These assistants are probabilistic, not deterministic. Results are not always correct or even consistent.

https://aicamp.so/blog/ai-output-inconsistency-enterprise-so...

Don't take my word for it, ask the assistants themselves:

   *Why are your results not consistent?*

   *My results may not be consistent because I am a probabilistic AI model.*

Considering the underlying processes, it would be much more surprising if they were highly accurate and consistent.

They really excel at only one thing --- generating verbiage that sounds plausible. Misleading the unsuspecting. Being factual is not their forte.

So before incorporating one of the most powerful foot guns in the world into your workflow, you should ask yourself one question, "Do you feel lucky?".

Some 72% of responses by Gemini, Google's AI assistant, had significant sourcing issues, compared to below 25% for all other assistants, it said.

Issues of accuracy were found in 20% of responses from all AI assistants studied