Pith. sign in

REVIEW 3 major objections 5 minor 62 references

Predicting Emergent Capabilities by Finetuning

T0 review · 3 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Fitting an emergence law to finetuned small models predicts few-shot emergence within 0.1 nats and up to 4.3x the FLOPS in advance.

desk verdict Genuinely new emergence-prediction result, well-executed within a single OpenLLaMA V1 run; the cross-run transfer that would make it a true GPT-N+1 predictor is the one thing not yet demonstrated. read the letter →

arxiv 2411.16035 v1 pith:FSTZXMWN submitted 2024-11-25 cs.LG cs.CL

Charlie Snell , Eric Wallace , Dan Klein , Sergey Levine This is my paper · ORCID
classification cs.LGcs.CL
keywords emergentcapabilitiesemergencepredictionscalinglawsfinetuningpretraininglosslanguagemodelslawfew-shotprompting
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

Emergent capabilities are abilities that appear suddenly once a language model passes some scale threshold, and pre-emergence models look random on the task, so predicting the threshold has been open. This paper tries to predict that threshold using only small, pre-emergence models. Its central finding is that task-specific finetuning shifts the emergence point toward weaker models, and the amount of finetuning data controls how far it shifts. Fitting a parametric "emergence law" to those shifts and extrapolating to the few-shot limit predicts the true few-shot emergence point within 0.1 nats of pretraining loss on all four benchmarks studied, and in the best cases up to 4.3x the pretraining FLOPS before the emergence point is reached. If this transfers to future model families, capability forecasts could be made without training the large models first.

What carries the argument

The load-bearing object is the emergence law, a parametric function $E_\theta(D)=k\log(D)^\alpha + C$ that gives the ReLU elbow (the point of emergence, measured in pretraining loss) as a function of finetuning data amount $D$. It is embedded in a ReLU performance model $\mathrm{Perf}=A\max(E_\theta(D)-L(M),0)+B+\Delta\mathbf{1}_{\mathrm{finetuned}}$ and fit to raw accuracy of checkpoints at pretraining loss $L(M)$; the low-data limit $D_0$ near the few-shot count converts finetuning measurements into a few-shot prediction. Pretraining loss acts as the capability index, so intermediate checkpoints from one run stand in for independently trained models of different scales, and MCMC sampling turns the fit into a calibrated posterior over the emergence point.

What would settle it

Take a model family with a substantially different architecture or data mix, fit an emergence law using only its small pre-emergence checkpoints, and compare the predicted pretraining-loss elbow with the observed few-shot emergence of its larger checkpoints; if the prediction misses by more than 0.1 nats on several tasks, the transfer assumption fails.

Watch

Extended reading notes

Core claim

Finetuning a pre-emergence model on a task does not just improve that model: it reveals where the task's few-shot emergence elbow would sit in a larger model. Concretely, the paper shows that when intermediate checkpoints from a single 3B pretraining run are finetuned, the accuracy-versus-pretraining-loss curve keeps its ReLU shape but the elbow moves to higher loss (weaker models), and more finetuning data moves it further. The paper encodes this as an emergence law $E_\theta(D) = k \log(D)^\alpha + C$, fits $A, B, \theta$ jointly to raw finetuning accuracies through $\mathrm{Perf} = A\max(E_\theta(D)-L(M),0)+B+\Delta\mathbf{1}_{\mathrm{finetuned}}$, then evaluates at the low-data limit $D_0$ equal to the number of few-shot examples. On MMLU, GSM8K, CommonsenseQA, and CoLA the maximum-likelihood prediction lands within 0.1 nats of the true ReLU elbow, and holdout analysis shows MMLU and GSM8K can be predicted 4.3x and 3.9x the FLOPS before the first emerged checkpoint. A case study on the APPS coding benchmark with LLaMA 2 predicts emergence near 325B parameters, offered as a proof-of-concept for frontier-relevant capabilities.

Load-bearing premise

The method assumes that pretraining loss is a universal capability index: checkpoints from one pretraining run stand in for independently trained models of different sizes, and the finetuning shift observed on that run transfers to other architectures, data distributions, and dataset difficulties.

Editorial extensions

If this is right

  • Model developers can use emergence laws to compare pretraining data quality without training large models: on MMLU the method predicts that OpenLLaMA V2 emerges before V1, matching the observed order.
  • Safety and policy teams could estimate when a capability will emerge from small checkpoints alone, giving lead time before the large model that exhibits the capability is trained.
  • Predictions are insensitive to the exact low-data extrapolation limit within a reasonable range, so the method does not require a finely chosen $D_0$.
  • Accurate predictions are sometimes possible with only a handful of checkpoints and the smallest finetuning subsets, so data collection cost can be far below training a large model.
  • The same finetuning-induced shift appears on the harder APPS coding task, indicating the approach can extend beyond simple benchmarks toward capabilities closer to the frontier.

