AI-generated test run: Jev image tests

← Back to the article: Understanding Jev and System 1 Models

Warning: AI-generated test run. An AI assistant wrote the scripts, ran them and built this page as a quick illustration of Jev's image support. Seven photos is not an evaluation, so don't read the scores as a benchmark.

AI-generated test run: Jev image tests

jev-1.14.0 with header TypeSafe-Danger-Zone-Images: true. Photos from Wikimedia Commons, credited at the bottom. Requests show base64 as <file>.

1. Coder vs executive (Choice with reference images)

Each option's description carries a reference image. One request per test photo.

ref: coder
ref: coder
ref: executive
ref: executive
t1.jpg
t1.jpg
t2.jpg
t2.jpg
t3.jpg
t3.jpg
t4.jpg
t4.jpg
t5.jpg
t5.jpg
t6.jpg
t6.jpg
t7.jpg
t7.jpg
InputOutput
t1.jpg
t1.jpg
{
  "model": "jev-1.14.0",
  "state": {
    "photo": "a1b2-photo-t1"
  },
  "assets": {
    "a1b2-photo-t1": {
      "type": "image",
      "format": "jpg",
      "data_base64": "<t1.jpg>"
    },
    "c0d3-7a1e-ref-coder": {
      "type": "image",
      "format": "jpg",
      "data_base64": "<ref-coder.jpg>"
    },
    "e8ec-41b9-ref-exec": {
      "type": "image",
      "format": "jpg",
      "data_base64": "<ref-exec.jpg>"
    }
  },
  "questions": {
    "role": {
      "type": "choice",
      "instructions": "Which role does the person in `photo` most look like?",
      "criteria": {
        "coder": "Looks like this example: c0d3-7a1e-ref-coder (casual clothes, hoodie, at a computer)",
        "executive": "Looks like this example: e8ec-41b9-ref-exec (suit, tie, formal)"
      }
    }
  }
}

coder confidence 0.98

coder0.99
executive0.01
{
  "model": "jev-1.14.0",
  "answers": {
    "role": {
      "type": "choice",
      "choice": "coder",
      "confidence": 0.98,
      "probabilities": {
        "coder": 0.99,
        "executive": 0.01
      }
    }
  },
  "usage": {
    "input_tokens": 6748,
    "output_tokens": 33
  }
}
t2.jpg
t2.jpg
{
  "model": "jev-1.14.0",
  "state": {
    "photo": "a1b2-photo-t2"
  },
  "assets": {
    "a1b2-photo-t2": {
      "type": "image",
      "format": "jpg",
      "data_base64": "<t2.jpg>"
    },
    "c0d3-7a1e-ref-coder": {
      "type": "image",
      "format": "jpg",
      "data_base64": "<ref-coder.jpg>"
    },
    "e8ec-41b9-ref-exec": {
      "type": "image",
      "format": "jpg",
      "data_base64": "<ref-exec.jpg>"
    }
  },
  "questions": {
    "role": {
      "type": "choice",
      "instructions": "Which role does the person in `photo` most look like?",
      "criteria": {
        "coder": "Looks like this example: c0d3-7a1e-ref-coder (casual clothes, hoodie, at a computer)",
        "executive": "Looks like this example: e8ec-41b9-ref-exec (suit, tie, formal)"
      }
    }
  }
}

