

    Delétang, G., Ruoss, A., Duquenne, P.A., Catt, E., Genewein, T., Mattern, C., Grau-Moya, J., Wenliang, L.K., Aitchison, M., Orseau, L., Hutter, M. & Veness, J. (2023) Language modeling is compression. arXiv preprint arXiv:2309.10668. Available at: https://arxiv.org/abs/2309.10668

  5. [13]

    (1961) Project Ozma

    Drake, F.D. (1961) Project Ozma. Physics Today 14(4), 40--46. https://doi.org/10.1063/1.3057498

  6. [14]

    DeBoer, D

    Enriquez, J.E., Siemion, A., Foster, G., Gajjar, V., Hellbourg, G., Hickish, V., Isaacson, H., Price, D.C., Croft, S. DeBoer, D. et al. (2017) The Breakthrough Listen search for intelligent life: 1.1–1.9 GHz observations of 692 nearby stars. The Astrophysical Journal 849, 104....

  7. [15]

    (2016) The Galactic Club or Galactic Cliques? Exploring the limits of interstellar hegemony and the Zoo Hypothesis

    Forgan, D.H. (2016) The Galactic Club or Galactic Cliques? Exploring the limits of interstellar hegemony and the Zoo Hypothesis. International Journal of Astrobiology 16, 349--354. https://doi.org/10.1017/S1473550416000392

  8. [16]

    & Ritter, M

    Gemmeke, J.F., Ellis, D.P.W., Freedman, D., Jansen, A., Lawrence, W., Moore, R.C., Plakal, M. & Ritter, M. (2017) Audio set: An ontology and human-labeled dataset for audio events. IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), New Orleans, ...

  9. [17]

    & Marvel, K.B

    Gray, R.H. & Marvel, K.B. (2001) A VLA search for the Ohio State ``Wow''. The Astrophysical Journal 546, 1171--1177. https://doi.org/10.1086/318272

  10. [18]

    & Ellingsen, S

    Gray, R.H. & Ellingsen, S. (2002) A search for periodic emissions at the Wow locale. The Astrophysical Journal 578, 967--971. https://doi.org/10.1086/342646

  11. [19]

    Harp, G.R., Richards, J., Shostak, S., Tarter, J.C., Mackintosh, G., Scargle, J.D., Henze, C., Nelson, B., Cox, G.A., Egly, S. et al. (2019) Machine vision and deep learning for classification of radio SETI signals. arXiv preprint arXiv:1902.02426. Available at: https://arxiv....

  12. [20]

    & Choi, Y

    Holtzman, A., Buys, J., Du, L., Forbes, M. & Choi, Y. (2020) The curious case of neural text degeneration. In: Proceedings of the International Conference on Learning Representations (ICLR 2020). Available at: https://openreview.net/forum?id=rygGQyrFvH

  13. [21]

    & Mohamed, A

    Hsu, W.-N., Bolte, B., Tsai, H.Y.-H., Lakhotia, K., Salakhutdinov, R. & Mohamed, A. (2021) HuBERT: Self-supervised speech representation learning by masked prediction of hidden units. IEEE/ACM Transactions on Audio, Speech, and Language Processing 29, 3451--3460. https://doi.o...

  14. [22]

    & Fung, P

    Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., Ishii, E., Bang, Y.J., Madotto, A. & Fung, P. (2023) Survey of hallucination in natural language generation. ACM Computing Surveys 55(1), 1--38. https://doi.org/10.1145/3571730

  15. [23]

    (2023) XC763801: Phylloscopus trochilus birdsong recording

    Kirkeby, J. (2023) XC763801: Phylloscopus trochilus birdsong recording. Available at: https://www.xeno-canto.org/763801. Licensed under CC BY-NC-SA 4.0

  16. [24]

    & Khudanpur, S

    Ko, T., Peddinti, V., Povey, D. & Khudanpur, S. (2015) Audio augmentation for speech recognition. Proceedings of Interspeech 2015, 3586--3589. https://doi.org/10.21437/Interspeech.2015-711

  17. [25]

    (1994) The Tantalizing “WOW!” Signal

    Kraus, J.D. (1994) The Tantalizing “WOW!” Signal. NRAO/AUI Archives. https://www.nrao.edu/archives/items/show/3684

  18. [26]

    & Goldberg, Y

    Levy, M., Jacoby, A. & Goldberg, Y. (2024) Same task, more tokens: The impact of input length on the reasoning performance of large language models. In: Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pp. 15339--...

  19. [27]

    Available at: https://librivox.org/pages/about-librivox/

    LibriVox (2024) About LibriVox. Available at: https://librivox.org/pages/about-librivox/

  20. [28]

    & Liang, P

    Liu, N.F., Lin, K., Hewitt, J., Paranjape, A., Bevilacqua, M., Petroni, F. & Liang, P. (2024) Lost in the middle: How language models use long contexts. Transactions of the Association for Computational Linguistics, 12:157--173. https://doi.org/10.1162/tacl_a_00638

  21. [29]

    & Pardo, B

    Liu, X., Seetharaman, P. & Pardo, B. (2020) Model selection for deep audio source separation via clustering. Proceedings of the DCASE 2020 Workshop, 89--93. https://doi.org/10.48550/arXiv.1910.12626

  22. [30]