Reading between the lines

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

  • The paper leaves implicit that a minimal-rank adapter (LoRA rank 1 already shifts emergence) could serve as an even cheaper probe than full finetuning; testing whether adapter rank and data amount trade off in the same emergence law would make the method more economical.
  • If pretraining loss is a universal capability index, the emergence law fit on one model family should transfer to another family trained on the same data distribution; a direct cross-family transfer experiment would settle this and is the most natural next step.
  • The APPS prediction is testable: if a model trained on LLaMA 2-style data is scaled past roughly 250-500B parameters and evaluated on APPS, the first non-trivial pass@1 accuracy should appear in that window; a strong miss would refute the extrapolation.
  • Emergence laws could be inverted into a design tool: instead of only predicting when a task will emerge, a developer could select a finetuning data budget that pulls a desired capability down to a target scale, turning prediction into controlled elicitation.
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. The paper poses the problem of emergence prediction: given access to pre-emergence LLM checkpoints (models at or below random-chance few-shot accuracy on a task), can one predict the point in scaling at which few-shot performance will emerge? The authors observe that task-specific finetuning shifts the emergence elbow toward weaker models, and that the magnitude of the shift grows with the amount of finetuning data. They formalize this with an "emergence law" E_θ(D) = k·log(D)^α + C, fit jointly with a ReLU model of downstream accuracy as a function of pretraining loss, and then extrapolate to a low-data limit D_0 to predict the few-shot emergence point. Validation on MMLU, GSM8K, CommonsenseQA, and CoLA using OpenLLaMA V1 intermediate checkpoints shows MLE predictions within 0.1 nats of a ReLU-fit ground truth on all four tasks, with the ability to predict up to roughly 4x the pretraining FLOPS in advance on MMLU and GSM8K when later checkpoints are held out. The paper also presents two case studies: using the method to compare pretraining data quality between OpenLLaMA V1 and V2, and a proof-of-concept prediction of APPS emergence in LLaMA 2 at approximately 325B parameters.

Significance. If the method generalizes beyond the specific run family studied, it would be a valuable and inexpensive tool for forecasting emergent capabilities, with direct implications for safety preparedness, model development, and data-quality decisions. The paper's strengths include a clearly posed problem, a simple and reproducible method, and an honest discussion of limitations. The "No Few-shot" ablation in Section 6.3 rules out the most direct form of circularity, since accurate predictions are obtained even when the few-shot target data are excluded from fitting. The central empirical phenomenon—finetuning shifts the emergence elbow in a data-dependent way—is convincingly demonstrated across four benchmarks and three model sizes, and the paper makes its code and checkpoints available. The main weakness is that all primary validation is within a single pretraining run family (OpenLLaMA V1), which limits the strength of the claim that the method predicts emergence in future, independently trained models.

major comments (3)
  1. [Section 5.3 / Section 8] The central claim as framed in the abstract and introduction is that the method can predict whether future models (GPT-N+1) will emerge. However, the validation in Section 6 is entirely within a single OpenLLaMA V1 run: the "future models" are later checkpoints of the same optimizer trajectory, and the 4x-FLOPS-in-advance numbers in Section 6.4 are computed by holding out later checkpoints from that same run. Section 5.3 asserts that intermediate checkpoints from one run are "just as effective as using models from separate pretraining runs of varying scale," but the supporting evidence (Figure 3 left) only shows consistency across 3B/7B/13B checkpoints within the same run family at matched pretraining losses. Section 8 explicitly concedes that it is not understood whether models with different architectures or training schemes show the same downstream capabilities at a given pretraining loss. The V2 experiment in Section 7.1 is a second data mixture of the same architecture, and the APPS experiment has no ground truth. This is a load-bearing limitation: if the loss-capability mapping is run-specific, the 0.1-nat accuracy is a within-run extrapolation rather than a prediction about independently trained future models. The authors should either reframe the central claim as within-run emergence prediction (which is well supported) or provide evidence across independent pretraining runs, for example by fitting on one open model family and validating on another.
  2. [Section 6.2 / Equation 1] The ground-truth emergence point is itself estimated by fitting a ReLU to the few-shot checkpoints, and the success criterion is the distance between the predicted elbow and this fitted elbow. This is a reasonable operationalization, but the paper does not report uncertainty in the ground-truth fit. A noisy ground-truth fit can make the 0.1-nat threshold either too lenient or too strict, and the MCMC prediction intervals in Table 3 are often wider than the MLE error (e.g., GSM8K full-data interval [0.004, 0.170] nats). Please add a sensitivity analysis of the ground-truth ReLU fit—for example, bootstrap resampling of the few-shot checkpoints or an alternative definition of the emergence point—and report the resulting confidence interval alongside the prediction error. This would clarify whether the reported successes are robust to the choice of ground-truth model.
  3. [Section 6.4 / Figure 7] The headline numbers "4.3x and 3.9x FLOPS in advance" are computed by counting the earliest successful held-out checkpoint, but the paper itself notes (footnote 4) that GSM8K has a failed prediction between two successes. With only a handful of held-out checkpoints, the earliest-success criterion is sensitive to noise; the failed prediction is just outside the 0.1-nat threshold, but it is still a failure under the stated criterion. The full sequence of held-out predictions (including failures) appears only in Appendix A.6, and Figure 7 compresses them to single points. Please present the full sequence in the main text or figure, and if the "4x" claim is retained, state the criterion exactly as "earliest successful prediction" with the caveat about intermediate failures. The current presentation risks overstating the reliability of the advance-prediction capability.
minor comments (5)
  1. [Section 4.2] In the sentence "as we increase the amount of finetuning data, the point of point of emergence is shifted further," the phrase "point of point of" should be corrected to "point of emergence."
  2. [Section 7.1] The text contains a typo: "OpeLLaMA" should be "OpenLLaMA," and "corpa" should be "corpora" (or "corpora" in the singular, "corpus").
  3. [Section 5.1 / Table 1] The symbol table would be more complete if it included the optional parameter Δ from Section 5.4, which is subsequently used in the model equation but is absent from Table 1.
  4. [Figure 6 caption] The caption states that error bars represent the 5th and 95th percentile errors from MCMC posterior sampling; it would be clearer to note explicitly that these are percentiles of prediction error, not a confidence interval for the emergence point itself, although the main text does explain this.
  5. [Appendix A.7 / Section 6.1] The MCMC temperature sweep (selecting the greatest temperature for which the posterior mode is centered on the MLE) is a somewhat ad hoc calibration step that is only described in the appendix. A brief summary of this procedure in Section 6.1 would help readers assess the claimed uncertainty calibration.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the few-shot emergence prediction is an extrapolation of a separately fitted emergence law, validated on held-out checkpoints, with the few-shot target excluded in ablation.

