A programmer works with instructions.
A designer works with visual and interactive elements.
A shaper works with intelligence itself.
SHAPING is the proposed creative practice of developing machine intelligence through intention, examples, experimentation, observation, and refinement.
Rather than simply instructing an intelligent system to perform a task, shaping concerns the development of what that system can do, how it behaves, and the boundaries within which it operates.
A person who practises shaping is a shaper.
The persistent, evolving work produced through shaping is called a Creation.
Anyone can shape intelligence.
This repository introduces, documents, and develops the conceptual framework behind SHAPING.
It is an open proposal for a human creative practice, not a claim to have invented the underlying technologies or established a new scientific standard.
Human beings have developed creative disciplines around the materials and technologies available to them.
We write with language. We design with visual and interactive elements. We program computational systems. We compose music, construct buildings, and create digital worlds.
As machine intelligence becomes increasingly adaptable, another possibility emerges:
What if developing intelligence itself becomes a creative activity?
Not merely using an existing AI system.
Not necessarily programming a new application.
Not necessarily training a neural network from scratch.
But intentionally developing intelligent capabilities and behaviours into something that reflects a person's goals, judgments, preferences, and imagination.
That is the idea behind SHAPING.
| Practice | Primary concern | Creative result |
|---|---|---|
| Writing | What should be expressed? | Written works |
| Programming | How should this execute? | Software and computational systems |
| Designing | How should this look and behave? | Visual and interactive experiences |
| Shaping | What should this intelligence become? | Intelligent capabilities and Creations |
These distinctions describe different creative concerns, not mutually exclusive professions.
A person may program, design, and shape the same system.
Programming traditionally concerns the construction of executable instructions and systems.
AI-assisted coding allows people to describe desired software and iterate on its implementation.
Shaping proposes a different creative emphasis.
The question is no longer only what the machine should do, but what the intelligence should become.
A shaper might want to develop an intelligence that:
- Explains complicated scientific ideas in a particular way.
- Understands and follows a specialised research methodology.
- Makes decisions within carefully defined constraints.
- Recognises when it lacks sufficient evidence.
- Develops expertise in a narrow professional domain.
- Collaborates with people according to particular preferences and boundaries.
The shaper expresses an intention, provides examples, experiments with interventions, observes the results, and progressively refines the system.
The outcome is not necessarily a new foundation model.
It may be a configured model, an adapted model, a composed agent, or an intelligent system assembled from multiple components.
The creative object is the capability and behaviour being developed.
The technical implementation is how that intention is realised.
The proposed practice can be described through seven recurring movements:
Intend → Demonstrate → Experiment → Observe → Evaluate → Refine → Preserve
Describe what you want the intelligence to become.
Define the capability, behaviour, purpose, or outcome you are seeking.
Show the system what you mean.
Provide examples, counterexamples, preferences, evaluation criteria, or boundaries.
Introduce an intervention that may influence the system.
Depending on the implementation, this could involve prompting, retrieval, model selection, fine-tuning, tools, configurations, or other engineering techniques.
Interact with the resulting system.
Examine what it actually does, including unexpected behaviour, limitations, and failures.
Compare the observed results with the original intention.
Identify improvements, regressions, trade-offs, and uncertainty.
Revise the intention, examples, implementation, or constraints.
Continue experimenting as your understanding develops.
Retain the resulting work, its configuration, and the evidence of how it developed.
That evolving work becomes a Creation.
These seven movements are an exploratory description of the practice, not a mandatory workflow or prescribed software interface.
A Creation is a persistent, evolving artefact developed through shaping.
It represents more than a conversation, a single prompt, or an isolated model execution.
A Creation may contain:
- Intention: What the creator wants the intelligence to achieve.
- Capabilities: The behaviours and functions being developed.
- Examples: Demonstrations, preferences, and reference material.
- Composition: Models, adapters, prompts, tools, retrieval systems, and configurations.
- Boundaries: Restrictions, safeguards, permissions, and intended limitations.
- Evidence: Experiments, evaluations, failures, and observed improvements.
- History: Versions, changes, provenance, and the reasoning behind decisions.
Not every Creation requires all these elements.
The concept does not prescribe a universal storage format or architecture.
A Creation need not be a newly trained neural network. It can be realised through several different technical methods, and preserving a Creation does not automatically confer ownership of its underlying models, data, or third-party components.
What matters is that the work can develop, be examined, and continue beyond a single interaction.
Imagine someone wants to develop an AI study companion.
Not simply a chatbot that answers questions, but an intelligence that helps students understand physics.
The creator wants it to explain difficult ideas without unnecessary complexity, recognise common misconceptions, acknowledge uncertainty, and avoid giving students a false impression of understanding.
They begin by expressing those intentions.
They demonstrate helpful and unhelpful explanations.
They introduce examples of common misunderstandings.
They experiment with different configurations and interventions.
They evaluate how the system responds to unfamiliar questions, ambiguous answers, and difficult edge cases.
They identify failures and refine its behaviour.
Over time, they preserve the resulting configuration, evaluation evidence, and development history as an evolving Creation.
The creator may not need to understand gradient descent.
The underlying infrastructure may nevertheless involve sophisticated machine-learning and software-engineering techniques.
The distinction is important:
Making the practice accessible does not make the engineering challenges disappear.
This is a hypothetical illustration of shaping, not evidence of a working implementation.
SHAPING is intended to describe a creative practice, not replace established technical disciplines.
