AI as the judge,
not the writer.
UH OH is a social-survival game where the player gets one sentence to escape an awkward situation. The player writes every word. Jev, TypeSafe's System One model, sits behind the game as a fast semantic judge.
No multiple choice and no AI-generated rescue line.
Narrow signals such as tact, empathy, boundaries and social risk.
The game's own rules convert those judgments into Social HP.
A decision model makes the game mechanic possible.
TypeSafe describes Jev as its first public System One Model: a model designed for fast, typed, probabilistic decisions inside software rather than free-form text generation. UH OH needs exactly that kind of behavior. A single player reply can contain several social signals at once, and the game needs structured judgments quickly enough that submitting a line still feels like playing, not waiting for a chatbot.
✅ What Jev does
- Reads the situation and the player's sentence.
- Makes several narrow semantic judgments.
- Returns structured decisions with uncertainty.
- Acts as the game's fast semantic referee.
🚫 What Jev does not do
- It does not write the player's answer.
- It does not decide a person's real “EQ.”
- It does not invent the game's social rules.
- It does not choose the final HP change by itself.
A different kind of AI game.
Most AI games put a generative model in the role of character, storyteller or opponent. UH OH explores a different pattern: AI as a semantic physics engine. The player creates the action; the model measures what that action means; ordinary game logic decides what happens next.
You get one sentence. Jev can judge whether your reply is affectionate, tactful, evasive or dangerously comparative. UH OH then decides whether you heal or take social damage.
See real player-reply examples with recorded Jev scores, including the model version and observation date, or read the examples in Chinese.
Jev × UH OH
What is the Jev use case here?
Jev is used as a real-time semantic judge for player-written replies. The game needs multiple small judgments rather than a long generated answer.
Does Jev generate any dialogue?
No. Keeping the answer entirely player-written is central to the game. Jev only judges after submission.
Is the score a scientific EQ score?
No. UH OH is entertainment. Each level has game-specific social goals, and the resulting Social HP is a game score, not a psychological assessment.
Why not use a normal chatbot?
The interaction needs fast structured judgments that software can immediately turn into game state. That is the Jev capability UH OH is exploring.
Try the Jev use case yourself 😬
Ten awkward situations. One sentence each. See whether your social HP survives.