No Training. No LLM. No Guessing.
Watch the Cube Solve Itself.
A live, human-scale demonstration of Zero-Training AI™ converting a moving 3D object
into a validated decision state and then executing a legal path to a defined goal.
Most people see a Rubik’s Cube as a puzzle. I see it as a decision problem you can hold in your hand.
Once the cube is scrambled, the objective is perfectly clear: return it to the solved state. But the cube cannot jump directly to that goal. Every turn changes several pieces at once. Every action changes the value of future actions. A move that appears helpful locally can create a worse decision surface several steps later. The system must know exactly where it is, distinguish legal states from impossible ones, evaluate the consequences of permitted actions, choose a path, and then execute that path without losing track of reality.
That is why this is such a powerful public demonstration of Zero-Training AI™. The cube is familiar enough for anyone to understand, yet unforgiving enough that the system cannot bluff. The cube either reaches the solved state through legal moves, or it does not. Every decision is visible.
This Is Not a Chatbot Pretending to Solve a Cube
The application does not send a picture to a language model and ask for a plausible answer. It does not call an online solving service. It does not rely on training data, a neural network, or a probabilistic text generator. It reads the live three-dimensional cube, converts the visual geometry into a precise internal decision state, and identifies the position and orientation of every corner and edge.
Before it makes a single solving move, it checks whether the state is physically possible. It looks for missing or duplicated pieces, illegal corner twists, illegal edge flips, and parity contradictions. If the state violates the cube’s physical invariants, the engine can reject it instead of confidently acting on corrupted information.
Only after the state is validated does the Decision Engine™ evaluate legal future actions and derive a sequence that returns the cube to its defined goal. Each decision is translated back into a visible cube movement. The system can pause after the current action, continue automatically, advance exactly one move, reverse a completed move, and independently confirm that the final physical state is solved.
subject to: every NextState must remain physically legal
What Makes This Zero-Training AI™
Zero-Training AI is a term I coined to describe artificial intelligence created directly from expert-defined variables, equations, rules, constraints, objectives, and safety boundaries—without training on large datasets or relying on a generative language model to make operational decisions.
In this demonstration, the variables are the cube pieces, their locations, and their orientations. The rules define legal turns. The constraints define states that cannot physically exist. The objective is an exact solved configuration. The safety boundary is simple but absolute: never treat an impossible or intermediate state as valid. The engine is therefore not guessing which move sounds right. It is deriving actions from a formal decision environment.
The important invention is not the existence of a Rubik’s Cube solver. Solvers have existed for decades. The importance of this demonstration is that it makes the complete decision architecture visible: observation, state conversion, validation, constraint enforcement, future-state evaluation, action selection, controlled execution, and verification. That same architecture is the foundation of serious decision software.
Why the Demonstration Is So Human
AI is often demonstrated with outputs that are difficult to verify. A paragraph can sound intelligent while being wrong. An image can look convincing while containing physical impossibilities. A confidence score can appear scientific even when the underlying decision cannot be explained.
A Rubik’s Cube removes that ambiguity. A person can scramble it with one click and watch the engine confront a new legal state. The live panel exposes the measurable decision problem. The viewer can stop the system, inspect the next action, advance it one move at a time, undo the action, and then let the engine continue. There is no need to trust a promotional claim. The audience becomes the test operator.
That is what makes this more than a technical demo. It gives people an immediate, physical understanding of deterministic decision intelligence. They can see the difference between generating an answer and controlling a real state through a sequence of constrained actions.
A Demonstration Should Prove the Capability—Not Surrender the Implementation
The public article and live application intentionally explain what the engine accomplishes without publishing the proprietary implementation details that make it work. The demonstration reveals the inputs, constraints, objectives, controls, and verifiable outcome. It does not need to reveal the internal piece mapping, orientation recovery logic, decision tables, pruning structures, search organization, or source code.
That is the right balance. Serious readers receive enough information to understand and test the claim. Potential customers and partners can see the operational value. At the same time, the implementation remains protected. A public demonstration should create proof, curiosity, and commercial interest—not become a free engineering specification for competitors.
From a Cube to Real Decision Systems
The Rubik’s Cube is deliberately small. It lets people understand the architecture without needing to understand aviation, transportation, medicine, logistics, or financial markets. But the decision pattern is much larger than the puzzle.
Replace cubie positions with sensor readings. Replace legal cube turns with permitted control actions. Replace the solved face pattern with a flight objective, a collision-free train path, a medical underwriting decision, an equipment allocation plan, or a media-buying objective. Then define the equations, rules, constraints, costs, and safety boundaries that govern the domain.
The cube demonstration does not, by itself, prove that every aircraft, train, or medical system is safe. What it proves is the underlying capability: a system can convert a live condition into a discrete decision state, reject impossible states, evaluate legal futures, select actions toward a defined objective, execute those actions in sequence, and verify the result—without an LLM and without model training.
That is the larger idea behind Zero-Training AI™: when expert knowledge can be explicitly defined, the intelligence can be built directly into the decision system instead of being statistically inferred from a mountain of historical examples.
Try It Yourself
Open the live demonstration and click SCRAMBLE. Then click SOLVE. Pause the engine, inspect it one move at a time, undo a move, or validate the cube before allowing it to continue.
The cube is small enough to understand, complex enough to expose bluffing, and precise enough to make every decision visible. That is why it may be the most human demonstration yet of what Zero-Training AI™ can do.