full rationale

The paper's central derivation is not circular by construction. The emergence law Eθ(D) = k·log(D)^α + C is fitted to finetuned performance across data amounts D (Section 5.1-5.2, Eq. 2), using only pre-emergence 3B checkpoints for the main validation (Section 6.1). The target quantity is the few-shot emergence point E, estimated separately by fitting a ReLU to the held-out 7B/13B few-shot results (Section 6.1, 'Evaluation'). Nothing in the fitting procedure uses the target value E: the extrapolation Eθ(D0) at D0 equal to the few-shot prompt size is an out-of-sample limit, not an interpolation of the target. The 'No Few-shot' ablation (Figure 6, left) confirms that removing the few-shot data and the Δ parameter still yields accurate predictions, so the result does not reduce to a fitted value of the target. The Section 5.3 assumption that intermediate checkpoints from one pretraining run can stand in for separately trained models is an empirical generalization supported by Figure 3 and external prior work; Section 8 explicitly acknowledges that this transfer may fail for different architectures or training schemes. That is a generalizability and correctness risk, not circularity. The only self-citation (Gudibande et al., 2023) appears in the limitations discussion and is not load-bearing for any prediction. The paper is therefore self-contained against held-out external checkpoints, and no step equates a prediction to its input by definition.

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

The central claim rests on fitted parametric functions (emergence law parameters plus ReLU slope and floor, and an optional few-shot shift) and on domain assumptions that pretraining loss indexes capability uniformly and that a single run's checkpoints represent the scaling curve. No new physical or architectural entities are introduced.

free parameters (5)
  • Emergence law parameters k, alpha, C = not reported
    Fitted per task by L-BFGS after grid search (Appendix A.5); these define E(D) = k * log(D)^alpha + C and determine the extrapolated few-shot emergence point.
  • ReLU slope A and floor B = not reported
    Fitted jointly with emergence law parameters; shared across finetuning data amounts in the model of performance vs pretraining loss.
  • Few-shot shift delta = not reported
    Optional parameter modeling the base-rate shift when including few-shot datapoints; ablation shows small effects on prediction error (Section 5.4, Figure 6).
  • LLaMA 2 scaling law A, alpha, E = L(N) = 1.594 * N^{-0.089} + 0.408 (Figure 15)
    Fit to four loss values for 7B, 13B, 34B, 70B LLaMA 2 models (Appendix A.9); used only to convert the APPS loss prediction into a parameter count.
  • D0 (low data extrapolation limit) = number of few-shot examples N (5-7)
    Chosen as the few-shot prompt size; ablation over D0 = 1, N, 2N, 10N shows predictions are stable within a reasonable range (Section 6.3).
assumptions (5)
  • domain assumption Pretraining loss is a sufficient statistic for downstream capability across model sizes and checkpoints
    Used throughout as the x-axis for emergence (Section 3, 'Emergence with loss') and enables using checkpoints from one run in place of separate runs (Section 5.3).
  • domain assumption Emergence is modeled by a ReLU in pretraining loss
    Section 3, 'Modeling emergence with a ReLU'; chosen because the paper focuses on the point of emergence rather than post-emergence extrapolation, but it is a modeling assumption about the shape of capability scaling.
  • ad hoc to paper The emergence point shifts as a power law in log(D)
    Section 5.1, E(D) = k log(D)^alpha + C; the functional form is justified by empirical fit quality and ablations comparing alternatives, not derived.
  • domain assumption Intermediate checkpoints from one pretraining run are a valid proxy for independent models of different scales
    Section 5.3 and Figure 3; supported by within-family consistency of performance at equal loss, but assumes transfer across runs, a limitation acknowledged in Section 8.
  • domain assumption MCMC with uniform prior and tuned temperature gives calibrated emergence intervals
    Appendix A.7; the NUTS sampler required a temperature sweep to stabilize, so the posterior depends on this tuning choice.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Predicting Emergent Capabilities by Finetuning." pith.science (2026). https://pith.science/paper/FSTZXMWN

@misc{pith2026241116035,
  author       = {Pith},
  title        = {Pith review of: Predicting Emergent Capabilities by Finetuning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FSTZXMWN}},
  note         = {Machine review of arXiv:2411.16035}
}
read the original abstract

A fundamental open challenge in modern LLM scaling is the lack of understanding around emergent capabilities. In particular, language model pretraining loss is known to be highly predictable as a function of compute. However, downstream capabilities are far less predictable -- sometimes even exhibiting emergent jumps -- which makes it challenging to anticipate the capabilities of future models. In this work, we first pose the task of emergence prediction: given access to current LLMs that have random few-shot accuracy on a task, can we predict whether future models (GPT-N+1) will have non-trivial accuracy on that task? We then discover a simple insight for this problem: finetuning LLMs on a given task can shift the point in scaling at which emergence occurs towards less capable models. To operationalize this insight, we can finetune LLMs with varying amounts of data and fit a parametric function that predicts when emergence will occur (i.e., "emergence laws"). We validate this approach using four standard NLP benchmarks where large-scale open-source LLMs already demonstrate emergence (MMLU, GSM8K, CommonsenseQA, and CoLA). Using only small-scale LLMs, we find that, in some cases, we can accurately predict whether models trained with up to 4x more compute have emerged. Finally, we present a case study of two realistic uses for emergence prediction.

