Comment by rainprincess

I feel like it's because of the way they are internally structured. They have some people who are machine learning trained on post-training, and other people who have nice product management resumes. The post-training people want more compute, well-formatted data, honestly more ways to try whatever technique of reinforcement learning they want.

The product people are building things, but OpenAI has literally been throwing stuff at the wall and it hasn't been sticking. They seem to be behind in terms of everything user interface. Canvas came after Anthropic had artifacts. Codex came after Anthropic had Claude Code.

Some of their researchers (okay one) have (has) stated they believe in interface work. That's because GUIs help engage the person beyond thought, and help the person work with more complexity (perception, physics, form, 3D). But they're playing catchup, or they're trying to incubate wins in science / math.