It is not simply prompting.
A prompt can express an intention or influence behaviour. Shaping concerns the sustained development, evaluation, and preservation of intelligent capabilities. Prompting can be one of its techniques.
It is not synonymous with machine learning.
Training and model adaptation may be important parts of shaping. But intelligent behaviour can also be developed through configuration, composition, retrieval, routing, and tools.
It is not another name for AI-assisted coding.
AI-assisted coding focuses on producing software. Shaping focuses on developing the capabilities and behaviour of intelligent systems, whether or not producing software is involved.
It does not imply machine consciousness.
The term intelligence is used here in relation to observable machine capabilities and behaviours, not as an assertion about subjective experience.
It does not replace engineering or verification.
A convincing demonstration is not proof of reliability. Intended behaviour must be distinguished from observed behaviour, and observations must be evaluated against appropriate evidence.
SHAPING proposes a way to understand and organise the human creative activity around these technologies.
It does not claim that the underlying techniques are new.
The SHAPING framework is guided by several working principles.
- Intention gives the work direction. Begin with the capability or behaviour being sought, rather than assuming a particular implementation.
- Examples are creative material. Demonstrations, counterexamples, and preferences help translate intentions into something that can be developed and evaluated.
- Experimentation is part of creation. The work emerges through intervention, observation, and refinement rather than specification alone.
- Boundaries are part of intelligence. What a system must not do is as important as what it should be capable of doing.
- Evidence matters. A system should not be considered reliable merely because its outputs appear convincing.
- Creations should have continuity. Development history, configurations, failures, and discoveries should be preservable where technically and legally possible.
- Accessibility and technical rigour should coexist. More people should be able to participate in developing machine intelligence without concealing its complexity, limitations, or responsibilities.
The ambition behind SHAPING is broader than a particular application or model architecture.
It asks whether developing machine intelligence can become a recognisable creative practice in its own right.
One in which people do more than consume intelligent capabilities designed by others.
One in which they can participate in developing, examining, refining, and sharing those capabilities.
There are substantial unanswered questions.
What makes a Creation genuinely distinct from an ordinary configuration?
How much creative control can a person exercise over a model's behaviour?
How should that control be measured?
How can evidence, ownership, attribution, permissions, and portability be preserved?
What kinds of interfaces might make the practice meaningful to people without specialist machine-learning training?
These are research and design questions, not settled conclusions.
SHAPING is an invitation to investigate them.
SHAPING emerged in connection with the long-term vision for modelparadox, an intelligence engineering company.
The vision is to develop an environment in which people can create, adapt, evaluate, operate, and share machine intelligence.
Within that vision:
- SHAPING is the proposed human creative practice.
- A shaper is the person engaging in that practice.
- A Creation is the persistent, evolving work developed through it.
- modelparadox is the company exploring the product and infrastructure required to support that experience.
The distinction is deliberate.
SHAPING is proposed as an open conceptual framework that can be discussed, studied, criticised, and developed beyond any one company's products.
This repository does not represent a completed SHAPING application, an implemented Creation format, or a promise that any particular product capability currently exists.
| Document | Purpose |
|---|---|
| DEFINITION.md | The proposed formal definition and conceptual boundaries |
| PRACTICE.md | The creative process, examples, and working ideas |
| GLOSSARY.md | Shared terminology |
| ORIGIN.md | Origin, authorship, and provenance |
| FUTURE-B.md | Relationship between the concept and the product vision |
| CONTRIBUTING.md | How to contribute research, criticism, and improvements |
| LICENSE.md | Documentation licence and rights |
| CITATION.cff | Citation information |
Current status: Conceptual proposal, version 0.1.
The repository contains written definitions and exploratory material. It does not currently provide executable SHAPING software, a runtime, a model, or a technical standard.
The definitions and examples are open to refinement as research and practical experimentation progress.
This repository welcomes serious discussion of the proposed practice.
Useful contributions include conceptual criticism, related research, counterexamples, practical demonstrations, accessibility considerations, and proposals for evaluating the development of intelligent capabilities.
Contributions should distinguish clearly between hypotheses, proposed designs, observed results, and established facts.
See CONTRIBUTING.md for guidance.
The objective is not to defend terminology at all costs.
It is to develop an account of the practice that becomes more useful, precise, and testable over time.
This particular framing of SHAPING as a creative practice for developing machine intelligence was proposed and documented by Sineth Madduma in October 2026.
The repository preserves the definition, associated terminology, conceptual development, and subsequent revisions.
See ORIGIN.md for the recorded provenance and its evidentiary limits. The date of any earlier public presentation remains an author-reported claim until original publication links are supplied.
The proposal does not claim exclusive originality over the ordinary English words shaping or shaper, or over established machine-learning methods and earlier related research.
Documentation: Creative Commons Attribution 4.0 International (CC BY 4.0).
Copyright © 2026 Sineth Madduma.
The original copyrightable documentation in this repository may be shared and adapted, including commercially, subject to the terms of CC BY 4.0.
Attribution, a reference to the licence, and an indication of changes are required as applicable.
This licence does not grant exclusive rights to the words shaping or shaper, underlying ideas or methods, trademarks, patents, or third-party material. Future software, datasets, model weights, and other assets may require separate licensing.
For academic or technical references, see CITATION.cff.
Don't program intelligence. Shape it.
SHAPING — The Creative Practice of Developing Machine Intelligence.
Proposed by Sineth Madduma · 2026