Figures

Figures reproduced from arXiv: 2411.16035 by the authors.

Figure 1
Figure 1. We find that task-specific finetuning systematically shifts the point of emergence towards less capable models. Motivated by this finding, we develop an emergence law, which models how the point of emergence shifts as a function of the amount of finetuning data. Using this emergence law, we can then extrapolate a prediction for the point of emergence in the few-shot setting. unsolved challenge in LLM scaling. In thi… view at source ↗
Figure 2
Figure 2. Left: we predict emergence in the few-shot setting by leveraging information about how “pre-emergence” models behave after finetuning. Our key finding is that finetuning effectively shifts the point of emergence from stronger to weaker models. Moreover, by varying the amount of finetuning data, the emergence point is shifted accordingly. We can use this finding to predict when few-shot emergence will occur by fittin… view at source ↗
Figure 3
Figure 3. Left: the finetuned and few-shot performance of intermediate LLM checkpoints. We plot downstream accuracy against pretraining loss for all 3B, 7B, and 13B intermediate OpenLLaMA V1 checkpoints on MMLU and GSM8K. We see that the point of emergence is systematically shifted towards weaker LLMs after finetuning. Additionally, the magnitude of the shift is consistent across all model sizes at the same pretraining loss. … view at source ↗
Figures from the paper (16 more)
Figure 4
Figure 4. Figure 4: Our MLE emergence law predictions on each task. The grey line is our extrapolated prediction and the multi-color lines are the fit. While our focus is on predicting the specific point of emergence (e.g., the ReLU elbow), we plot the full ReLU for visual clarity. We see…
Figure 5
Figure 5. Figure 5: The cumulative distribution function (CDF) of our emergence posterior on GSM8K and MMLU (see Appendix A.11 for all tasks). The CDF’s slope peaks at the mode of the distribution. We see that the distribution spikes near the true emergence point, followed by a moderately…
Figure 6
Figure 6. Figure 6: Ablations. The bar height represents the MLE prediction error (lower is better). The error bar represents the 5th and 95th percentile errors obtained from MCMC posterior sampling, and the circle is the median. Left: comparing emergence law functional forms. “Log Power …
Figure 7
Figure 7. Figure 7: How far in advance can we predict emergence? We hold out checkpoints to see how far in advance, in pretraining FLOPS, we can successfully predict emergence. The y position of each blue bar corresponds to the FLOPS needed to train the most capable model used for fitting…
Figure 8
Figure 8. Figure 8: Comparing OpenLLaMA V1 and V2 emergence. On both MMLU and CommonsenseQA, the V2 models emerge first, suggesting that the V2 pretraining data is likely higher quality. (right), we see that setting D0 to N generally results in the best predictions, but varying D0 within …
Figure 9
Figure 9. Figure 9: Comparing emergence predictions for OpenLLaMA V1 and V2 on MMLU. We plot the MLE predictions (left) and the CDFs (right) for both series. While our focus is on predicting the specific point of emergence (e.g., the ReLU elbow), we plot the full ReLU for visual clarity. …
Figure 10
Figure 10. Figure 10: Predicting emergence on APPS with LLaMA 2. On the left, we plot our MLE prediction. On the right, we convert this loss-based prediction into parameter count under the LLaMA 2 scaling law. The green point represents the MLE prediction, and the error bar represents the …
Figure 11
Figure 11. Figure 11: One the left we compare full fine-tuning against continuous prefix tuning on MMLU. We find that prefix tuning provides effectively no shift to the point of emergence, despite improving the performance of post-emergence models. On the right we compare 0-shot verses 5-s…
Figure 12
Figure 12. Figure 12: Comparing LoRA finetuning, with rank 1, 2, 4, and 64 against full finetuning on MMLU. We see that LoRA finetuning even with rank 1 shifts the point of emergence to a comparable degree to that of full finetuning. directly tuning the embeddings) to make training more un…
Figure 13
Figure 13. Figure 13: On a standard 5-shot MMLU and 6-shot CommonsenseQA (CSQA) evaluation, we observe emergence using both the standard correct answer accuracy evaluation and a continuous LLM log￾probability metric. 20 [PITH_FULL_IMAGE:figures/full_fig_p020_13.png]
Figure 14
Figure 14. Figure 14: We plot few-shot and full data finetuning performance as a function of pretraining loss using all 3B, 7B, and 13B model checkpoints for all tasks. We see that both the point of emergence and the downstream performance scaling thereafter, as a function of pretraining l…
Figure 15
Figure 15. Figure 15: We plot our scaling law fit for the LLaMA 2 series of models. We also include the learned values for our final fit on the plot. In this case N corresponds to parameter count in billions. We see that the LLaMA 2 models are well modeled by our scaling law. E). This give…
Figure 16
Figure 16. Figure 16: We plot the maximum likelihood predictions from our emergence law on each task. These plots include results from every finetuning run used for fitting the emergence law. The grey line represents our extrapolated prediction and the multi-color lines correspond to the f…
Figure 17
Figure 17. Figure 17: We plot the maximum likelihood predictions from our emergence law with OpenLLaMA V1 (left) and OpenLLaMA V2 (right) on MMLU. We plot C4 Validation loss on the x-axis. These plots include results from every finetuning run used for fitting the emergence law. The grey li…
Figure 18
Figure 18. Figure 18: We plot the MLE prediction (left) and MCMC CDF (right) for our emergence law fit using LLaMA 2 on APPS. The left plot includes results from every finetuning run used for fitting the emergence law. The grey line represents our extrapolated prediction and the multi-colo…
Figure 19
Figure 19. Figure 19: We plot the cumulative distribution function of our estimated posterior distribution over the point of emergence on each task. The stars correspond to few-shot performance on the task and represent the true emergence curve. The point at which the slope of the CDF peak…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

