Part 4/10:
Once the contexts are established, the next step is to generate potential actions or decisions appropriate for each scenario. Again, the Curie instruct engine is used, given its strength at following directives. The process involves:
Replacing placeholders in prompts with specific contexts
Asking the model to list plausible responses or decisions
Employing different GPT-3 engines (like Curie or Ada) for balancing quality and cost-effectiveness
For example, in the scenario involving Ned and his wife, possible actions include "increase Ned's satisfaction" or "learn about the role of fathers," which are then evaluated relative to the core objectives.