    Ma, P.X., Ng, C., Rizk, L., Croft, S., Siemion, A.P.V., Brzycki, B., Czech, D., Drew, J., Gajjar, V., Hoang, J. et al. (2023) A deep-learning search for technosignatures from 820 nearby stars. Nature Astronomy 7, 492--502. https://doi.org/10.1038/s41550-022-01872-z

  23. [31]

    (2015) Design for minimum energy in interstellar communication

    Messerschmitt, D.G. (2015) Design for minimum energy in interstellar communication. Acta Astronautica 107, 20--39. https://doi.org/10.1016/j.actaastro.2014.11.007

  24. [32]

    (2023) Mistral 7B

    Mistral, A.I. (2023) Mistral 7B. Available at: https://mistral.ai/news/announcing-mistral-7b/

  25. [33]

    Navy (2021) SanctSound Raw Passive Acoustic Data

    NOAA Office of National Marine Sanctuaries and U.S. Navy (2021) SanctSound Raw Passive Acoustic Data. NOAA National Centers for Environmental Information. https://doi.org/10.25921/saca-sp25

  26. [34]

    arXiv preprint arXiv:2303.08774

    OpenAI (2023) GPT-4 technical report. arXiv preprint arXiv:2303.08774. Available at: https://arxiv.org/abs/2303.08774

  27. [35]

    & Schafer, R.W

    Oppenheim, A.V. & Schafer, R.W. (2009) Discrete-Time Signal Processing (3nd ed.). Prentice Hall

  28. [36]

    & Sutskever, I

    Radford, A., Wu, J., Child, R., Luan, D., Amodei, D. & Sutskever, I. (2019) Language models are unsupervised multitask learners. OpenAI Technical Report. Available at: https://api.semanticscholar.org/CorpusID:160025533

  29. [37]

    (1948) A mathematical theory of communication

    Shannon, C.E. (1948) A mathematical theory of communication. Bell System Technical Journal 27(3), 379--423

  30. [38]

    & Worden, S.P

    Sheikh, S.Z., Smith, S., Price, D.C., DeBoer, D., Lacki, B.C., Czech, D.J., Croft, S., Gajjar, V., Isaacson, H., Lebofsky, M., MacMahon, D.H.E., Ng, C., Perez, K.I., Siemion, A.P.V., Webb, C.I., Zic, A., Drew, J. & Worden, S.P. (2021) Analysis of the Breakthrough Listen signal...

  31. [39]

    (2011) Project Ozma: The birth of observational SETI

    Shuch, H.P. (2011) Project Ozma: The birth of observational SETI. In: Searching for Extraterrestrial Intelligence. The Frontiers Collection. Springer, Berlin, Heidelberg. Available at: https://doi.org/10.1007/978-3-642-13196-7_2

  32. [40]

    & van Leeuwen, J

    Siemion, A.P.V., Bower, G.C., Foster, G., McMahon, P.L., Wagner, M.I., Werthimer, D., Backer, D., Cordes, J. & van Leeuwen, J. (2012) The Allen Telescope Array Fly's Eye Survey for fast radio transients. The Astrophysical Journal, 744(2):109. https://doi.org/10.1088/0004-637X/...

  33. [41]

    & Zic, A

    Smith, S., Price, D.C., Sheikh, S.Z., Czech, D.J., Croft, S., DeBoer, D., Gajjar, V., Isaacson, H., Lacki, B.C., Lebofsky, M., MacMahon, D.H.E., Ng, C., Perez, K.I., Siemion, A.P.V., Webb, C.I., Drew, J., Worden, S.P. & Zic, A. (2021) A radio technosignature search towards Pro...

  34. [42]

    (2001) The search for extraterrestrial intelligence (SETI)

    Tarter, J.C. (2001) The search for extraterrestrial intelligence (SETI). Annual Review of Astronomy and Astrophysics 39, 511--548. https://doi.org/10.1146/annurev.astro.39.1.511

  35. [43]

    & Engel, T

    Tawakuli, A., Havers, B., Gulisano, V., Kaiser, D. & Engel, T. (2024) Time-series data preprocessing: A survey and an empirical analysis. Journal of Engineering Research (in press). Available at: https://doi.org/10.1016/j.jer.2024.02.018

  36. [44]

    Touvron, H., Lavril, T., Izacard, G., Martinet, X., Lachaux, M.A., Lacroix, T., Rozi\` e re, B., Goyal, N., Hambro, E., Azhar, F. et al. (2023) LLaMA: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971. Available at: https://doi.org/10.48550/arXiv.2...

  37. [45]

    E.H., Hashimoto, T., Vinyals, O., Liang, P., Dean, J

    Wei, J., Tay, Y., Bommasani, R., Raffel, C., Zoph, B., Borgeaud, S., Yogatama, D., Bosma, M., Zhou, D., Metzler, D., Chi. E.H., Hashimoto, T., Vinyals, O., Liang, P., Dean, J. & Fedus, W. (2022) Emergent abilities of large language models. Transactions on Machine Learning Rese...

  38. [46]

    & Oman-Reagan, M.P

    Wright, J.T. & Oman-Reagan, M.P. (2018) Visions of human futures in space and SETI. International Journal of Astrobiology 17, 177--188. https://doi.org/10.1017/S1473550417000222

  39. [47]

    & Dai, L.R

    Xu, H.T., Zhang, J. & Dai, L.R. (2022) Differential time-frequency log-mel spectrogram features for infant cry recognition. Proceedings of Interspeech 2022, 1963--1967. ISCA. https://doi.org/10.21437/Interspeech.2022-18

Pith tools

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