Theory : Prompt Engineering

[head-iie-vnr]

Background: Context plays important role in getting right answer.

Chip could mean

• Potato chip
• Semiconductor chip
• Wooden chip.

Prompting Strategy would go from Least to Most Least : simple & stright forward prompt Most : Experiment with multiple and refine approach.

Below are various levels

• Basic : simple & stright forward
• Temperature & Top-P : Higher temperature more ranodmness (adding conext, Positive reinoforcements, add example, add task-speciic-keywords, Add prefixes/instructions, Fine uning (speciic to a style), Ierative.
• Self-Ask Prompting : How would you like to get asked, to give best answer.
• Meta Prompt : Influence the way chat-gpt by giving metadata (data about data). Eg: Tax vs VAT

Few examples that you can play with. https://experiments.withgoogle.com/ https://experiments.withgoogle.com/drum-machine https://musiclab.chromeexperiments.com/Song-Maker

[head-iie-vnr]

Other types of prompts (leveling continued)

• Chain of Thought : Build on top of the previous prompt. Example you say (Building on the previous description, do xyz)
• ReAct : Not very clear. Seems to be based on our acknowledgement (eg: Yes, No, explain etc prefixes before you ask next)
• Condtional Prop : Prepare biryani without usage of the biryani leaf. Optimise the code by reducing number of lines.

Below are some well-known prompting techniques, often used to enhance the effectiveness of interactions with AI models like ChatGPT. Here’s a brief explanation of each:

1. Instructional Prompt:

   • Description: Provides specific instructions or tasks for the model to complete.
   • Example: "Explain the process of osmosis in plant cells."

2. Conditional Prompt:

   • Description: Uses conditional statements to direct the model’s response based on certain conditions.
   • Example: "If a student asks about Newton’s third law, explain it with an example. Otherwise, provide a general overview of Newton’s laws."

3. Contextual Prompt:

   • Description: Provides context or background information to the model to frame the response appropriately.
   • Example: "As an introduction to a lesson on thermodynamics, describe the laws of thermodynamics with real-world examples."

4. Role Prompt:

   • Description: Specifies a role for the model to play, which guides the style and tone of the response.
   • Example: "You are a history teacher explaining the causes of World War I to high school students."

5. Iterative Prompt:

   • Description: Involves refining the prompt through multiple iterations to improve the quality of the response.
   • Example: "Describe the water cycle. Now, explain each stage in detail. Finally, provide examples of how the water cycle affects climate."

6. Perfect Prompt:

   • Description: An ideal or highly optimized prompt that generates the best possible response from the model.
   • Example: "Summarize the key points of the attached research paper in 200 words, focusing on the methodology and findings."

These prompting techniques can be extremely useful for guiding AI interactions to achieve specific goals, whether for educational purposes, content creation, or problem-solving.

[head-iie-vnr]

https://chat.lmsys.org/

It shows two model's responses side by side Lets you compare them.

[head-iie-vnr]

End of Day

Day1 Summary

• Basics of Gen AI
• Theory
• Hands on exercises : Generation of PPT, ChatPDF,

Objective : Understand the latest technologies so that we can go with them.

Next

Day2: Text to {Image, Audio, Video etc} Day3: GeneraiveAI into Research & Publications Day4-6 : Create our own models LLM, RankModel # Handson

All contents will be shared. Will help in the future.