OpenAI is making a wild number of product plays at once, trying to leverage the value of the frontier model, brand value, and massive number of eyeballs they own. Sora is just one of many. Some will fail and maybe some will succeed.
It seems true that no company has used frontier models to create a product with business value commensurate with the cost it takes to train and run them. That what OpenAI is trying to do with Sora, and with Codex, Apps, "Agent" flows, etc. I don't think there's more to read into it than that.
On some level they know that LLMs alone won't lead to AGI so they have to take a shotgun approach to diversify, and also because integrating some parts of all these paths is more likely to lead to the outcome they want than going all in on one.
Also because they have the funding to do it.
Reminds me a bit of the early Google days, Microsoft, Xerox, etc,
This is just what the teenage stage of the top tech startup/company in an important new category looks like.
The massive cost of this product is unique though (not even counting the copyright lawsuits/settlements coming). I can't think of any side projects that require this level of investment.
>It seems true that no company has used frontier models to create a product with business value commensurate with the cost it takes to train and run them. That what OpenAI is trying to do with Sora, and with Codex, Apps, "Agent" flows, etc. I don't think there's more to read into it than that.
It makes them look desperate though. Nothing like starting tons of services at once to show you have a vision
> It seems true that no company has used frontier models to create a product with business value commensurate with the cost it takes to train and run them.
Anthropic has said that every model they've trained has been profitable. Just not profitable enough to pay to train the next model.
I bet that's true for OpenAI's LLMs too, or would be if they reduced their free tier limits.
> OpenAI is making a wild number of product plays at once
It's similar to the process of electrification. Every existing machine/process needed to be evaluated to see if electricity would improve it: dish washing, clothes drying, food mixing, etc.
OpenAI is not alone. Every one of their products has an (sometimes superior) equivalent from Google (e.g. Veo for Sora) and other competitors.
It seems true that no company has used frontier models to create a product with business value commensurate with the cost it takes to train and run them. That what OpenAI is trying to do with Sora, and with Codex, Apps, "Agent" flows, etc. I don't think there's more to read into it than that.