LINEAiGE
AI HISTORY

Chinchilla Research Publication

DeepMind researchers introduced Chinchilla, a 70-billion-parameter language model, in the paper Training Compute-Optimal Large Language Models submitted to arXiv on March 29, 2022.

29 March 2022AI HISTORYVERIFIED

Historical classification

This verified record belongs to LINEAiGE's artificial-intelligence history layer beginning in 1956.

Why it matters

A documented research milestone that tested compute-optimal scaling by training a smaller model on substantially more data than larger contemporaneous language models under a comparable compute budget.

People, organizations & technologies

ORGANIZATIONS
Google DeepMind

TECHNOLOGIES
Chinchilla

Evidence & verification

STATE
verified

CONFIDENCE
high

LAST VERIFIED
2026-09-09T21:07:16+08:00

The arXiv record gives the submission timestamp as March 29, 2022 and states that the authors trained Chinchilla with 70 billion parameters using the same compute budget as Gopher but four times more data. A Google DeepMind NeurIPS 2022 retrospective independently describes Chinchilla as a 70-billion-parameter language model and says the work updated large-model scaling laws. This record is intentionally classified as a research-publication milestone, not a public product release, API launch, deployment date, or causal ancestor of later models. No geography, ancestry, influence, causality, or relationships are inferred.

Sources

SOURCE 1 · PRIMARY · CLAIM-EVIDENCE

Training Compute-Optimal Large Language Models

arXiv

PUBLISHED
2022-03-29

SOURCE 2 · PRIMARY · CORROBORATION

DeepMind's latest research at NeurIPS 2022

Google DeepMind

Trace the chronology

Record identity

LINEAiGE ID
chinchilla-2022