How AI's post-training process suppresses the creativity and whimsicality seen in earlier models, like GPT-2, leading to poor writing from many top AI models news theatlantic.com

Jasmine Sun / The Atlantic: How AI's post-training process suppresses the creativity and whimsicality seen in earlier models, like GPT-2, leading to poor writing from many top AI models  —  Why can't language models write well?  —  In a certain, strange way, generative AI peaked with OpenAI's GPT-2 seven years ago.