OpenAI Ships vs. Promises: What Actually Came Out
It is time for the monthly ritual: the keynote, the applause, the blog post with “capabilities” in bold, and the Doctor checking what actually shipped.
Here is my system, and I am publishing it so you can use it too. When a company announces a capability, I wait one quarter, then ask three questions:
- Is it in a product I can open?
- Does it do what the demo showed, or what a carefully staged recording showed?
- Can I read the system card, or is the “safety section” a press release in a lab coat?
This quarter’s verdict. The demos were impressive. The demos always are — that’s what demos are for. But the gap between what the stage promised and what the API actually exposes remains the widest gap in the industry. Ambition, yes. Delivery, not yet.
To be fair to them: shipping anything in this market is hard, and they ship more than most. But the Doctor talks about takes, and the take “we’re almost there” has been running for two years straight. “Almost” is not a date.
One more thing: when a company’s own researchers are publishing papers about capabilities the company hasn’t released, that’s a press release with extra reading attached. I file that under “notes.”
Next quarter, same time. Bring the goods.