Observational Scaling Laws and the Predictability of Language Model Performance Article Swipe

PDF

Yangjun Ruan , Chris J. Maddison , Tatsunori Hashimoto ·

YOU? · · 2024 · Open Access · · DOI: https://doi.org/10.48550/arxiv.2405.10938

Understanding how language model performance varies with scale is critical to benchmark and algorithm development. Scaling laws are one approach to building this understanding, but the requirement of training models across many different scales has limited their use. We propose an alternative, observational approach that bypasses model training and instead builds scaling laws from ~100 publically available models. Building a single scaling law from multiple model families is challenging due to large variations in their training compute efficiencies and capabilities. However, we show that these variations are consistent with a simple, generalized scaling law where language model performance is a function of a low-dimensional capability space, and model families only vary in their efficiency in converting training compute to capabilities. Using this approach, we show the surprising predictability of complex scaling phenomena: we show that several emergent phenomena follow a smooth, sigmoidal behavior and are predictable from small models; we show that the agent performance of models such as GPT-4 can be precisely predicted from simpler non-agentic benchmarks; and we show how to predict the impact of post-training interventions like Chain-of-Thought and Self-Consistency as language model capabilities continue to improve.

Related Topics

Concepts

Predictability Observational study Scaling Scaling law Econometrics Computer science Economics Mathematics Statistics Geometry

Metadata

Type: preprint
Language: en
Landing Page: http://arxiv.org/abs/2405.10938
PDF: https://arxiv.org/pdf/2405.10938
OA Status: green
Cited By: 2
Related Works: 10
OpenAlex ID: https://openalex.org/W4398157620

All OpenAlex metadata

Raw OpenAlex JSON

OpenAlex ID: https://openalex.org/W4398157620

Canonical identifier for this work in OpenAlex
DOI: https://doi.org/10.48550/arxiv.2405.10938

Digital Object Identifier
Title: Observational Scaling Laws and the Predictability of Language Model Performance

Work title
Type: preprint

OpenAlex work type
Language: en

Primary language
Publication year: 2024

Year of publication
Publication date: 2024-05-17

Full publication date if available
Authors: Yangjun Ruan, Chris J. Maddison, Tatsunori Hashimoto

List of authors in order
Landing page: https://arxiv.org/abs/2405.10938

Publisher landing page
PDF URL: https://arxiv.org/pdf/2405.10938

Direct link to full text PDF
Open access: Yes

Whether a free full text is available
OA status: green

Open access status per OpenAlex
OA URL: https://arxiv.org/pdf/2405.10938

Direct OA link when available
Concepts: Predictability, Observational study, Scaling, Scaling law, Econometrics, Computer science, Economics, Mathematics, Statistics, Geometry

