The talk will cover three key areas for product integration: using the API, mastering prompt engineering, and implementing function calls for external data.
This section covers the basic code structure for making a chat completion request, including the different message roles and the stateless nature of the API.
Key factors for selecting a model include its training dataset cutoff date, context length measured in tokens, and the number of model parameters.
Improve LLM outputs by writing clear instructions, providing context with personas and references, and breaking down complex tasks into smaller steps.
Prevent users from bypassing system instructions by reinforcing the original rules with a post-prompt at the end of the message history.
Function calling allows the LLM to request help from your own code to access external data or perform actions like searching a database.
The talk concludes with a recap of core concepts and provides resources for advanced prompting techniques and retrieval-augmented generation (RAG).
The Q&A covers practical concerns like managing context length, prompt testing costs, implementing function call logic, and ensuring reliable JSON output.
26:30Simon A.T. Jiménez
32:26Krzysztof Cieślak
26:54Duan Lightfoot
25:17Scott Hanselman
33:28Maxim Salnikov
58:00Vijay Krishan Gupta & Gauravdeep Singh Lotey
28:38Tomislav Tipurić
29:50Christopher Walles
Jobs that call for the skills explored in this talk.
