The "Stochastic Nature" of LLMs

FAQ DeepLearning.AI Learning Platform FAQ Short Courses FAQ
, ,
1

Remember: Understanding the LLM output variability and not just report “Bug: My output is different from Andrew’s video.”

  • Explanation: There are different parameters such as Temperature, Top-P, and Seeds, and also LLM models that due to its “stochastic nature” will output different results (answers) from the ones seen in the instructor’s notebook’s video. That is expected!

  • Expected vs. concerning differences:

    • :white_check_mark: Expected: Slight wording variations, creative differences

    • :warning: Review needed: Completely wrong answers, errors, non-functional code

  • Troubleshooting checklist:

    • Verify model version matches course

    • Test multiple runs

The Reality Check: In Generative AI, different outputs are a feature, not a bug. Unless the code throws an error, a variation in the wording of the response is expected behavior.

1 Like