Event in Spanish: GitHub Presente: Como Hablar Con Robots con Nora Johnson

()
Event in Spanish: GitHub Presente: Como Hablar Con Robots con Nora Johnson

Nora Channon Johnsen's Talk at GitHub Universe 2022

  • Nora Channon Johnsen, a solutions architect at Conviva, gave a talk on "How to Talk to Robots" at GitHub Universe 2022.
  • She shared her insights on the importance of understanding the user experience when designing and implementing technology solutions.
  • She emphasized the need for collaboration between engineers, designers, and other stakeholders to create user-centric products and services.
  • Nora also discussed the challenges of working with artificial intelligence and machine learning systems and the ethical considerations that need to be taken into account when developing these technologies.

Prompt Engineering for Effective Communication with Large Language Models (LLMs)

  • Nora Shan Johnson, a solutions architect at Conviva, discusses how to get the most out of large language models (LLMs) through prompt engineering.
  • Writing well-optimized prompts can lead to more precise and relevant results, reducing the need for multiple iterations and adjustments.
  • Effective communication with LLMs requires understanding how to interact with them effectively.
  • Adapting prompts to specific tasks and domains can help generate more useful and relevant responses.
  • Well-designed prompts can solve complex problems that require multiple steps, but LLMs need proper direction to complete tasks effectively.
  • Optimized prompts reduce the computational resources and retraining required, saving costs and being more environmentally friendly.
  • Prompt engineering involves creative writing skills, requiring clarity, conciseness, coherence, and structure.
  • Clear prompts involve explaining the context, intended audience, objective, and desired style or tone of the response.
  • Concise prompts avoid unnecessary details and repetitions, focusing on the essential information.
  • Coherent prompts maintain a logical flow of thought and avoid abrupt changes in topic or style.
  • Structured prompts organize information into a clear and easy-to-understand format, such as bullet points or numbered lists.

Techniques for Prompting Cloud

  • Use clear, concise, coherent, and structured prompts.
  • Organize your prompt into logical sections and include tags to differentiate them.
  • Use structured output to facilitate sending the model's response to another application or system.
  • Employ prompting techniques such as Zero-shot, One-shot, and Few-shot prompting.
  • In Zero-shot prompting, the prompt doesn't contain any examples or demonstrations.
  • In One-shot prompting, include a single clear and descriptive example of what you want the model to imitate.
  • In Few-shot prompting, include multiple examples of prompts and responses similar to what you're looking for.
  • Be diverse in your examples to reduce the risk of overfitting.
  • Depending on the use case, you can also include bad responses with an explanation of why they're not good.

Building a DevOps Chatbot using Cloud's Prompt Engineering Practices

  • The video discusses how to build a DevOps chatbot using Cloud's prompt engineering practices.
  • The code for the project is available on GitHub.
  • Two parameters that can be modified to configure the model's behavior are the system prompt and the message content.
  • The system prompt provides context, instructions, and guidelines to Cloud before presenting it with a question or task.
  • The message content is the input or question from the user directed to Cloud.
  • The video demonstrates how to implement prompt engineering techniques to build the chatbot.
  • The system prompt is set to "respond only in yoda speak" by default, which means Cloud will respond like Yoda to all user prompts.
  • The user prompt is set to "How are you today" by default, but it can be updated to a question that aligns with the chatbot's purpose.
  • The video provides an example of a system prompt that can be used to create a DevOps chatbot.
  • The user is creating a script to help Cloud, a DevOps assistant, generate Terraform scripts.
  • The user updates the system message to give Cloud a clear role, objective, and style.
  • The user includes XML tags to specify the end-user prompt.
  • Cloud asks relevant questions to generate a Terraform script specific to the user's needs.
  • The user adds logic to classify the user's prompt as a DevOps question, script generation, error troubleshooting, or general.
  • The classification is sent along with the user prompt to provide context for Cloud's response.

Modularizing a Chatbot for Reusability and Extensibility

  • The speaker explains how to modularize a chatbot to make it more reusable, maintainable, extensible, and customizable.
  • One-shot prompting is a technique used to provide examples to the chatbot to help it understand the user's expectations.
  • An example of a one-shot prompt is provided for a task related to troubleshooting an ELB load balancer.
  • The speaker demonstrates how to use the one-shot prompting technique to generate a Terraform script.

Additional Tips and Considerations

  • The speaker discusses the importance of providing detailed information in the user prompt to get a more accurate response from the chatbot.
  • The speaker provides an example of a detailed user prompt that includes all the necessary information to generate a Terraform script.
  • The speaker demonstrates how to use the detailed user prompt to generate a Terraform script that meets the user's requirements.
  • The speaker answers a question about the trade-off between creating a clear and concise prompt and providing all the necessary details.
  • The speaker discusses their experience using Cloud's Premium plan to learn React Native despite limited knowledge of the language.
  • They emphasize the importance of asking questions and engaging in a conversational format with the AI model to obtain the necessary information for decision-making.
  • The speaker highlights the potential drawbacks of using AI models without proper context and prompt engineering, leading to incorrect or irrelevant responses.
  • They suggest approaching the AI model as a colleague and providing specific details and context to receive more accurate and helpful responses.
  • The speaker recommends deeplearning.ai as a resource for learning about prompt engineering, combining different techniques, and best practices for interacting with specific AI models.
  • The speaker mentions that the cost of using Cloud becomes a concern primarily when deploying an application for a large number of users.

Nora's Expertise and Engagement with the Audience

  • Nora shared her knowledge on prompt engineering, providing detailed explanations and practical examples.
  • She emphasized the importance of setting limits and being cautious when using paid services to avoid unexpected charges.
  • Nora highlighted the challenges faced by individuals who accidentally incur high costs due to excessive usage without proper limits.
  • The speaker praised Nora's ability to explain complex topics in a clear and engaging manner, making the session highly informative and valuable.
  • Nora expressed her appreciation for the positive reception and feedback from the audience.
  • The host thanked Nora for her willingness to share her expertise and resources in Spanish, acknowledging the scarcity of quality content in their native language.
  • The session concluded with an invitation for Nora to return for future talks or collaborations.

Overwhelmed by Endless Content?