Lesson 1.6 · Beginner

Why the same question can produce different answers

Understand generation settings and evaluate reasoning by its support.

3 min read

The short version

The model picks each next word from a list of likely candidates. Sometimes it takes the top one every time; sometimes it rolls the dice. That is why the same question can get different answers on different days, and why getting the same answer twice does not prove it is right.

On this page +

The idea What to check Try it

The idea

A text model scores possible next tokens. A token is a piece of text. The system then chooses one and repeats the process. Some methods choose the highest score. Others allow a choice from several likely tokens.

What to check

That is one reason the same question can get different answers. Settings can narrow the choices, but a low temperature does not guarantee the same result on every run. A repeated answer can still be wrong.

Some models spend extra work before answering. This may help with a hard task. It does not make the result true. A written chain of steps is also output that needs checking.

Try it

Two runs give the same case name and the same quote. What should you do next?

Check your understanding

Open the case in a trusted source. Confirm that it exists, find the quoted words, and read the surrounding passage. Agreement between runs is not a substitute for those checks.

Examples are teaching material. For legal rules, read the current primary text linked below. A revision date does not certify every claim.

Sources and further reading

Linked sources support the technical background identified below. The legal workflows, prompts, and fictional examples are original editorial exercises; they are not claims of endorsement by the linked authors.

CourseHugging Face · Text generation inference

Decoding, sampling, and generation settings.

← Back to chapter