[energy123]

It's a joint ignorance of how these frontier models get baked and what consumers want.

Many pundits think it's just a matter of scraping the internet and having a few ML scientists run ablation experiments to tune hyperparameters. That hasn't been true for over a year. The current requirements are more org-scale, more payoff from scale, more moat. The main legitimate competitive threat is adversarial distillation.

Many pundits also think that consumers don't want to pay a premium for small differences on the margin. That is very wrong-headed. I pay $200/month to a frontier lab because, even though it's only a few % higher in benchmark scores, it is 5x more useful on the margin.

[svnt]

It is the benchmark error rate, not the benchmark success %, that we actually trip up on.

Going from 85% to 90% is possibly 1/3 fewer errors or even higher, depending on the distribution of work you’re doing.

[lelanthran]

> The current requirements are more org-scale, more payoff from scale, more moat.

What moat? None of the AI providers have a moat at the moment, and the trend doesn't indicate that any of them will in the near future.

[energy123]

I made 2 posts in this thread regarding why I think they have a moat. Was there anything ambiguous or that you disagreed with?

[lelanthran]

> I made 2 posts in this thread regarding why I think they have a moat. Was there anything ambiguous or that you disagreed with?

I'm afraid I don't see those posts; I see 2x posts from you asserting they have a moat, but not why you think they have a moat.

I distinguish between "They have a moat." and "This is why $FOO, $BAR and $BAZ forms a moat."

Maybe you think brand recognition is a moat, but that didn't work out for incumbents before (too many examples to list).

[energy123]

It was kind of buried in my second post:

> They have internal scale and scope economies as the breadth of synthetic data expands.

These frontier labs will have a hundred or a thousand teams of people+AI working in parallel generating synthetic data to solve different niches. A few teams solve computer use. A few teams solve math. A few teams solve various games. So the org is basically a big machine that mints data, and model research is only a small part of it. Scale then is the moat.

The second leg of the moat thesis is that open weights competition will die off soon because the cost to keep up with the scale will be too excessive.

The third leg of the moat thesis is that customers are happy to pay big margins for differences that appear small if the benchmark is the measuring stick.

If the paradigm was still scrape internet -> train model, I'd agree that there is no moat.

[lelanthran]

I disagree that the model is a moat; distillation of models is going to happen, and even without it all the current players have models that are virtually indistinguishable for the use-case.

Model capbilities have converged over time, and I don't see this trend reversing. OpenAI owns only the model.

The provider who does have a moat is Google - they own the entire vertical, from the hardware, to the training data, they have it all.

OpenAI has to buy GPUs, Google makes them.

OpenAI has to rent data centers. Google owns them.

OpenAI has to scrape the web for all training data. Google's collection of user emails (not counting their Android data harvesting, ad data harvesting user-tracking, etc) alone gives them a ton of training data which will never be available to scrapers.

Google has billions of signed-in users, OpenAI has to market to and attract users (800m user count last I checked, but also last I checked that growth was asymptotic and flattening out).

Thats what a moat looks like. Better technology and/or results has never been, in my memory, a moat.

[energy123]

Good points about Google.

I think where I don't agree is about the model. You're mostly correct right now, and your view is supported by how close everyone is.

Where I am more optimistic about the 2-4 biggest labs (not just OpenAI) is what the next 2 years looks like.

I expect this to happen:

- Synthetic data goes from 30% of training data to 90-97%+ of training data.

- Synthetic data becomes hugely varied, and the production of it is factory-like and parallelized.

The moat here is the data factory, and the scale/scope economies behind it.

Thoughts?

[nick32661123]

You pay to OpenAI or which one do you use? Do you switch regularly?

[energy123]

I pay OpenAI but I would also be a happy Anthropic customer.

My view is that OpenAI, Anthropic and Google have a good moat. It's now an oligopolistic market with extreme barriers to entry due to needed scale. The moat will keep growing as the payoffs from scale keep growing. They have internal scale and scope economies as the breadth of synthetic data expands. The small differences between the labs now are the initial conditions that will magnify the differences later.

It wouldn't be surprising to also see consolidation of the industry in the next 2 years which makes it even more difficult to compete, as 2 or 3 winners gobble up everyone and solidify their leads.

When people worry about frontier lab's moat, they point to open weights models, which is really a commentary that these models have zero cost to replicate (like all software). But I think the era of open weights competition cannot be sustained, it's a temporary phenomenon tied to the middle-ground scale we're in where labs can still do that affordably. The absolute end of this will be the end-game of nation state backed competition.