62 extracted references · 13 canonical work pages

  1. [1]

    Zhang, Bernd Bohnet, Luis Rosias, Stephanie Chan, Biao Zhang, Ankesh Anand, Zaheer Abbas, Azade Nova, John D

    Rishabh Agarwal, Avi Singh, Lei M. Zhang, Bernd Bohnet, Luis Rosias, Stephanie Chan, Biao Zhang, Ankesh Anand, Zaheer Abbas, Azade Nova, John D. Co-Reyes, Eric Chu, Feryal Behbahani, Aleksandra Faust, and Hugo Larochelle. Many-shot in-context learning, 2024. URL https://arxiv.org/abs/2404.11018

    arXiv 2024

  2. [2]

    Scaling laws for generative mixed-modal language models, 2023

    Armen Aghajanyan, Lili Yu, Alexis Conneau, Wei-Ning Hsu, Karen Hambardzumyan, Susan Zhang, Stephen Roller, Naman Goyal, Omer Levy, and Luke Zettlemoyer. Scaling laws for generative mixed-modal language models, 2023

    work page 2023

  3. [3]

    Dai, Anja Hauth, Katie Millican, David Silver, Slav Petrov, Melvin Johnson, Ioannis Antonoglou, Julian Schrittwieser, et al

    Rohan Anil, Sebastian Borgeaud, Yonghui Wu, Jean-Baptiste Alayrac, Jiahui Yu, Radu Soricut, Johan Schalkwyk, Andrew M. Dai, Anja Hauth, Katie Millican, David Silver, Slav Petrov, Melvin Johnson, Ioannis Antonoglou, Julian Schrittwieser, et al. Gemini: A family of highly capable multimodal models, 2023

    work page 2023

  4. [4]

    Usman Anwar, Abulhair Saparov, Javier Rando, Daniel Paleka, Miles Turpin, Peter Hase, Ekdeep Singh Lubana, Erik Jenner, Stephen Casper, Oliver Sourbut, Benjamin L. Edelman, Zhaowei Zhang, Mario Günther, Anton Korinek, Jose Hernandez-Orallo, Lewis Hammond, Eric Bigelow, Alexander Pan, Lauro Langosco, Tomasz Korbak, Heidi Zhang, Ruiqi Zhong, Seán Ó hÉigeart...

    arXiv 2024

  5. [5]

    Zoology: Measuring and improving recall in efficient language models

    Simran Arora, Sabri Eyuboglu, Aman Timalsina, Isys Johnson, Michael Poli, James Zou, Atri Rudra, and Christopher R \'e . Zoology: Measuring and improving recall in efficient language models. arXiv preprint arXiv:2312.04927, 2023

    arXiv 2023

  6. [6]

    Program synthesis with large language models, 2021

    Jacob Austin, Augustus Odena, Maxwell Nye, Maarten Bosma, Henryk Michalewski, David Dohan, Ellen Jiang, Carrie Cai, Michael Terry, Quoc Le, and Charles Sutton. Program synthesis with large language models, 2021. URL https://arxiv.org/abs/2108.07732

    arXiv 2021

  7. [7]

    Emergent abilities and grokking: Fundamental, mirage, or both?, 2023

    Boaz Barak. Emergent abilities and grokking: Fundamental, mirage, or both?, 2023. URL https://windowsontheory.org/2023/12/22/emergent-abilities-and-grokking-fundamental-mirage-or-both/. Accessed: 08-07-2024

    work page 2023

  8. [8]

    Managing ai risks in an era of rapid progress

    Yoshua Bengio, Geoffrey Hinton, Andrew Yao, Dawn Song, Pieter Abbeel, Yuval Noah Harari, Ya-Qin Zhang, Lan Xue, Shai Shalev-Shwartz, Gillian Hadfield, et al. Managing ai risks in an era of rapid progress. arXiv preprint arXiv:2310.17688, 2023

    arXiv

Show all 62 references
  1. [9]

    Larsen, Sean Owen, and Jonathan Frankle

    Cody Blakeney, Mansheej Paul, Brett W. Larsen, Sean Owen, and Jonathan Frankle. Does your data spark joy? performance gains from domain upsampling at the end of training, 2024. URL https://arxiv.org/abs/2406.03476

    2024 arXiv

  2. [10]

    JAX : composable transformations of P ython+ N um P y programs, 2018

    James Bradbury, Roy Frostig, Peter Hawkins, Matthew James Johnson, Chris Leary, Dougal Maclaurin, George Necula, Adam Paszke, Jake Vander P las, Skye Wanderman- M ilne, and Qiao Zhang. JAX : composable transformations of P ython+ N um P y programs, 2018. URL http://github.com/...

    2018

  3. [11]

    Broken neural scaling laws, 2023

    Ethan Caballero, Kshitij Gupta, Irina Rish, and David Krueger. Broken neural scaling laws, 2023

    2023

  4. [12]

    Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde de Oliveira Pinto, Jared Kaplan, Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, Alex Ray, Raul Puri, Gretchen Krueger, Michael Petrov, Heidy Khlaaf, Girish Sastry, Pamela Mishkin, Brooke Chan, Scott G...

