Physics meets learning.

↙ Return to the open questionsBright record 009 · EarthInspect the source ↓

The human problem

Models must balance physical realism and computational cost.

What changed

The constraint: Parameterizations approximate processes too small to resolve directly.
AI’s role: Learned components worked inside a differentiable physical model.
The observed result: Researchers reported competitive weather forecasting and stable climate simulations.

The human possibility

The approach could expand tools for understanding climate.

Possible implications are separate from measured outcomes.

Keep following the question.

What holds this up

Neural general circulation models for weather and climate ↗

Event date: Not separately established
Source publication: 2024-07-22
Capture date: Not retained in the original record
Claim confidence: Unassessed. Separate from maturity; no numeric certainty inferred.
Review provenance: Legacy source check. No named human editorial approval is recorded.
Evidence state: Demonstrated · What this means
Source checked: 2026-09-05Compared with the published record. No independent replication by Bright.
Replication: Not assessed by Bright
Conflicting findings: Not systematically reviewed

The limits are part of the record.

This is not a complete replacement for comprehensive Earth-system models.

What this classification does—and does not—establish

Bright’s classification applies to the result described here. It does not certify a product, recommend a treatment, or predict a future outcome. Source publication and real-world deployment are different milestones. This opening record may not include every later study.

A record that can change

2026-09-05
Added to the source-checked opening collection.

This is Bright’s catalog history, not a complete history of the research. Future corrections and evidence revisions will be recorded here.

What comes next

Evaluate additional climate processes and changing conditions.

How corrections change the record ↗