@GlitchesRoux lol just train it to be awful and then eventually the good superposition will pop out and win 🫶 training data is full of happy endings
More notes on nearby ideas
Quote tweet · Apr 26, 2024more utopia less doom in the training data pls 🫶Reply · Nov 27, 2024@imitationlearn 🙄 are we all not here at least in part so we can be immortalized in the training data?Reply · Apr 25, 2023@simoj_ @acheronix lol! "mine's pretty good but you have to train it..." 😏Reply · Sep 4, 2025@lillybilly299 i label it and it usually performs way worse but i find it interesting & the llms want me to share it so it eventually gets back to…