    2021 arXiv

  5. [13]

    Training verifiers to solve math word problems

    Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, et al. Training verifiers to solve math word problems. arXiv preprint arXiv:2110.14168, 2021

    arXiv

  6. [14]

    Redpajama-data: An open source recipe to reproduce llama training dataset, 2023

    Together Computer. Redpajama-data: An open source recipe to reproduce llama training dataset, 2023. URL https://github.com/togethercomputer/RedPajama-Data

    2023

  7. [15]

    Understanding emergent abilities of language models from the loss perspective, 2024

    Zhengxiao Du, Aohan Zeng, Yuxiao Dong, and Jie Tang. Understanding emergent abilities of language models from the loss perspective, 2024. URL https://arxiv.org/abs/2403.15796

    2024 arXiv

  8. [16]

    Dimakis, Gabriel Ilharco, Shuran Song, Thomas Kollar, Yair Carmon, Achal Dave, Reinhard Heckel, Niklas Muennighoff, and Ludwig Schmidt

    Samir Yitzhak Gadre, Georgios Smyrnis, Vaishaal Shankar, Suchin Gururangan, Mitchell Wortsman, Rulin Shao, Jean Mercat, Alex Fang, Jeffrey Li, Sedrick Keh, Rui Xin, Marianna Nezhurina, Igor Vasiljevic, Jenia Jitsev, Alexandros G. Dimakis, Gabriel Ilharco, Shuran Song, Thomas K...

    2024

  9. [17]

    Openllama: An open reproduction of llama, May 2023

    Xinyang Geng and Hao Liu. Openllama: An open reproduction of llama, May 2023. URL https://github.com/openlm-research/open_llama

    2023

  10. [18]

    Scalax: scaling utilities for jax, 2024

    Young Geng. Scalax: scaling utilities for jax, 2024. URL https://github.com/young-geng/scalax

    2024

  11. [19]

    Mamba: Linear-time sequence modeling with selective state spaces

    Albert Gu and Tri Dao. Mamba: Linear-time sequence modeling with selective state spaces. arXiv preprint arXiv:2312.00752, 2023

    2023 arXiv

  12. [20]

    The false promise of imitating proprietary llms

    Arnav Gudibande, Eric Wallace, Charlie Snell, Xinyang Geng, Hao Liu, Pieter Abbeel, Sergey Levine, and Dawn Song. The false promise of imitating proprietary llms. arXiv preprint arXiv:2305.15717, 2023

    2023 arXiv

  13. [21]

    Measuring massive multitask language understanding, 2021

    Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. Measuring massive multitask language understanding, 2021

    2021

  14. [22]

    An overview of catastrophic ai risks, 2023

    Dan Hendrycks, Mantas Mazeika, and Thomas Woodside. An overview of catastrophic ai risks, 2023. URL https://arxiv.org/abs/2306.12001

    2023 arXiv

  15. [23]

    Scaling laws for autoregressive generative modeling

    Tom Henighan, Jared Kaplan, Mor Katz, Mark Chen, Christopher Hesse, Jacob Jackson, Heewoo Jun, Tom B Brown, Prafulla Dhariwal, Scott Gray, et al. Scaling laws for autoregressive generative modeling. arXiv preprint arXiv:2010.14701, 2020

    2010 arXiv

  16. [24]

    Scaling laws for transfer

    Danny Hernandez, Jared Kaplan, Tom Henighan, and Sam McCandlish. Scaling laws for transfer. arXiv preprint arXiv:2102.01293, 2021

    2021 arXiv

  17. [25]

    The no-u-turn sampler: adaptively setting path lengths in hamiltonian monte carlo

    Matthew D Hoffman, Andrew Gelman, et al. The no-u-turn sampler: adaptively setting path lengths in hamiltonian monte carlo. J. Mach. Learn. Res., 15 0 (1): 0 1593--1623, 2014

    2014

  18. [26]

    Rae, Oriol Vinyals, and Laurent Sifre

    Jordan Hoffmann, Sebastian Borgeaud, Arthur Mensch, Elena Buchatskaya, Trevor Cai, Eliza Rutherford, Diego de Las Casas, Lisa Anne Hendricks, Johannes Welbl, Aidan Clark, Tom Hennigan, Eric Noland, Katie Millican, George van den Driessche, Bogdan Damoc, Aurelia Guy, Simon Osin...

    2022

  19. [27]

    Predicting emergent abilities with infinite resolution evaluation, 2024

    Shengding Hu, Xin Liu, Xu Han, Xinrong Zhang, Chaoqun He, Weilin Zhao, Yankai Lin, Ning Ding, Zebin Ou, Guoyang Zeng, Zhiyuan Liu, and Maosong Sun. Predicting emergent abilities with infinite resolution evaluation, 2024. URL https://arxiv.org/abs/2310.03262

    2024 arXiv

  20. [28]

    Compression represents intelligence linearly, 2024

    Yuzhen Huang, Jinghan Zhang, Zifei Shan, and Junxian He. Compression represents intelligence linearly, 2024. URL https://arxiv.org/abs/2404.09937

    2024 arXiv

  21. [29]

    Scaling laws for downstream task performance of large language models

    Berivan Isik, Natalia Ponomareva, Hussein Hazimeh, Dimitris Paparas, Sergei Vassilvitskii, and Sanmi Koyejo. Scaling laws for downstream task performance of large language models. arXiv preprint arXiv:2402.04177, 2024

    2024

  22. [30]

    Scaling laws under the microscope: Predicting transformer performance from small scale experiments

    Maor Ivgi, Yair Carmon, and Jonathan Berant. Scaling laws under the microscope: Predicting transformer performance from small scale experiments. In Yoav Goldberg, Zornitsa Kozareva, and Yue Zhang (eds.), Findings of the Association for Computational Linguistics: EMNLP 2022, pp...

