@felurian8 @BasedBeffJezos latent space is the model's internal representation of what it's learned from all the training data, compressed via the training cycles to capture the most important features
More notes on nearby ideas
Reply · Jul 31, 2023@mage_ofaquarius maybe... idk, it's like pawing thru the latent space directly- kinda disturbingReply · Aug 18, 2023@HTTPSNoWhere idk, I feel like manic and depressed episodes are each their own vibe that would get trained into the latent space... would be…Reply · Apr 23, 2024@jd_pressman i love ur threads; incidentally always good to see past-me correctly evaluated the OP's intellectual seriousness 😏Reply · Nov 27, 2023@Nominus9 ahhh yeah, lol interpretability stuff... 120 layers, 100K different tokens, what is really going on inside the weights wrt features & world…