Description: Using DartmouthChat, submit a sample prompt and regenerate. Use the controls to adjust the parameters (temperature, top-p, and top-k) and see the influence of these on the outputs. Compare the variety of outputs generated from the same prompt.
Structure:
Ask students what they know about the reliability and reproducibility of GenAI. Do we also see the same response from an application? What are common reasons for using GenAI? Are these compatible with a technology that works this way?
Share definitions for:
Token probabilities
Stochastic
Temperature
Top-k and Top-p sampling
The instructor should select a short but complex prompt, for example, "What are two different common definitions of the liberal arts and how do these two definitions compare with each other?" Copy the output to a blank document (Word, Google docs, etc).
Click on the regenerate button to submit the prompt again. Compare the output with what was copied into the document. Ask students if these are significant or insignificant differences.
Ask the students to compare their outputs with those generated by a peer.
Use the controls feature on the top right of the interface to modify parameters.
First, examine the default values for temperature, top-p, and top-k. What is the scale and what are the defaults?
Lower temperature variable to 0.0 and regenerate. Raise and regenerate.
Achieved learning outcome: Become familiar with contemporary chatbot models and some of their major affordances concerning randomness and variation in generated output.