<aside>

What Does "Non-Deterministic" Mean in This Context?

• Deterministic system: Given the same input and conditions, it always produces the same output.
• Non-deterministic system: Given the same input, it might produce different outputs on different runs.

Generative AI systems (like ChatGPT, DALL·E, or any large language model) often behave non-deterministically unless explicitly configured otherwise (e.g. by setting a fixed random seed and temperature).

</aside>

<aside>

Why Are Generative AI Systems Non-Deterministic?

<aside>

Sampling-based generation:

• These models generate output by sampling from a probability distribution over possible next tokens.
• Even slightly different sampling decisions can lead to significantly different final outputs. </aside>

<aside>

Temperature parameter:

• A higher temperature (e.g., 0.9) introduces more randomness and creativity.
• A lower temperature (e.g., 0.2) makes the model more conservative and deterministic. </aside>

<aside>

Top-k / Top-p sampling:

• These strategies limit which outputs can be sampled but still involve randomness.
• This further contributes to varied responses. </aside>

</aside>

<aside>

Example

Input →Write a poem about the moon.

Run 1:

The moonlight spills like silver wine, Across the lake so still, divine...

Run 2:

Pale guardian of the silent skies, You rise and fall with ancient sighs...

Same input → Different poetic outputs due to probabilistic token selection.

</aside>

<aside>

<aside>

Pros:

• Creativity, diversity, and spontaneity in outputs.
• Useful for brainstorming, writing, design, etc. </aside>

<aside>

Cons:

• Hard to reproduce exact outputs.
• Not ideal for tasks requiring exact consistency (e.g., legal or financial texts without further control). </aside>

</aside>

<aside>

Can It Be Made Deterministic?

Yes, by:

• Setting a fixed random seed.
• Using low temperature (e.g., 0.0 or 0.1).
• Enabling deterministic decoding like greedy search (always pick the most likely next token).

But this often reduces creativity and variability.

</aside>

<aside>

✅Topic Completed!

🌟Great work! You’re one step closer to your goal.

Ready to Move On →

</aside>

<aside>

• [ ] I have revised the topic at least once
• [ ] I want to practice more on this topic
• [ ] I have practiced enough and feel confident
• [ ] I need to revisit this topic later </aside>