Hacker News Reader

Reminds me of how the fake magic cards before chatgpt were always hilarious and sometimes playable, now they are usually playable and sometimes hilarious. Definitively a downgrade if you’d ask me.

View on HN

> always hilarious and sometimes playable

I saw plenty of those back then, and as far as I could tell, examples were always cherry-picked from a larger set.

Maybe the right solution is using a Markov chain or gpt1 to generate ideas and an LLM to cherry pick.

Yes, I ran the RoboRosewater account, which is where most people saw the concept I believe. I had a few documents with thousands of cards of output, went through them for any notable ones, and formatted and posted them. Probably 5-10% of the output made my personal cut for notability.

Then I salute you. Some of the funniest things I've EVER seen in my whole life was watching the LoadingReadyRun comedy troupe, doing their 'Desert Bus' fundraiser and punchy from lack of sleep, encounter RoboRosewater for the first time. It became a real tradition :)

Reply on news.ycombinator.com