Part 3/9:
The technical challenge involves prompting GPT-3 to pick a secret word and respond accurately to queries. Shapiro experiments with various techniques, including:
Using UUIDs (Universally Unique Identifiers) to ensure randomness.
Generating a list of common words—covering categories like food, places, famous people, and objects.
Selecting words from curated lists or random dictionaries to keep the game interesting and balanced in difficulty.
A key component is controlling the entropy of the seed to prevent GPT-3 from making predictable choices. The code samples involve leveraging randomness via Python's random module, reading word lists from external files, and customizing prompts with dynamic sampling.