[tootie]

Is anyone confident in what a winning strategy might even be? The market keeps getting more crowded and yet AI is still only creating incremental value for company's that adopt it. It's not clear to me that this is going to be quite as much of a sea change in how businesses run as people like Altman have been pitching.

Nor do any of the top 10 AI companies have any kind of moat. The fact that Elon Musk can found a competitor out of spite and have a plausible competitor in 6 months with a multi-billion dollar value actually just dilutes the perceived value any of the market leaders. OpenAI is still riding the high of being first and having the ChatGPT brand be so strong.

Google doesn't need to win on all the benchmarks, they just need to embed themselves in enough enterprises and they have a huge leg up in that regard.

[Workaccount2]

I think google will end up winning enterprise, and the fact that Apple didn't sign a sole partnership with OpenAI, but kept Gemini in the mix lends a lot of credence to this.

Google is the only "classic" org in the SOTA model space, and the only one in the whole race who doesn't have to kiss the ground Jensen Huang walks on. They are big enough to be able to "pay you back" if they colossally fuck up, and the chances of them going belly up are pretty slim. They also have the cheapest models to boot.

From a business standpoint, Google is the safest play on many levels, even if their models are just good enough.

[eitally]

The winning strategy is that Google has so many surfaces that they can embed their own AI in (also use as training data sources) that they essentially can't lose unless their models are just terrible. Fortunately, their models are great and they've truly been moving very fast to integrate AI into both enterprise (via Cloud) and consumer products this past year. As a xoogler, I'm truly impressed.

[stormfather]

I think the winning strategy is making a small model that's good enough to embed in search, etc, at no cost. Google has the best small models. But whoever wins this will be whoever makes a more efficient computing substrate for LLM inference, like running a transformer with lenses or something like that. Winning strategy there is acquisition.

[tootie]

You lost me. Lenses? Are you being fanciful or is this a thing?

[kridsdale1]

Maybe the substrate will be flesh.