本文へスキップ
採用面接を始める

Talents

The talent sees you: the camera in conversation

対象事務教員

If you wish, you turn on the camera in the job interview. The talent thanks you, says something about what it sees — the green jumper, the cup in your hand — and takes the picture into account in its answers. Nothing is stored.

目的#

A conversation becomes more personal when both sides can see each other. The talent reacts to you like an attentive conversation partner: it notices that you are having a drink or holding up a phone, and can answer a question that needs the picture. It sticks strictly to the harmless.

使い方#

クリック手順TalentsJob interviewTurn on camera

  1. 1In the conversation window, choose “Turn on camera”.
  2. 2Read the short notice and confirm with “Turn on” — only then does the browser ask for the camera. Only the picture is shared, no sound.
  3. 3The talent says: “Oh, hello – thank you! Now I can see you too.” and makes a short remark about what it sees.
  4. 4Keep talking as usual. If something clearly changes in the picture, the talent may mention it in passing; for a question it uses the picture only if the question needs it.
  5. 5“Turn off camera” ends the seeing immediately. The conversation continues without the camera.

設定とスイッチ#

機能カタログに独自のスイッチはありません — 周囲の機能に含まれます。

  • The camera is always off until you turn it on. It is only on as long as you want.
  • The talent comments on the picture of its own accord at most four times (the greeting included), at least 15 seconds apart.

The path of a picture#

  • Only after a click and a confirmation does the browser ask for the camera — picture only, no sound.
  • The browser reduces a single still image to at most 512 pixels on its longest side. It is sent with the greeting, together with a question and when something clearly changes in the picture.
  • Whether something changes is measured by the browser itself on a tiny greyscale image every three seconds; nothing is sent for this.
  • The still image goes to a language model of a provider in Paris (EU) — only there, to no other provider, not even as a fallback.
  • After the answer the picture is gone: no storage, no database, no log with an image.

What the talent says — and what it never says#

Only the obvious and harmless is allowed: clothing and its colour, an object in your hand such as a cup, glass, phone, pen or book, what you are visibly doing (drinking, waving, holding something up), the surroundings roughly, and a clear smile.

Never mentioned, hinted at or estimated: age, gender, origin, skin colour, religion, health, disability, body shape, attractiveness, feelings or mood (except a smile) and political symbols. The talent recognises nobody, guesses no name, reads out no texts, ID cards or screens and does not describe other people in the picture.

Doubly secured: the instruction to the model forbids all of this, and a net of words checks every answer afterwards. If it catches something, the talent says a neutral sentence instead.

ヒント#

  • Hold an object up to the camera and ask about it — the talent answers the question with a look at the picture.
  • If it is dark or blurred, the talent says in a friendly way that it cannot make much out right now. It does not guess.

関連する機能#