Thankfully @igorTaber was just as insane as me and led our second round after chatting with AI Matt!
@mattshumer_
-
AI Voice Clone for VCs: 2021 Innovation Ahead of Its Time
By
–
Haha we did this back in 2021, complete with an AI clone of my voice that VCs could chat with People thought I was insane back then
-
Structured Output Limitations in AI Systems
By
–
JSON output != JSON input I've had quite bad luck w/ forced structured outputs in general though
-
Claude Models Performance Evaluation by Serious Practitioners
By
–
almost every model, especially claude models try it. almost any serious practitioner will tell you the same.
-
Structured Tags in AI: From Simple to Complex Use Cases
By
–
Our use cases range from extremely simple (i.e. output three options in tags) to absurdly complex (super nested tags with counters, metadata, etc.) Prompt clearly, and it'll work.
-
Prompt engineering beats validation libraries for reliable outputs
By
–
We don't really. It just works. I've seen so many validation libraries and techniques come and go, but most serious builders I know don't bother with them. They're essentially just nerd traps. Prompt the model well, and it'll output the right format 99%+ of the time.
-
Using XML for Prompt Engineering Techniques
By
–
We actually use XML for writing prompts 🙂 There are ways to get it in the right space, just needs a bit of exploration!
-
XML in Production: Stability Through Traditional Data Format
By
–
Not true at all. We use XML for almost everything in prod because it keeps things super stable.
-
XML Dramatically Improves Complex AI Task Performance
By
–
Sorry, but this is (mostly) wrong. Do they *need* XML? No. But does it improve outputs dramatically for complex tasks? Hell yes.
-
Markdown Performance Limitations with Claude AI Models
By
–
Agree, though I've found Markdown performs slightly worse on my tasks, esp. if using Claude