Eric Siu adapted Andrej Karpathy's autonomous research loops to generate and score marketing copy. The system creates 10 variants, ranks them with a five-expert panel, and evolves the top 3 until hitting a target score around 85+, all without needing traffic or A/B testing infrastructure.
We applied Andrej Karpathy's auto research for marketing, and we open sourced it. Check it out. So you can see here, this is autonomous research loops that generate, evaluate, and evolve solutions, but we applied this to marketing copy. So instead of optimizing for machine learning experience, it optimizes for landing pages, emails, ads, and forms, okay? So here's how it works. Check out this diagram over here. So we will generate the 10 variants, right? Let's say we're doing like emails, for example. We'll score with a 5 expert panel, we'll evolve the top 3, And we'll repeat this loop until the score is above 85 or so, okay? So obviously this is a little more subjective. It's not the same as like running a machine learning experiment, but you can see how this loop works. No traffic needed, no A/B testing infrastructure, minutes instead of weeks. So this allows you to move a lot faster and not have to question yourself when it comes to running marketing experiments. Now, if you want access to this repo, there's way more skills in it than just this. Just comment newsletter and we'll send it over to you.