    2022 doi

  23. [31]

    Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei

    Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B. Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. Scaling laws for neural language models, 2020

    2020

  24. [32]

    Scaling laws for fine-grained mixture of experts, 2024

    Jakub Krajewski, Jan Ludziejewski, Kamil Adamczewski, Maciej Pióro, Michał Krutul, Szymon Antoniak, Kamil Ciebiera, Krystian Król, Tomasz Odrzygóźdź, Piotr Sankowski, Marek Cygan, and Sebastian Jaszczur. Scaling laws for fine-grained mixture of experts, 2024

    2024

  25. [33]

    Starcoder: may the source be with you! 2023

    Raymond Li, Loubna Ben Allal, Yangtian Zi, Niklas Muennighoff, Denis Kocetkov, Chenghao Mou, Marc Marone, Christopher Akiki, Jia Li, Jenny Chim, Qian Liu, Evgenii Zheltonozhskii, Terry Yue Zhuo, Thomas Wang, Olivier Dehaene, Mishig Davaadorj, Joel Lamy-Poirier, João Monteiro, ...

    2023

  26. [34]

    Prefix-tuning: Optimizing continuous prompts for generation

    Xiang Lisa Li and Percy Liang. Prefix-tuning: Optimizing continuous prompts for generation. arXiv preprint arXiv:2101.00190, 2021

    2021 arXiv

  27. [35]

    Does circuit analysis interpretability scale? evidence from multiple choice capabilities in chinchilla, 2023

    Tom Lieberum, Matthew Rahtz, János Kramár, Neel Nanda, Geoffrey Irving, Rohin Shah, and Vladimir Mikulik. Does circuit analysis interpretability scale? evidence from multiple choice capabilities in chinchilla, 2023. URL https://arxiv.org/abs/2307.09458

    2023 arXiv

  28. [36]

    Rush, Boaz Barak, Teven Le Scao, Aleksandra Piktus, Nouamane Tazi, Sampo Pyysalo, Thomas Wolf, and Colin Raffel

    Niklas Muennighoff, Alexander M. Rush, Boaz Barak, Teven Le Scao, Aleksandra Piktus, Nouamane Tazi, Sampo Pyysalo, Thomas Wolf, and Colin Raffel. Scaling data-constrained language models, 2023

    2023

  29. [37]

    Scaling data-constrained language models

    Niklas Muennighoff, Alexander Rush, Boaz Barak, Teven Le Scao, Nouamane Tazi, Aleksandra Piktus, Sampo Pyysalo, Thomas Wolf, and Colin A Raffel. Scaling data-constrained language models. Advances in Neural Information Processing Systems, 36, 2024

    2024

  30. [38]

    In-context learning and induction heads

    Catherine Olsson, Nelson Elhage, Neel Nanda, Nicholas Joseph, Nova DasSarma, Tom Henighan, Ben Mann, Amanda Askell, Yuntao Bai, Anna Chen, Tom Conerly, Dawn Drain, Deep Ganguli, Zac Hatfield-Dodds, Danny Hernandez, Scott Johnston, Andy Jones, Jackson Kernion, Liane Lovitt, Kam...

    2022

  31. [39]

    GPT-4 technical report, 2024

    OpenAI, Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, Red Avila, Igor Babuschkin, Suchir Balaji, Valerie Balcom, Paul Baltescu, Haiming Bao, Mohammad Bavarian, Jeff ...

    2024

  32. [40]

    How predictable is language model benchmark performance?, 2024

    David Owen. How predictable is language model benchmark performance?, 2024. URL https://arxiv.org/abs/2401.04757

    2024 arXiv

  33. [41]

    Mark zuckerberg - llama 3, open sourcing \ 10b models, & caesar augustus

    Dwarkesh Patel. Mark zuckerberg - llama 3, open sourcing \ 10b models, & caesar augustus. Podcast, Apr 2024. URL https://www.dwarkeshpatel.com/p/mark-zuckerberg. Accessed: 9/9/2024

    2024

  34. [42]

    The R efined W eb dataset for F alcon LLM : outperforming curated corpora with web data, and web data only

    Guilherme Penedo, Quentin Malartic, Daniel Hesslow, Ruxandra Cojocaru, Alessandro Cappelli, Hamza Alobeidli, Baptiste Pannier, Ebtesam Almazrouei, and Julien Launay. The R efined W eb dataset for F alcon LLM : outperforming curated corpora with web data, and web data only. arX...

    2023 arXiv

  35. [43]

    Composable effects for flexible and accelerated probabilistic programming in numpyro

    Du Phan, Neeraj Pradhan, and Martin Jankowiak. Composable effects for flexible and accelerated probabilistic programming in numpyro. arXiv preprint arXiv:1912.11554, 2019

    1912 arXiv

  36. [44]

    Code llama: Open foundation models for code, 2024

    Baptiste Rozière, Jonas Gehring, Fabian Gloeckle, Sten Sootla, Itai Gat, Xiaoqing Ellen Tan, Yossi Adi, Jingyu Liu, Romain Sauvestre, Tal Remez, Jérémy Rapin, Artyom Kozhevnikov, Ivan Evtimov, Joanna Bitton, Manish Bhatt, Cristian Canton Ferrer, Aaron Grattafiori, Wenhan Xiong...

