What makes ChatGPT and Claude Great

Misha Laskin on why the best benchmark scores don't mean the best product: "When things really fit together, it's a tight coupling between a product and a model. It's not just the model alone." "ChatGPT was post-trained specifically for the prompts coming in from users. When I saw the first coding blog post it produced for me, that was insane." He thinks Claude Code is the next example of that.