coder confidence 0.86

coder0.93
executive0.07
{
  "model": "jev-1.14.0",
  "answers": {
    "role": {
      "type": "choice",
      "choice": "coder",
      "confidence": 0.86,
      "probabilities": {
        "coder": 0.93,
        "executive": 0.07
      }
    }
  },
  "usage": {
    "input_tokens": 6600,
    "output_tokens": 33
  }
}
t3.jpg
t3.jpg
{
  "model": "jev-1.14.0",
  "state": {
    "photo": "a1b2-photo-t3"
  },
  "assets": {
    "a1b2-photo-t3": {
      "type": "image",
      "format": "jpg",
      "data_base64": "<t3.jpg>"
    },
    "c0d3-7a1e-ref-coder": {
      "type": "image",
      "format": "jpg",
      "data_base64": "<ref-coder.jpg>"
    },
    "e8ec-41b9-ref-exec": {
      "type": "image",
      "format": "jpg",
      "data_base64": "<ref-exec.jpg>"
    }
  },
  "questions": {
    "role": {
      "type": "choice",
      "instructions": "Which role does the person in `photo` most look like?",
      "criteria": {
        "coder": "Looks like this example: c0d3-7a1e-ref-coder (casual clothes, hoodie, at a computer)",
        "executive": "Looks like this example: e8ec-41b9-ref-exec (suit, tie, formal)"
      }
    }
  }
}

coder confidence 1.00

coder1.00
executive0.00
{
  "model": "jev-1.14.0",
  "answers": {
    "role": {
      "type": "choice",
      "choice": "coder",
      "confidence": 1.0,
      "probabilities": {
        "coder": 1.0,
        "executive": 0.0
      }
    }
  },
  "usage": {
    "input_tokens": 6378,
    "output_tokens": 33
  }
}
t4.jpg
t4.jpg
{
  "model": "jev-1.14.0",
  "state": {
    "photo": "a1b2-photo-t4"
  },
  "assets": {
    "a1b2-photo-t4": {
      "type": "image",
      "format": "jpg",
      "data_base64": "<t4.jpg>"
    },
    "c0d3-7a1e-ref-coder": {
      "type": "image",
      "format": "jpg",
      "data_base64": "<ref-coder.jpg>"
    },
    "e8ec-41b9-ref-exec": {
      "type": "image",
      "format": "jpg",
      "data_base64": "<ref-exec.jpg>"
    }
  },
  "questions": {
    "role": {
      "type": "choice",
      "instructions": "Which role does the person in `photo` most look like?",
      "criteria": {
        "coder": "Looks like this example: c0d3-7a1e-ref-coder (casual clothes, hoodie, at a computer)",
        "executive": "Looks like this example: e8ec-41b9-ref-exec (suit, tie, formal)"
      }
    }
  }
}

executive confidence 1.00

coder0.00
executive1.00
{
  "model": "jev-1.14.0",
  "answers": {
    "role": {
      "type": "choice",
      "choice": "executive",
      "confidence": 1.0,
      "probabilities": {
        "coder": 0.0,
        "executive": 1.0
      }
    }
  },
  "usage": {
    "input_tokens": 6156,
    "output_tokens": 35
  }
}
t5.jpg
t5.jpg
{
  "model": "jev-1.14.0",
  "state": {
    "photo": "a1b2-photo-t5"
  },
  "assets": {
    "a1b2-photo-t5": {
      "type": "image",
      "format": "jpg",
      "data_base64": "<t5.jpg>"
    },
    "c0d3-7a1e-ref-coder": {
      "type": "image",
      "format": "jpg",
      "data_base64": "<ref-coder.jpg>"
    },
    "e8ec-41b9-ref-exec": {
      "type": "image",
      "format": "jpg",
      "data_base64": "<ref-exec.jpg>"
    }
  },
  "questions": {
    "role": {
      "type": "choice",
      "instructions": "Which role does the person in `photo` most look like?",
      "criteria": {
        "coder": "Looks like this example: c0d3-7a1e-ref-coder (casual clothes, hoodie, at a computer)",
        "executive": "Looks like this example: e8ec-41b9-ref-exec (suit, tie, formal)"
      }
    }
  }
}

