AI-Generated Homogeneity
This is a really great write up from Gwern on LLM ensloppification/mode collapse, its implications for alignment, and how to dis-incentivize it
Incidentally, I spent last semester working with @aidan_mclau and @heyanuja to tackle this exact problem with AidanBench, which we’ve turned into a proper benchmark and are rele... See more
Honour your creativity
In many ways, I feel fortunate that we built Poolsuite, Vacation® and other projects before AI became as accessible as it is now. That time gave me a runway to hone the skill of being creative. But the other day, I found myself slipping into a terrible new habit, and turning to ChatGPT before I even gave my brain 30 seconds to... See more
In many ways, I feel fortunate that we built Poolsuite, Vacation® and other projects before AI became as accessible as it is now. That time gave me a runway to hone the skill of being creative. But the other day, I found myself slipping into a terrible new habit, and turning to ChatGPT before I even gave my brain 30 seconds to... See more