Embedded Files
Before you start:
π Complete the pre-class exercise. [30 min]Β
π Download the class slides here.
1. Introduction
1. Introduction
How can we decide whether AI can help us achieve our specific goals? In this video, we'll focus on two use cases: an HKS admissions assistant (which you might have developed in the last class), and a policy drafter.
2. Task Criteria
2. Task Criteria
When assessing our scenarios, we can look at some fundamental criteria: personalization, interaction, size of the corpus of text, creativity and availability of demonstration data. In this video, we'll look at some examples and highlight some tips and tricks. Warning: AI jokes involved.
π Look back at the two examples discussed above (HKS admissions and policy drafting) and rate them according to the criteria discussed. Which application is better suited for generative AI?
3. Practical Consideration (Part 1)
3. Practical Consideration (Part 1)
Just because an application can be developed with generative AI, it doesn't mean it should. In this video, we look at some key considerations (and mitigations) we should keep in mind as we make this decision, such as privacy, clarity of alignment guidelines, cost of false information and comparison with the alternative.
π Think about the NYC bot example again. What level of performance would you consider satisfactory?
4. Practical Considerations (Part 2)
4. Practical Considerations (Part 2)
In this video, we explore some additional examples that show how LLMs struggle with race and gender bias.Β
π Look back at the two examples discussed above (HKS admissions and policy drafting) and rate them according to the considerations we just discussed. Which applications are safest to develop?
5. Key Takeaways
5. Key Takeaways
Remember: when we think of a task that we want to tackle with generative AI, we want to ask ourselves: are LLMs going to be useful in this scenario? Should they be deployed? Are they going o be successful and effective?
Page updated
Google Sites
Report abuse