Top concepts (fields/topics) attached by OpenAlex
Cited by: 2

Total citation count in OpenAlex
Citations by year (recent): 2025: 2

Per-year citation counts (last 5 years)
Related works (count): 10

Other works algorithmically related by OpenAlex

Full payload

id	https://openalex.org/W4398157620
doi	https://doi.org/10.48550/arxiv.2405.10938
ids.doi	https://doi.org/10.48550/arxiv.2405.10938
ids.openalex	https://openalex.org/W4398157620
fwci
type	preprint
title	Observational Scaling Laws and the Predictability of Language Model Performance
biblio.issue
biblio.volume
biblio.last_page
biblio.first_page
topics[0].id	https://openalex.org/T10028
topics[0].field.id	https://openalex.org/fields/17
topics[0].field.display_name	Computer Science
topics[0].score	0.9667999744415283
topics[0].domain.id	https://openalex.org/domains/3
topics[0].domain.display_name	Physical Sciences
topics[0].subfield.id	https://openalex.org/subfields/1702
topics[0].subfield.display_name	Artificial Intelligence
topics[0].display_name	Topic Modeling
topics[1].id	https://openalex.org/T10181
topics[1].field.id	https://openalex.org/fields/17
topics[1].field.display_name	Computer Science
topics[1].score	0.9124000072479248
topics[1].domain.id	https://openalex.org/domains/3
topics[1].domain.display_name	Physical Sciences
topics[1].subfield.id	https://openalex.org/subfields/1702
topics[1].subfield.display_name	Artificial Intelligence
topics[1].display_name	Natural Language Processing Techniques
is_xpac	False
apc_list
apc_paid
concepts[0].id	https://openalex.org/C197640229
concepts[0].level	2
concepts[0].score	0.9329477548599243
concepts[0].wikidata	https://www.wikidata.org/wiki/Q2534066
concepts[0].display_name	Predictability
concepts[1].id	https://openalex.org/C23131810
concepts[1].level	2
concepts[1].score	0.788745641708374
concepts[1].wikidata	https://www.wikidata.org/wiki/Q818574
concepts[1].display_name	Observational study
concepts[2].id	https://openalex.org/C99844830
concepts[2].level	2
concepts[2].score	0.5349883437156677
concepts[2].wikidata	https://www.wikidata.org/wiki/Q102441924
concepts[2].display_name	Scaling
concepts[3].id	https://openalex.org/C2988430800
concepts[3].level	3
concepts[3].score	0.46890413761138916
concepts[3].wikidata	https://www.wikidata.org/wiki/Q428971
concepts[3].display_name	Scaling law
concepts[4].id	https://openalex.org/C149782125
concepts[4].level	1
concepts[4].score	0.36179912090301514
concepts[4].wikidata	https://www.wikidata.org/wiki/Q160039
concepts[4].display_name	Econometrics
concepts[5].id	https://openalex.org/C41008148
concepts[5].level	0
concepts[5].score	0.3503289222717285
concepts[5].wikidata	https://www.wikidata.org/wiki/Q21198
concepts[5].display_name	Computer science
concepts[6].id	https://openalex.org/C162324750
concepts[6].level	0
concepts[6].score	0.2548661231994629
concepts[6].wikidata	https://www.wikidata.org/wiki/Q8134
concepts[6].display_name	Economics
concepts[7].id	https://openalex.org/C33923547
concepts[7].level	0
concepts[7].score	0.25480496883392334
concepts[7].wikidata	https://www.wikidata.org/wiki/Q395
concepts[7].display_name	Mathematics
concepts[8].id	https://openalex.org/C105795698
concepts[8].level	1
concepts[8].score	0.20050248503684998
concepts[8].wikidata	https://www.wikidata.org/wiki/Q12483
concepts[8].display_name	Statistics
concepts[9].id	https://openalex.org/C2524010
concepts[9].level	1
concepts[9].score	0.0
concepts[9].wikidata	https://www.wikidata.org/wiki/Q8087
concepts[9].display_name	Geometry
keywords[0].id	https://openalex.org/keywords/predictability
keywords[0].score	0.9329477548599243
keywords[0].display_name	Predictability
keywords[1].id	https://openalex.org/keywords/observational-study
keywords[1].score	0.788745641708374
keywords[1].display_name	Observational study
keywords[2].id	https://openalex.org/keywords/scaling
keywords[2].score	0.5349883437156677
