Item — HN Reader

Just heard Steve today from Builder.io who did an impressive launch of Figma -> code powered by AI.

They trained a custom model for this. Better accuracy, sure, but I was a little surprised to watch how much faster it is than GPT4.

Based on their testing, they’ve become believers in domain specific smaller models, especially for performance.

1 comments

Codegen is an area that seems to have the best breakout performance compared to the big FMs. In hindsight it should be obvious given that openai created Codex.

However, I am skeptical that smaller and more focused models will do well for more general tasks. I’ve found gpt-4 to just be flatly excellent for tasks that involve emitting a custom DSL and customer-provided business context (the latter being something we cannot train for, not without immense expense).