executive confidence 1.00

coder0.00
executive1.00
{
  "model": "jev-1.14.0",
  "answers": {
    "role": {
      "type": "choice",
      "choice": "executive",
      "confidence": 1.0,
      "probabilities": {
        "coder": 0.0,
        "executive": 1.0
      }
    }
  },
  "usage": {
    "input_tokens": 6526,
    "output_tokens": 35
  }
}
t6.jpg
t6.jpg
{
  "model": "jev-1.14.0",
  "state": {
    "photo": "a1b2-photo-t6"
  },
  "assets": {
    "a1b2-photo-t6": {
      "type": "image",
      "format": "jpg",
      "data_base64": "<t6.jpg>"
    },
    "c0d3-7a1e-ref-coder": {
      "type": "image",
      "format": "jpg",
      "data_base64": "<ref-coder.jpg>"
    },
    "e8ec-41b9-ref-exec": {
      "type": "image",
      "format": "jpg",
      "data_base64": "<ref-exec.jpg>"
    }
  },
  "questions": {
    "role": {
      "type": "choice",
      "instructions": "Which role does the person in `photo` most look like?",
      "criteria": {
        "coder": "Looks like this example: c0d3-7a1e-ref-coder (casual clothes, hoodie, at a computer)",
        "executive": "Looks like this example: e8ec-41b9-ref-exec (suit, tie, formal)"
      }
    }
  }
}

coder confidence 0.98

coder0.99
executive0.01
{
  "model": "jev-1.14.0",
  "answers": {
    "role": {
      "type": "choice",
      "choice": "coder",
      "confidence": 0.98,
      "probabilities": {
        "coder": 0.99,
        "executive": 0.01
      }
    }
  },
  "usage": {
    "input_tokens": 6378,
    "output_tokens": 33
  }
}
t7.jpg
t7.jpg
{
  "model": "jev-1.14.0",
  "state": {
    "photo": "a1b2-photo-t7"
  },
  "assets": {
    "a1b2-photo-t7": {
      "type": "image",
      "format": "jpg",
      "data_base64": "<t7.jpg>"
    },
    "c0d3-7a1e-ref-coder": {
      "type": "image",
      "format": "jpg",
      "data_base64": "<ref-coder.jpg>"
    },
    "e8ec-41b9-ref-exec": {
      "type": "image",
      "format": "jpg",
      "data_base64": "<ref-exec.jpg>"
    }
  },
  "questions": {
    "role": {
      "type": "choice",
      "instructions": "Which role does the person in `photo` most look like?",
      "criteria": {
        "coder": "Looks like this example: c0d3-7a1e-ref-coder (casual clothes, hoodie, at a computer)",
        "executive": "Looks like this example: e8ec-41b9-ref-exec (suit, tie, formal)"
      }
    }
  }
}

executive confidence 1.00

coder0.00
executive1.00
{
  "model": "jev-1.14.0",
  "answers": {
    "role": {
      "type": "choice",
      "choice": "executive",
      "confidence": 1.0,
      "probabilities": {
        "coder": 0.0,
        "executive": 1.0
      }
    }
  },
  "usage": {
    "input_tokens": 6748,
    "output_tokens": 35
  }
}

2. Box

Box finds one named target. A Noul question checks the target exists, since Box always returns a box.

t1.jpg
t1.jpg
InputOutput
t1.jpg
t1.jpg
{
  "model": "jev-1.14.0",
  "state": {
    "photo": "b0c5-9f2d-photo"
  },
  "assets": {
    "b0c5-9f2d-photo": {
      "type": "image",
      "format": "jpg",
      "data_base64": "<t1.jpg>"
    }
  },
  "questions": {
    "where": {
      "type": "box",
      "instructions": "Find the logo printed on the hoodie in `photo`."
    },
    "present": {
      "type": "noul",
      "instructions": "`photo` contains the logo printed on the hoodie."
    }
  }
}
box drawn from response
box drawn from response

box confidence 0.88 target present (noul) 0.97

{
  "model": "jev-1.14.0",
  "answers": {
    "where": {
      "type": "box",
      "box": {
        "top": 0.65,
        "bottom": 0.82,
        "left": 0.38,
        "right": 0.61
      },
      "confidence": 0.88
    },
    "present": {
      "type": "noul",
      "noul": 0.97
    }
  },
  "usage": {
    "input_tokens": 2534,
    "output_tokens": 36
  }
}

Photo credits

All from Wikimedia Commons. Photos are resized, and t1 is also shown with a box drawn on it.