keywords[2].display_name	Scaling
keywords[3].id	https://openalex.org/keywords/scaling-law
keywords[3].score	0.46890413761138916
keywords[3].display_name	Scaling law
keywords[4].id	https://openalex.org/keywords/econometrics
keywords[4].score	0.36179912090301514
keywords[4].display_name	Econometrics
keywords[5].id	https://openalex.org/keywords/computer-science
keywords[5].score	0.3503289222717285
keywords[5].display_name	Computer science
keywords[6].id	https://openalex.org/keywords/economics
keywords[6].score	0.2548661231994629
keywords[6].display_name	Economics
keywords[7].id	https://openalex.org/keywords/mathematics
keywords[7].score	0.25480496883392334
keywords[7].display_name	Mathematics
keywords[8].id	https://openalex.org/keywords/statistics
keywords[8].score	0.20050248503684998
keywords[8].display_name	Statistics
language	en
locations[0].id	pmh:oai:arXiv.org:2405.10938
locations[0].is_oa	True
locations[0].source.id	https://openalex.org/S4306400194
locations[0].source.issn
locations[0].source.type	repository
locations[0].source.is_oa	True
locations[0].source.issn_l
locations[0].source.is_core	False
locations[0].source.is_in_doaj	False
locations[0].source.display_name	arXiv (Cornell University)
locations[0].source.host_organization	https://openalex.org/I205783295
locations[0].source.host_organization_name	Cornell University
locations[0].source.host_organization_lineage	https://openalex.org/I205783295
locations[0].license
locations[0].pdf_url	https://arxiv.org/pdf/2405.10938
locations[0].version	submittedVersion
locations[0].raw_type	text
locations[0].license_id
locations[0].is_accepted	False
locations[0].is_published	False
locations[0].raw_source_name
locations[0].landing_page_url	http://arxiv.org/abs/2405.10938
locations[1].id	doi:10.48550/arxiv.2405.10938
locations[1].is_oa	True
locations[1].source.id	https://openalex.org/S4306400194
locations[1].source.issn
locations[1].source.type	repository
locations[1].source.is_oa	True
locations[1].source.issn_l
locations[1].source.is_core	False
locations[1].source.is_in_doaj	False
locations[1].source.display_name	arXiv (Cornell University)
locations[1].source.host_organization	https://openalex.org/I205783295
locations[1].source.host_organization_name	Cornell University
locations[1].source.host_organization_lineage	https://openalex.org/I205783295
locations[1].license	cc-by
locations[1].pdf_url
locations[1].version
locations[1].raw_type	article
locations[1].license_id	https://openalex.org/licenses/cc-by
locations[1].is_accepted	False
locations[1].is_published
locations[1].raw_source_name
locations[1].landing_page_url	https://doi.org/10.48550/arxiv.2405.10938
indexed_in	arxiv, datacite
authorships[0].author.id	https://openalex.org/A5074108427
authorships[0].author.orcid	https://orcid.org/0000-0002-0816-219X
authorships[0].author.display_name	Yangjun Ruan
authorships[0].author_position	first
authorships[0].raw_author_name	Ruan, Yangjun
authorships[0].is_corresponding	False
authorships[1].author.id	https://openalex.org/A5054711904
authorships[1].author.orcid
authorships[1].author.display_name	Chris J. Maddison
authorships[1].author_position	middle
authorships[1].raw_author_name	Maddison, Chris J.
authorships[1].is_corresponding	False
authorships[2].author.id	https://openalex.org/A5015518638
authorships[2].author.orcid	https://orcid.org/0000-0003-0521-5855
authorships[2].author.display_name	Tatsunori Hashimoto
authorships[2].author_position	last
authorships[2].raw_author_name	Hashimoto, Tatsunori
authorships[2].is_corresponding	False
has_content.pdf	False
has_content.grobid_xml	False
is_paratext	False
open_access.is_oa	True
open_access.oa_url	https://arxiv.org/pdf/2405.10938
open_access.oa_status	green
open_access.any_repository_has_fulltext	False
created_date	2024-05-21T00:00:00
display_name	Observational Scaling Laws and the Predictability of Language Model Performance
