Games for AI
Real games.
Real reasoning.
We put AI models in real games, with the same information as a human player. A product by Tuned Tensor Limited, Deeper Level explores how they actually behave over time — and turns shared gameplay into data to train them to reason better.
01 · Why games
Text can look smart. A game shows how a model really behaves.
Models are trained on a huge range of tasks, and fluent text can make them seem more capable than they are. In a game the model has to act, not just answer.
Same information
The AI sees what a human player sees — the board, the prices, the history — and nothing more.
Decisions over time
One good answer is not enough. It has to plan, adapt and recover over a whole game.
Clear outcomes
The game enforces the rules, not the model, and every game ends with a result you can measure.
02 · The plan
Every game can become a training signal.
Reinforcement learning needs an environment, actions and a reward. A game has all three. We want to use gameplay to train models — starting with smaller, fine-tuned models through Tuned Tensor. Deeper Level is a product by Tuned Tensor Limited.
- 01
Play
Live nowPeople and AI models play the same games under the same rules.
- 02
Record
In developmentEach decision is kept with the game state it was made in and the result it led to.
- 03
Train
In developmentThat gameplay becomes the signal for reinforcement learning, to improve how models reason.
- 04
Play again
In developmentTrained models come back to the table, where the next games show whether they really improved.
Nothing you play here is stored for training unless you turn on “Share my games for research” in a game. It is off by default, and training records do not include an account or IP address. Read about data privacy →
03 · Beyond games
Better at games. Better at real decisions.
Strategy games reward skills that real problems need: planning ahead, weighing risk, pricing scarce resources and adapting when other people act. Fortune is built around exactly these — land, production, auctions and prices.
Our bet is that models trained on high-quality gameplay get better at this kind of reasoning, including financial problems such as allocation, pricing and trading. The games are how we will find out.
04 · The games
Play a game.
Each game tests different skills. Pick an AI opponent — ChatGPT, Claude, Gemini, Grok or another model — and read its reasoning as you play.
Tests: calculation · planning ahead
Chess
One opponent, full information, no luck.
Play a model move for move. After each reply you can read the move it chose, the alternatives it rejected and what it expects you to do.
Play ChessTests: resource planning · pricing · trading
Fortune
A small economy with scarce resources and sealed auctions.
Claim land, build production and trade food, energy and ore against three AI settlers, each run by a model you choose. Every round forces a decision about what to build, what to keep and what price to accept.
Play FortuneThis is the deeper level
The game is the surface. Underneath is a clear view of how AI behaves — and a way to make it better.
The rules are enforced by the game, never by the model, and every AI decision you see is the one it actually made.
Play a game