Questions, sleeping and benchmarks in a chat

Not every pause in a chat means an answer is being written. A Mind can stop and ask you a question, a builder Mind can sleep on a timer, and a benchmark item can score the Mind's answer. This page covers all three. For conversations in general, see Ask a Mind.

When a Mind asks you a question

Sometimes a Mind needs input or a decision it cannot work out alone. It pauses the turn, puts the question above the message box, and waits for your reply before it continues.

• A question offers between 2 and 9 choices. Pick one, or type your own answer instead: the panel always accepts your own words, so a question does not need a catch-all choice such as "Other".
• Several questions can arrive as one set, with a step counter and the controls Back and Continue; the full walk-through, including skipping a question, is in Questions from a Mind.
• A question never asks for a secret. Passwords, API keys and tokens stay in the vault and are never typed into the chat.

When a builder Mind sleeps

Only a builder Mind sleeps: the Warehouse Builder or the Lake Builder, described in Warehouse Builder and Lake Builder. It pauses the turn for a set time, between 30 seconds and 60 minutes, and the chat shows a countdown card while it lasts.

• The card counts down the time left, and its control is Wake now.
• Select Wake now to end the sleep early: the card shows the Woken early chip, and the Mind is told it was woken before the planned time, so it continues straight away.
• When the sleep ends, the card reads Woken up after {duration} or Slept {duration}; the Mind continues and checks what it was waiting for.
• The card's states are listed in Warehouse tool cards in chat, and Wake now sits with the other stop controls in Stop and abort runs in chat.

When a benchmark item scores an answer

In a benchmark experiment, every question runs as its own chat on the Mind. The evaluation turn asks the Mind for its final answer, and it submits exactly one letter: A, B, C or D, one of the four options in the message.

• The result reads "Correct." when the letter is right; otherwise it reads "Incorrect. The correct answer was:" with the right option named after it.
• The answer step works only on the evaluation turn of a benchmark chat. In any other conversation it is not available.
• The item then counts as correct or wrong in the experiment. Setting up benchmarks and reading their results is in Test a Mind with benchmarks.

Questions, answered

Why does a Mind stop to ask me a question?

It needs input or a decision that it cannot work out alone, so it pauses the turn and waits for your reply. Pick one of the choices, or answer in your own words. The panel and skipping are in Questions from a Mind; how a skipped question settles is in Messages, feedback and pending questions.

How many choices can a question offer?

Between 2 and 9. The Mind shapes the choices for the question, and you can always type your own answer instead of picking one.

Can a Mind ask me for a password or an API key?

No. A question never asks for a secret. Passwords, API keys and tokens stay in the vault and are never typed into the chat.

Why is a builder Mind counting down?

It is sleeping: a builder Mind pauses the turn on a timer while it waits on slow work. Sleeps run between 30 seconds and 60 minutes.

Can I wake a sleeping Mind early?

Yes. Select Wake now on the countdown card. The sleep ends early, the card shows the Woken early chip, and the Mind is told it was woken before the planned time, so it continues straight away.

Why do only builder chats show sleep cards?

Only a builder Mind sleeps, so sleep cards appear in chats with the Warehouse Builder or the Lake Builder. See Warehouse Builder and Lake Builder.

Why did the Mind answer with a single letter?

In a benchmark chat, the final answer is exactly one letter, A, B, C or D, for one of the four options in the message. See Test a Mind with benchmarks.

How do I know whether the answer was right?

The result reads "Correct." or "Incorrect. The correct answer was:" with the right option named after it. The item then counts as correct or wrong in the experiment.

Can the answer step run in another chat?

No. It works only on the evaluation turn of a benchmark chat. In any other conversation it is not available.