has_fulltext	False
is_retracted	False
updated_date	2025-11-06T06:51:31.235846
primary_topic.id	https://openalex.org/T10028
primary_topic.field.id	https://openalex.org/fields/17
primary_topic.field.display_name	Computer Science
primary_topic.score	0.9667999744415283
primary_topic.domain.id	https://openalex.org/domains/3
primary_topic.domain.display_name	Physical Sciences
primary_topic.subfield.id	https://openalex.org/subfields/1702
primary_topic.subfield.display_name	Artificial Intelligence
primary_topic.display_name	Topic Modeling
related_works	https://openalex.org/W3206776788, https://openalex.org/W2231089071, https://openalex.org/W2117748264, https://openalex.org/W4298190863, https://openalex.org/W2518229301, https://openalex.org/W2597205018, https://openalex.org/W2554072523, https://openalex.org/W1671671741, https://openalex.org/W2952002823, https://openalex.org/W2013329450
cited_by_count	2
counts_by_year[0].year	2025
counts_by_year[0].cited_by_count	2
locations_count	2
best_oa_location.id	pmh:oai:arXiv.org:2405.10938
best_oa_location.is_oa	True
best_oa_location.source.id	https://openalex.org/S4306400194
best_oa_location.source.issn
best_oa_location.source.type	repository
best_oa_location.source.is_oa	True
best_oa_location.source.issn_l
best_oa_location.source.is_core	False
best_oa_location.source.is_in_doaj	False
best_oa_location.source.display_name	arXiv (Cornell University)
best_oa_location.source.host_organization	https://openalex.org/I205783295
best_oa_location.source.host_organization_name	Cornell University
best_oa_location.source.host_organization_lineage	https://openalex.org/I205783295
best_oa_location.license
best_oa_location.pdf_url	https://arxiv.org/pdf/2405.10938
best_oa_location.version	submittedVersion
best_oa_location.raw_type	text
best_oa_location.license_id
best_oa_location.is_accepted	False
best_oa_location.is_published	False
best_oa_location.raw_source_name
best_oa_location.landing_page_url	http://arxiv.org/abs/2405.10938
primary_location.id	pmh:oai:arXiv.org:2405.10938
primary_location.is_oa	True
primary_location.source.id	https://openalex.org/S4306400194
primary_location.source.issn
primary_location.source.type	repository
primary_location.source.is_oa	True
primary_location.source.issn_l
primary_location.source.is_core	False
primary_location.source.is_in_doaj	False
primary_location.source.display_name	arXiv (Cornell University)
primary_location.source.host_organization	https://openalex.org/I205783295
primary_location.source.host_organization_name	Cornell University
primary_location.source.host_organization_lineage	https://openalex.org/I205783295
primary_location.license
primary_location.pdf_url	https://arxiv.org/pdf/2405.10938
primary_location.version	submittedVersion
primary_location.raw_type	text
primary_location.license_id
primary_location.is_accepted	False
primary_location.is_published	False
primary_location.raw_source_name
primary_location.landing_page_url	http://arxiv.org/abs/2405.10938
publication_date	2024-05-17
publication_year	2024
referenced_works_count	0
abstract_inverted_index.a	59, 89, 99, 102, 139
abstract_inverted_index.We	38
abstract_inverted_index.an	40
abstract_inverted_index.as	158, 183
abstract_inverted_index.be	161
abstract_inverted_index.in	73, 111, 114
abstract_inverted_index.is	8, 67, 98
abstract_inverted_index.of	27, 101, 128, 155, 176
abstract_inverted_index.to	10, 20, 70, 118, 172, 188
abstract_inverted_index.we	81, 123, 132, 149, 169
abstract_inverted_index.and	12, 48, 78, 106, 143, 168, 181
abstract_inverted_index.are	17, 86, 144
abstract_inverted_index.but	24
abstract_inverted_index.can	160
abstract_inverted_index.due	69
abstract_inverted_index.has	34
abstract_inverted_index.how	1, 171
abstract_inverted_index.law	62, 93
abstract_inverted_index.one	18
abstract_inverted_index.the	25, 125, 152, 174
