LLMs are not the product
summary
LLMs function as backend services to power features, not as stand-alone products: GitHub prototypes use an icebreaker button to turn an ambiguous chat box into a guided entry, and an AI-generated PR descriptions flow that lets users generate then refine output. The system runs hallucination mitigation—post-processing the model output with database lookups, filtering nonexistent items (example: issue #90) and rendering results as UI primitives instead of raw strings.
AI-written summary · about
chapters (10)
- 0:00Introduction and setup
- 1:09Who I am and my role
- 2:14LLMs are tools not magic
- 3:40Limitations of chat interfaces
- 5:04Why chatbots aren't the future
- 7:11Real-world examples at GitHub
- 9:17Generate refine generate loop
- 11:01Understanding hallucinations
- 12:23Hallucination mitigation example
- 14:50Wrap-up and key takeaways
related talks
15mLLMs for Data-Driven Decision Making in Product Development
29mWhat Is It Like to Be an LLM?
36mLLMs and AI agents demystified
01h 00mFrom LLM Theory to Practical Agentic Implementations
32mWhat can we expect of LLMs as software engineers?
19mLLM codegen fails and how to stop 'em
39mLLMs everywhere: The future is now and the past keeps repeating itself
40mLLMs - application and impact on business