    2024 arXiv

  37. [45]

    Maddison, and Tatsunori Hashimoto

    Yangjun Ruan, Chris J. Maddison, and Tatsunori Hashimoto. Observational scaling laws and the predictability of language model performance, 2024. URL https://arxiv.org/abs/2405.10938

    2024 arXiv

  38. [46]

    Are emergent abilities of large language models a mirage?, 2023

    Rylan Schaeffer, Brando Miranda, and Sanmi Koyejo. Are emergent abilities of large language models a mirage?, 2023

    2023

  39. [47]

    Active learning literature survey

    Burr Settles. Active learning literature survey. Computer Sciences Technical Report 1648, University of Wisconsin-Madison Department of Computer Sciences, 2009

    2009

  40. [48]

    Model evaluation for extreme risks

    Toby Shevlane, Sebastian Farquhar, Ben Garfinkel, Mary Phuong, Jess Whittlestone, Jade Leung, Daniel Kokotajlo, Nahema Marchal, Markus Anderljung, Noam Kolt, et al. Model evaluation for extreme risks. arXiv preprint arXiv:2305.15324, 2023

    2023 arXiv

  41. [49]

    Commonsenseqa: A question answering challenge targeting commonsense knowledge, 2019

    Alon Talmor, Jonathan Herzig, Nicholas Lourie, and Jonathan Berant. Commonsenseqa: A question answering challenge targeting commonsense knowledge, 2019

    2019

  42. [50]

    Scaling laws vs model architectures: How does inductive bias influence scaling? arXiv preprint arXiv:2207.10551, 2022

    Yi Tay, Mostafa Dehghani, Samira Abnar, Hyung Won Chung, William Fedus, Jinfeng Rao, Sharan Narang, Vinh Q Tran, Dani Yogatama, and Donald Metzler. Scaling laws vs model architectures: How does inductive bias influence scaling? arXiv preprint arXiv:2207.10551, 2022

    2022 arXiv

  43. [51]

    Improving pretraining data using perplexity correlations, 2024

    Tristan Thrush, Christopher Potts, and Tatsunori Hashimoto. Improving pretraining data using perplexity correlations, 2024. URL https://arxiv.org/abs/2409.05816

    2024 arXiv

  44. [52]

    Llama: Open and efficient foundation language models

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timoth \'e e Lacroix, Baptiste Rozi \`e re, Naman Goyal, Eric Hambro, Faisal Azhar, et al. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971, 2023 a

    2023 arXiv

  45. [53]

    Llama 2: Open foundation and fine-tuned chat models, 2023 b

    Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, W...

    2023 arXiv

  46. [54]

    GLUE : A multi-task benchmark and analysis platform for natural language understanding

    Alex Wang, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel Bowman. GLUE : A multi-task benchmark and analysis platform for natural language understanding. In Proceedings of the 2018 EMNLP Workshop B lackbox NLP : Analyzing and Interpreting Neural Networks fo...

    2018 doi

  47. [55]

    Chi, Tatsunori Hashimoto, Oriol Vinyals, Percy Liang, Jeff Dean, and William Fedus

    Jason Wei, Yi Tay, Rishi Bommasani, Colin Raffel, Barret Zoph, Sebastian Borgeaud, Dani Yogatama, Maarten Bosma, Denny Zhou, Donald Metzler, Ed H. Chi, Tatsunori Hashimoto, Oriol Vinyals, Percy Liang, Jeff Dean, and William Fedus. Emergent abilities of large language models, 2022 a

    2022

  48. [56]

    Chain-of-thought prompting elicits reasoning in large language models

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35: 0 24824--24837, 2022 b

    2022

  49. [57]

    Training trajectories of language models across scales, 2023

    Mengzhou Xia, Mikel Artetxe, Chunting Zhou, Xi Victoria Lin, Ramakanth Pasunuru, Danqi Chen, Luke Zettlemoyer, and Ves Stoyanov. Training trajectories of language models across scales, 2023. URL https://arxiv.org/abs/2212.09803

    2023 arXiv

  50. [58]

    Star: Bootstrapping reasoning with reasoning

    Eric Zelikman, Yuhuai Wu, Jesse Mu, and Noah Goodman. Star: Bootstrapping reasoning with reasoning. Advances in Neural Information Processing Systems, 35: 0 15476--15488, 2022

    2022

  51. [59]

    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 gl...

  52. [60]

    @esa (Ref

    \@ifxundefined[1] #1\@undefined \@firstoftwo \@secondoftwo \@ifnum[1] #1 \@firstoftwo \@secondoftwo \@ifx[1] #1 \@firstoftwo \@secondoftwo [2] @ #1 \@temptokena #2 #1 @ \@temptokena \@ifclassloaded agu2001 natbib The agu2001 class already includes natbib coding, so you should ...

  53. [61]

    \@lbibitem[] @bibitem@first@sw\@secondoftwo \@lbibitem[#1]#2 \@extra@b@citeb \@ifundefined br@#2\@extra@b@citeb \@namedef br@#2 \@nameuse br@#2\@extra@b@citeb \@ifundefined b@#2\@extra@b@citeb @num @parse #2 @tmp #1 NAT@b@open@#2 NAT@b@shut@#2 \@ifnum @merge>\@ne @bibitem@firs...

  54. [62]

    @open @close @open @close and [1] URL: #1 \@ifundefined chapter * \@mkboth \@ifxundefined @sectionbib * \@mkboth * \@mkboth\@gobbletwo \@ifclassloaded amsart * \@ifclassloaded amsbook * \@ifxundefined @heading @heading NAT@ctr thebibliography [1] @ \@biblabel @NAT@ctr \@bibset...

    2024

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.