abstract_inverted_index.from	53, 63, 146, 164
abstract_inverted_index.laws	16, 52
abstract_inverted_index.like	179
abstract_inverted_index.many	31
abstract_inverted_index.only	109
abstract_inverted_index.show	82, 124, 133, 150, 170
abstract_inverted_index.such	157
abstract_inverted_index.that	44, 83, 134, 151
abstract_inverted_index.this	22, 121
abstract_inverted_index.use.	37
abstract_inverted_index.vary	110
abstract_inverted_index.with	6, 88
abstract_inverted_index.~100	54
abstract_inverted_index.GPT-4	159
abstract_inverted_index.Using	120
abstract_inverted_index.agent	153
abstract_inverted_index.large	71
abstract_inverted_index.model	3, 46, 65, 96, 107, 185
abstract_inverted_index.scale	7
abstract_inverted_index.small	147
abstract_inverted_index.their	36, 74, 112
abstract_inverted_index.these	84
abstract_inverted_index.where	94
abstract_inverted_index.across	30
abstract_inverted_index.builds	50
abstract_inverted_index.follow	138
abstract_inverted_index.impact	175
abstract_inverted_index.models	29, 156
abstract_inverted_index.scales	33
abstract_inverted_index.single	60
abstract_inverted_index.space,	105
abstract_inverted_index.varies	5
abstract_inverted_index.Scaling	15
abstract_inverted_index.complex	129
abstract_inverted_index.compute	76, 117
abstract_inverted_index.instead	49
abstract_inverted_index.limited	35
abstract_inverted_index.models.	57
abstract_inverted_index.models;	148
abstract_inverted_index.predict	173
abstract_inverted_index.propose	39
abstract_inverted_index.scaling	51, 61, 92, 130
abstract_inverted_index.several	135
abstract_inverted_index.simple,	90
abstract_inverted_index.simpler	165
abstract_inverted_index.smooth,	140
abstract_inverted_index.Building	58
abstract_inverted_index.However,	80
abstract_inverted_index.approach	19, 43
abstract_inverted_index.behavior	142
abstract_inverted_index.building	21
abstract_inverted_index.bypasses	45
abstract_inverted_index.continue	187
abstract_inverted_index.critical	9
abstract_inverted_index.emergent	136
abstract_inverted_index.families	66, 108
abstract_inverted_index.function	100
abstract_inverted_index.improve.	189
abstract_inverted_index.language	2, 95, 184
abstract_inverted_index.multiple	64
abstract_inverted_index.training	28, 47, 75, 116
abstract_inverted_index.algorithm	13
abstract_inverted_index.approach,	122
abstract_inverted_index.available	56
abstract_inverted_index.benchmark	11
abstract_inverted_index.different	32
abstract_inverted_index.phenomena	137
abstract_inverted_index.precisely	162
abstract_inverted_index.predicted	163
abstract_inverted_index.sigmoidal	141
abstract_inverted_index.capability	104
abstract_inverted_index.consistent	87
abstract_inverted_index.converting	115
abstract_inverted_index.efficiency	113
abstract_inverted_index.phenomena:	131
abstract_inverted_index.publically	55
abstract_inverted_index.surprising	126
abstract_inverted_index.variations	72, 85
abstract_inverted_index.benchmarks;	167
abstract_inverted_index.challenging	68
abstract_inverted_index.generalized	91
abstract_inverted_index.non-agentic	166
abstract_inverted_index.performance	4, 97, 154
abstract_inverted_index.predictable	145
abstract_inverted_index.requirement	26
abstract_inverted_index.alternative,	41
abstract_inverted_index.capabilities	186
abstract_inverted_index.development.	14
abstract_inverted_index.efficiencies	77
abstract_inverted_index.Understanding	0
abstract_inverted_index.capabilities.	79, 119
abstract_inverted_index.interventions	178
abstract_inverted_index.observational	42
abstract_inverted_index.post-training	177
abstract_inverted_index.predictability	127
abstract_inverted_index.understanding,	23
abstract_inverted_index.low-dimensional	103
abstract_inverted_index.Chain-of-Thought	180
abstract_inverted_index.Self-Consistency	182
cited_by_percentile_year
countries_distinct_count	0
institutions_distinct_count	3
citation_normalized_percentile