ChatGPT & Prompt Engineering Podcast
Analyze stocks; write a whodunit mystery; can ChatGPT browsing do analysis? (Mastermind)
Sat Jun 10 2023Description
The episode discusses using web search functionality in AI systems like chat GPT for deeper research into different topics. The hosts share their experiences with building prompts and refining them to improve results. They explore challenges with the AI system's search capabilities, including limitations in finding relevant information and difficulties with specific instructions. The AI successfully retrieves information on retirement plans and AI tools, while also discussing trading patterns and building trading models. The hosts delve into creating a murder mystery story using Chat GPT, generating an outline with plot twists and hints towards the plot twist. They continue the story by summarizing chapter one and introducing chapter two. The episode concludes with an outline for the mystery story and a mention of an upcoming course on prompt engineering.
Insights
AI System's Limitations in Web Search
The AI system faces challenges in finding relevant information and providing helpful results. It spends time explaining unrelated topics and has limitations in its search capabilities.
Refining Prompts for Better Results
Narrowing down examples in prompts can yield better results compared to providing too many examples. Prompting the AI system with specific instructions does not significantly improve results.
Exploring Retirement Plans and AI Tools
The AI successfully retrieves information on the best 401k plans. The hosts discuss the idea of creating a mystery-solving activity for students and share their experiences with building financial models for stock analysis.
Using Algorithms and Models for Trading Patterns
The podcast explores using algorithms and models for detecting trading patterns based on criteria. Open source platforms are suggested for building trading models.
Creating a Murder Mystery Story with Chat GPT
The hosts discuss the challenge of structuring a mystery story with clues and critical thinking using Chat GPT. They generate an outline for a murder mystery short story with plot twists and hints towards the plot twist.
Continuing the Murder Mystery Story
Chapter one of the murder mystery story is summarized, and chapter two is introduced. The generator keeps track of characters and their actions throughout the story. GPT-4 is mentioned as a slower but enjoyable option for generating the story.
Outline for Mystery Story and Book Recommendations
An outline for the mystery story is provided, and the generation of chapter one and two is discussed. No book recommendations on prompt engineering are available yet, but a course is coming soon.
Chapters
- Using Web Search Functionality in AI Systems
- Challenges with AI System's Search Capabilities
- Exploring Retirement Plans and AI Tools
- Using Algorithms and Models for Trading Patterns
- Continuing the Murder Mystery Story
- Outline for Mystery Story and Book Recommendations
Using Web Search Functionality in AI Systems
00:00 - 08:08
- The podcast discusses using web search functionality in AI systems like chat GPT for deeper research into different topics.
- The host and guest share their experiences with building prompts and refining them to improve the results.
- They mention that providing too many examples in the prompts may not yield good results, and narrowing down the examples can be more effective.
- The guest expresses difficulty in getting the AI system to go deep into research and find specific information.
- The host suggests trying different prompts related to trends in retirement plans and AI tools.
- They discuss the browsing behavior of the AI system, which seems to analyze one webpage at a time rather than multiple pages.
- They decide to test a prompt asking for historical rate of return for top 10 retirement plans in the US.
- The guest mentions wanting to explore current trending risk areas for officers responsible for retirement plans.
Challenges with AI System's Search Capabilities
07:40 - 16:04
- The AI system is not successful in finding relevant information or providing helpful results.
- The system spends time explaining what a 401(k) is, which is not the desired information.
- There seems to be a time component that limits the system's search capabilities.
- The system fails to read certain websites and redirects to unrelated topics like credit cards.
- Prompting the system with specific instructions does not yield better results.
- The user suggests a strategy of searching multiple times and organizing the collected information.
Exploring Retirement Plans and AI Tools
15:45 - 24:03
- The AI is browsing the web and looking at credit cards.
- There are issues with providing a bulleted list prompt.
- The AI successfully retrieves information on the best 401k plans.
- The hosts discuss the idea of creating a mystery-solving activity for students.
- One host shares their attempts to get JTPT to perform a specific task.
- Another host talks about building a cup and handle financial model for stock analysis.
Using Algorithms and Models for Trading Patterns
23:39 - 32:16
- Chat GPT is not great at graphical reasoning
- There are ways to make technical analysis work with trading patterns and visual setups
- The podcast discusses using algorithms and models for detecting trading patterns based on criteria
- The host suggests looking into open source platforms for building trading models
- The host mentions a file from David that is different and leans into the Sherlock Holmes experience
- They discuss the challenge of getting Chat GPT to structure a mystery story with clues and critical thinking
- The host tries asking Chat GPT for ideas for clues, crimes, and contradictions in a mystery story
- They consider structuring the story by generating chapters one by one
- They mention characters like a PI, philanthropist, business tycoon, and scientist in the murder mystery story
- Chat GPT generates an outline for a murder mystery short story with plot twists and hints towards the plot twist
Continuing the Murder Mystery Story
31:48 - 40:48
- Axes and Murder Mystery, many best sellers write this chapter of a murder mystery short story using the outline crime plot twist
- Summarize that chapter into a single paragraph containing all the pertinent details to carry on into chapter two
- Replacing chapter one with chapter two in the summary
- The generator is able to keep track of characters and their actions throughout the story
- GPT-4 is not as fast as GPT-3.5 but provides a style that the reader can enjoy
- The final chapters are to be penned by the deceased himself, adding intrigue to the story
- The goal is for readers to try and figure out who did it before reaching the conclusion
- Subtle hints suggest it may be a suicide rather than a murder
Outline for Mystery Story and Book Recommendations
40:19 - 42:33
- Outline for mystery and then generate chapter one and two.
- No book recommendations on prompt engineering yet, but a course is coming soon.