Welcome to the community!
Do you have a question?
Hi all,
I am Vikramaditya aka Viktor.
I am almost done with the course. It was a great course for a beginner like me so thank you for creating this course. I learned many new things like - AI slop, progressive outline in writing, iterative brainstorming, when to use deep learning feature etc.
However, I felt a little overwhelmed by some concepts taught in this course.
Now I feel like I have a writer’s block while writing any prompt. I keep judging my own prompt.
I would like some clarity on these concepts so I can make use of this knowledge. Hopefully, it helps me start and use AI more rather than less.
-
AI critiquing
Context- Prof talks about using rubric for critiquing the writing done by AI. Then he shows two rubrics used to score one AI response, rubric -1 was not good at critiquing however a more detailed rubric-2 was able to judge the AI response better.
My question is- How will we be able to write such detailed rubric. Even just thinking about the example used, I could not think of what should be part of that rubric, especially, not with all the details like in the example. Should I use AI to write the rubric and then judge the response using that? Just thinking about writing such detailed rubric is giving me a writer’s block. -
Sycophancy
Using neutral prompts to control sycophantic responses from AI.
Questions- However, how to be sure that the prompt I wrote is neutral enough that AI will have to give objective responses? What if AI takes hint of slight inclination and then responds based on that?
What if I write a non neutral prompt but then ask AI to give only objective answers? Will that work? -
Context for better answers
Let’s take an example - A user wants an AI to compare several cars using specs, quotes, and insurance plans. What is the best way to improve the quality of the answer?
Based on the course answer is to give relevant files to the AI and context for quality answer.
Question -will AI not be more advanced than me at pulling the specs, quotes, insurance plans from web? Will it not be better to let it do the internet search and provide responses?
Looking forward to your responses.
Sure.
I understood the course, it was pretty basic. However, my concern is regarding the implementation of principles taught in this course. I have put all my questions in OG post.
My questions are related to - AI critiquing, Sycophancy and context setting.
- AI Critiquing - how to create such detailed rubric as shown in the video lecture.
- Sycophancy - as models are getting smarter, what if they can take a hint of inclination even from an objective prompt or chain of prompts.
- Context for better answers - course suggested to provide detailed context like excel comparative analysis sheet for buying car. How to create such detailed context? Won’t AI produce better analysis results?