Can we really achieve consistent & deterministic response from AI models?

Situation when an LLM produces very similar response (but not strictly the same) to the same prompt has long seemed to be a feature. Still, it is annoying and makes LLM output feel unreliable. In this video we are using a technique called "Batch invariance" to force the same response to the same prompt.