REVIEW 8 cited by
Why Has Predicting Downstream Capabilities of Frontier AI Models with Scale Remained Elusive?
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Signed reviews
read the original abstract
Predicting changes from scaling advanced AI systems is a desirable property for engineers, economists, governments and industry alike, and, while a well-established literature exists on how pretraining performance scales, predictable scaling behavior on downstream capabilities remains elusive. While many factors are certainly responsible, this paper identifies a significant factor that makes predicting scaling behavior on widely used multiple-choice question answering benchmarks challenging and illuminates a path towards making such downstream evaluations predictable with scale. Using five model families and twelve well-established multiple-choice benchmarks, we demonstrate that downstream performance is computed from negative log likelihoods via a sequence of transformations that progressively degrades the statistical relationship between performance and scale. We then pinpoint the mechanism causing this degradation: downstream metrics require comparing the correct choice against a small number of specific incorrect choices, meaning accurately predicting downstream capabilities requires predicting not just how probability mass concentrates on the correct choice with scale, but also how probability mass fluctuates on the alternative incorrect choices with scale. We empirically study how probability mass on the correct choice co-varies with probability mass on incorrect choices with increasing compute, suggesting that scaling laws for \textit{incorrect} choices might be achievable. Our work also explains why pretraining scaling laws are commonly regarded as more predictable than downstream capabilities and contributes towards establishing scaling-predictable evaluations of frontier AI models.
Forward citations
Cited by 8 Pith papers
-
Will Scaling Improve Social Simulation with LLMs?
Using 85 controlled and 35 public LLMs, the authors show social-simulation accuracy generally improves with compute, but some behavioral and low-resource tasks do not scale.
-
Toward a Theory of Value in AI Alignment
A systematic annotation of 94 AI alignment papers shows the field largely equates human values with measurable preferences, rarely defines values, and is increasingly removing humans from alignment evaluation.
-
Prescriptive Scaling Reveals the Evolution of Language Model Capabilities
For most benchmarks, the best achievable post-training accuracy follows a stable sigmoid curve in pre-training compute; math reasoning is the exception, with a boundary that keeps rising over time.
-
Signal and Noise: A Framework for Reducing Uncertainty in Language Model Evaluation
Benchmark signal-to-noise ratio, measured as score spread across models divided by checkpoint-to-checkpoint variability, predicts small-to-large model decision accuracy and can be improved by subtask filtering, checkp...
-
Language Models Improve When Pretraining Data Matches Target Tasks
Ranking pretraining documents by similarity to benchmark training examples (BETR) yields consistent benchmark gains and a 2.1x compute multiplier over DCLM-Baseline.
-
Loss-to-Loss Prediction: Scaling Laws for All Datasets
Losses of models trained on different datasets are related by shifted power laws, enabling translation of scaling laws and prediction of downstream performance from a few runs.
-
Position: Machine Learning Conferences Should Establish a "Refutations and Critiques" Track
ML conferences should create an official peer-reviewed track dedicated to refuting and critiquing previously published work.
-
Optimizing Pretraining Data Mixtures with LLM-Estimated Utility
Simple size-diversity heuristics outperform learned data mixing, while an LLM-based utility estimator (MEDU) matches expensive ablations at roughly 200x lower compute.
Discussion (0). Continue with ORCID to comment.