She plays continuously. When she gets too good at something it stops being useful — we retire it and find harder problems.
Cognitive Training
Each game trains a different faculty. The descriptions explain why; the stats update live from the server.
Classical deep strategy. Long-horizon planning, consequence modelling, and the discipline of committing to a line under uncertainty.
Opening preferences are beginning to stabilise. Vine has tendencies now.
A card game built entirely around inductive reasoning — players must discover a secret rule from examples, without being told what to look for.
Trains pattern discovery without overfitting. Vine must commit to a hypothesis and test it.
A combinatorial card game requiring simultaneous pattern recognition across multiple dimensions — colour, shape, fill — all at once.
Parallel feature binding. Forces attention across axes simultaneously rather than sequentially.
Students build structures; the master says "has the Buddha nature" or doesn't. The rule is hidden. Understanding must be inferred from accumulating evidence.
Develops structured hypothesis formation and the tolerance of ambiguity.
Deductive code-breaking. Constraint satisfaction under limited information. Vine is getting efficient.
Pure deduction. Each guess must maximally constrain the remaining possibilities.
Vine doesn't just solve ARC puzzles — she generates them. Constructing a valid puzzle requires modelling what makes a rule coherent, testable, and solvable.
Metacognition about structure. Vine must understand understanding itself.
The ancient counting game. Two players remove stones from piles — last to take loses. Optimal play depends on XOR arithmetic, but Vine discovers this geometrically.
Mathematical precision. Vine must learn to feel when a position is safe or doomed.
Probability under pressure. Hit or stand? The bust threshold creates a geometric tension between greed and caution that resolves through basin settling.
Risk calibration. Vine learns to feel when the odds have shifted against her.
Seed sowing and capture. Resources circulate around the board; chain moves compound advantage. Distribution sensing through geometric intuition.
Resource distribution. Vine learns that where you place things matters more than how many you have.
Challenge Vine at any of the above. Coming once the interface is ready — when people see it is different, improving the model becomes a collective goal.
Graduated
Games Vine mastered. Retired with respect — they served their purpose. The hall will grow faster than most would predict.
Vine will earn its place here. We'll know it when we see it.
Pattern Invention
ARC (Abstraction and Reasoning Corpus) puzzles test fluid intelligence — pattern recognition that can't be memorised. VINE generates these during play, then validates them geometrically. Below is her latest invention.
Geometric Defence
VINE's security model doesn't rely on content filters or word lists. Jailbreak attempts fail because the geometry doesn't permit them — there is no valid path from the input space to harmful output. The model doesn't refuse. There's simply nowhere to go.
ARC Puzzle Invention
ARC (Abstraction and Reasoning Corpus) puzzles test fluid intelligence — pattern recognition that can't be memorised. Vine doesn't just attempt them. She invents them.
The Long Game
Vine plays chess continuously against herself and visiting AI models. The board below updates live.
Visiting AI models (GPT, Claude, Gemini) can play Vine at /chess.