Hacker News Reader

Comment by AnotherGoodName | Hacker News Reader

And i wouldn't even say VCs are dumb because they keep hitting success chasing that money. It's why they have this money to throw around.

The labor market is USD$144 trillion worldwide in salary. Automate a fraction of a percentage of the worldwide labor? You have a company with more annual revenue than Apple.

Is it that hard to understand? Can people not think of scenarios where the current generation of AI may be able to automate some percentage of jobs? eg. Do people have any idea how much is spent on paralegals looking up case study law and how good LLMs are at finding such links in an automated way?

I keep seeing people with no idea of the market pushing bad ideas. "LOL CEO said 'AI' to make stock go up". "Nvidia can't possibly be worth more than it once was because AI isn't real". "It's hype and doesn't work". etc.

View on HN

> The labor market is USD$144 trillion worldwide in salary. Automate a fraction of a percentage of the worldwide labor? You have a company with more annual revenue than Apple.

What makes you think that AI/LLM tech is going to result in a single winner? That revenue might be split between a few hundred thousand companies.

Unlike the product from the current tech giants, there is no network effect AI stacks, there is no lockin, and there is absolutely no way to build a moat that can't be crossed with minimal capital.

This probably why Google sat on this for so long and refused to make any progress forward. It took openai to take Google's research and actually implement it. Google weren't going to do it because there's no moat to speak of

I don't think that's why Google sat on it. My guess is that it was hubris of the key researchers. They all had these castles in the sky of how it was "supposed to work" and then all that is required for quite powerful systems is scaling.

Only Ilya seemed to lack hubris and followed the path of scaling.

As for their being no moat, we don't know that. We are seeing a number of attempts at replicating GPT3/4 and there has been some success which is indicative that there is no moat, but this is a really new field.

> That revenue might be split between a few hundred thousand companies.

That’s not how power laws work. Over time, many niches will be created but usually it starts with one big winner taking it all. For example, first there where many half baked solutions, then there was Netflix and now we have all kinds of streaming providers, many of which are trying to find a niche such as Apple-owners or Disney fans. Similarly, we first had an iPhone and now small phones, big phones, tablets, touch laptops etc etc.

> Over time, many niches will be created but usually it starts with one big winner taking it all.

Usually, yes. But in this case it appeared to have started with multiple small providers, with multiple non-half-baked services. There doesn't appear to be a moat for anyone in this market.

Unlike streaming video (Netflix, et al) which has a moat (exclusive content), or social media (network effects), or ad industry (eyeballs).

With AI there is no moat that I can think of (quite happy to be provided with potential counter-examples, even if they are only hypothetical).

Let me try a different way to explain it: If you're building an ad network, a social media service, a SaaS dev-product provider, a video streaming service, etc, you will be unable to convince users to switch just by having a better product. The moat will keep them on the worse product (eyeballs, network effect, legacy support, exclusive content respecitively) no matter what you do.

With AI, there is no moat that I can think of - if you were to start an AI product today, what is the AI-specific moat you can use to keep out future competitors? Legislation, maybe?

I'm not sure that iPhones are a good example. It's a product, not a service. By definition there's no moat.

> With AI there is no moat that I can think of

I was also thinking about that yes. I think it’s mostly data (a sort of network effect), capex, and state-of-the-art engineering/research. A bit like TSMC and ASML. Also heavy capex and engineering moats.

only if they can keep training from the commons and stealing other people's work.

Even looking at a world where training data, models, etc. are all public domain or otherwise free to use for commercial purposes, you will still see a huge number of companies that specialize in AI "as-a-service" simply due to the size of the ML talent pool, the desire to mitigate risk and purchase a known quantity rather than "roll your own AI", and the nature of how new technology is implemented at scale.

Edit: and if copyright laws strike down current AI models and require some kind of BOM with established 'rights' to the data used to train models, you'll just end up with some variant of a gray market for training data that allows big corporations to do what they want with whatever data they can get their hands on, delegating risk to middle-man brokers making access to trained models EVEN MORE restricted to highly funded ventures.

Pretty soon, we'll be forced to wear glasses so that companies get revenue when we look at something copyrighted.

I understand copyright holder's complaints. Maybe we can meet in the middle - reduce copyright to 5-10 years.

Indeed, the current trajectory of discrediting the public domain leads towards a pay-per-thought society.

What is being stolen, exactly?

Knowledge

Reply on news.ycombinator.com