Why Measure AIs?
Generative intelligences have become critical infrastructure.
They write, recommend, guide, and decide.
Their responses now influence:
- healthcare,
- finance,
- legal,
- education,
- administration,
- autonomous systems.
Yet their real behavior in production remains largely unobserved.
A Public Instrument, a Documented Method
Like any scientific measurement instrument, NeoMundi must be understandable, discussable, and challengeable.
We publicly document how the instrument works: the observed signals, measurement modes, artifacts produced, interpretation limits, and data protection principles.
This documentation allows researchers, AI teams, auditors, lawyers, and compliance officers to understand how a runtime behavior becomes an observable signal.
Benchmarks are no longer enough.
Classic evaluations test isolated models in fixed contexts at a single point in time.
But a deployed generative service:
- evolves,
- drifts,
- changes infrastructure,
- changes load,
- changes behavior over time.
Measuring only static performance is no longer sufficient to understand the real stability of a generative system in production.
An AI that measures itself becomes observable
NeoMundi Research develops a runtime measurement approach applied to publicly deployed generative services.
The goal is not to judge an AI morally or politically.
The goal is to make certain signals observable:
- stability,
- drift,
- dispersion,
- critical regimes,
- behavior under informational load.
Why real-time (runtime) measurement changes everything
A correct answer can be produced in an unstable state.
Conversely, a stable answer can gradually drift under certain load, context, or interaction conditions.
The real behavior of a generative system cannot be understood solely from a static benchmark.
It must be observed over time, during its execution.
A logic similar to critical infrastructure
Electrical grids, financial markets, industrial systems, and aeronautical infrastructure are continuously monitored.
Generative systems are also becoming critical infrastructure.
They increasingly require:
- stability signals,
- observation protocols,
- versioned histories,
- traceability mechanisms.
What NeoMundi Research Publishes
NeoMundi Research progressively publishes:
- runtime mappings,
- public methodologies,
- versioned datasets,
- comparative signatures,
- reproducible observations.
All publications are released in open access.
Measure, Certify, Transmit
NeoMundi Research does not govern the observed services.
The association measures, documents, and publishes observable signals under defined experimental conditions.
Governance remains the responsibility of the operators.
The stability of AIs cannot remain invisible.
The thermodynamic mapping is a continuous program of public observation of generative systems.