I remember watching the AlphaGo documentary in 2017. What stood out to me was that the model got drastically better when it started competing against itself. GANs clicked for me similarly: a generator and discriminator competing, and somehow the competition is what produces something remarkable.

I've been curious whether this principle generalizes to today's agents.

So mehulkalia and I built Browser Brawl at the YC / BrowserUse hackathon last weekend and won first place. It is a fun experiment in which an attacker agent tries to complete tasks on live websites while a defender agent injects JavaScript to sabotage it.

The analogy isn't perfect, because browser tasks aren't zero-sum. But our hypothesis is that an agent faced with an adversary should produce more interesting training data than one navigating clean, static environments.

Try it on: http://browser-brawl.com

GitHub: https://github.com/RichardHruby/browser-brawl

Demo Video: https://youtu.be/NIoFXv-JvBY

(Skip to [0:55](https://www.youtube.com/watch?v=NIoFXv-JvBY&t=55s) to see the agents “brawling” in the arena :), [1:52](https://www.youtube.com/watch?v=NIoFXv-JvBY&t=1m52s) to see the browser traces generated)

Would love to chat with anyone building or training browser agents. Happy to dive in below!

投稿

Show HN: BrowseBrawl – What if browser agents battled to generate training data?

コメント