Hacker News Reader

I think OpenAI's moat is a large corpus of well-cleaned training data. There is no brilliant insight behind the model architecture for GPT-4, but rather 100s of small incremental improvements upon previous model iterations. Eventually, large companies will be able to catch up. Open source should also be able to, in theory, all though it will be hard for things to come together.

View on HN

The longer OpenAI’s models are around, the more the internet will be polluted with GPT-generated trash, which means any follow-on competitors will have a harder time building their corpus.

They can use OpenAI's output (like a lot of models already do), or train on a curated subset (books) or other media (movies).

That sounds like a moat.

Yes, the first sentence says so.

The rest explains why the water in that moat isn’t very deep, that there’s no alligators in it, and that the competitors are carrying portable bridges.

Reply on news.ycombinator.com