Skip to content
Longterm Wiki
resource

Predictability and Surprise in Large Generative Models

Metadata

Source Tableresources
Source ID370eb18a7329ea8b
DescriptionThe paper identifies a core tension in large generative models: while their training loss follows predictable scaling laws, their specific capabilities, behaviors, and outputs remain difficult to anticipate. This unpredictability creates challenges for safe deployment and policy regulation. The auth…
Source URLarxiv.org/abs/2202.07785
Children
CreatedApr 10, 2026, 9:35 PM
UpdatedApr 10, 2026, 9:35 PM
SyncedApr 10, 2026, 9:35 PM

Record Data

id370eb18a7329ea8b
urlarxiv.org/abs/2202.07785
titlePredictability and Surprise in Large Generative Models
typeweb
summaryThe paper identifies a core tension in large generative models: while their training loss follows predictable scaling laws, their specific capabilities, behaviors, and outputs remain difficult to anticipate. This unpredictability creates challenges for safe deployment and policy regulation. The auth
review
abstract
keyPoints
[
  "Large generative models exhibit predictable aggregate loss (scaling laws) but unpredictable specific capabilities and outputs, creating a fundamental safety challenge.",
  "The appearance of useful, predictable capabilities drives rapid development, while unpredictable qualities make it hard to…
publicationId
authors
authorEntityIds
publishedDate
tags
[
  "capabilities",
  "deployment",
  "governance",
  "policy",
  "evaluation",
  "ai-safety",
  "alignment",
  "red-teaming",
  "technical-safety"
]
localFilename
credibilityOverride
fetchedAt
contentHash
stableId
fetchStatusok
lastFetchedAtApr 10, 2026, 9:35 PM
archiveUrl
stance
contextNoteThis Anthropic paper examines the paradox of large generative models having predictable scaling laws but unpredictable emergent capabilities, with direct implications for AI safety governance and deployment risk assessment.
resourcePurposeanalysis
resourceSubtypeconference_paper
typeMetadata
publisherEntityId
relatedEntityIds
enrichmentStatusenriched
enrichmentDateApr 10, 2026, 9:35 PM
importanceScore0.72
contentLifecycle
Debug info

Thing ID: 370eb18a7329ea8b

Source Table: resources

Source ID: 370eb18a7329ea8b