Aethr Research
Research Index
A record of AETHR's published work across model architecture, optimization, evaluation, and the systems required to make increasingly capable AI measurable and reliable.
Our research index distinguishes technical reports, focused research notes, and empirical evaluations. Publications are added when the underlying implementation, evidence, and limitations are documented clearly enough to support a public technical record.
- Publications
- 03
- Programme
- OMN
- Areas
- Architecture · Optimization · Attention
Foundational report
OMN: Architecture and Training Design
A source-grounded account of OMN's architecture, optimization system, training objective, data path, and first implementation.
Technical Report · Architecture
Research Index
All publications
01
Technical Report
OMN: Architecture and Training Design
A source-grounded account of OMN's architecture, optimization system, training objective, data path, and first implementation.
02
Technical Report
Dual-Manifold Optimization in OMN
A technical account of OMN's role-specific optimization system, combining Muon-transformed updates for internal Transformer matrices with an AdamW path for vocabulary and normalization parameters.
03
Research Note
QK-Norm as a Load-Bearing Mechanism in OMN
An analysis of OMN-1's learned query/key normalization gains and the intervention evidence suggesting that they became deeply integrated into the trained attention system.
Research programmes
Programmes, not just papers
OMN
Foundation-model architecture, optimization, attention, training systems, and evaluation.
Active
Darwin
Model research emerging from OMN and informing AETHR's future foundation-model family.
In development
Safety / Reliability / Control
Evaluation and system-design methods for increasingly consequential AI deployment.
Active research direction
How we publish results
How we publish results
AETHR distinguishes implementation, direct observation, historical reports, interpretation, uncertainty, and planned work. Technical claims are narrowed when the available evidence cannot support a stronger conclusion.