Art of using and prompting LLMs

bolt.diy [bolt.diy] General Discussion
1

In Friday meeting one topic that was touches was prompting and using LLMs

I just listened in the morning best overview of how it looks today, it matches my approach I adopted over a year of working and integrating LLMs

Really recommend watching if you are digging into that:

Really hits on important topics like:

  • Different tier models for different tasks
  • Meta-prompting and how to iterate and test prompts, he shows his workflow which is almost exactly mine workflow too for some time
  • Test-driven Development with LLMs - something I argue for since 2022 since before ChatGPT :slight_smile: And still almost no one speaks about that… I think its next big thing in efficiently working with them
  • Distilling AI Model Performance - pointless waste of money without eval dataset
  • Model Routing and Multi-Model Orchestration - challenges
9 Likes
2

Phenomenal, thanks for sharing @wonderwhy.er! I totally agree that test driven development is the next big thing, especially because without it it’s way too dangerous to rely on LLMs editing anything in your codebase.

1 Like
3

This is great, Learned a lot from this Video. Thanks for sharing: )

2 Likes
4

Amazing video. Great insights. I totally get the 90-95% AI can do and last 5-10% is such a pain.

2 Likes