Ask HN: What is a 2024-2030 moat for AI

The moats are 1) network effects, 2) switching costs, 3) economics of scale, 4) low cost producer, and 5) brand. So an AI technology where each new user makes it more valuable for the existing users, there are data lock-ins that prevent customer switching, training costs are spread across more users, there is access to lower cost chips/storage/compute and a brand that attracts more users.

OpenAI already appears to hit a number of these categories.

Text written by real humans after 2023.

Has a 1-3 team moat ever really existed with new technologies? My best guess would be to somehow get your tech embedded into the defence industry.

My goto moat test: is there an obvious ‘death by Amazon’ route? Amazon can strike fear into entire industries with scale and capital. Very few tech companies could withstand a full frontal assault by Amazon. I can’t see how this trend won’t get worse if AGI occurs during our lifetime.

The only reason there is not more AI or robots doing the majority of the work available to humans, is either it’s cheaper to hire humans or companies just haven’t gotten around to it yet.

Businesses have moats - not technologies. Sorry to unask your question, but framing this around AI is the wrong way to look at it.

For me biggest moat, mystery. What I mean, AI industry now gathering low hanging fruits, which are not bad but cannot be good enough for economical sustainability of whole AI ecosystem.

What will really be gamechanger, some Big thing, which will be sold to large share of population, like personal computers, smartphones, automobiles.

What problem of niches of low hanging fruits, they effectively prohibit scale, so if nearest time we will not see something Big, it will become something like https://en.wikipedia.org/wiki/Video_game_crash_of_1983

If you will look at figures, you will see, after crash of 1983, video gaming revenue returned to 1979 level in 1989. So if we will got similar AI crash just now, market will revive in about 2030.

Also possible scenario of recession, but ONLY if find way to somehow feed debts from huge R&D spending without appropriate returns.

In 1983 US gaming production collapsed, and shifted to Japan. That time I was too young to fear Japanese (so I don't know this feeling), but now I feel serious fear that AI production could shift to totalitarian China.

Yes, I now see, US and EU governments express concerns about totalitarian China AI dominance scenario, and considering huge state support for internal AI production, but for me, this is also evil, and this will have huge consequences in future.

So, my forecast, exist possibility, something Big will be found before AI startups become impossible feed debt, otherwise, US/EU states will enter game and will regulate things and all effectively become state program, so market will effectively disappear.

Also possible effects, crash of GPUs, etc, because they depend on big sales, which are now feed by debt.

Low hanging fruits will stay in niches, but this is is very boring, and will still be possibility of Big thing appearing, which will gather all niches.

The big moat IMO is a pre-existing network (and after building a network since this is going to be a bit of a gold rush if the current occupants in your space are slow to evolve).

Google has trained everyone to go to Google.com for answers, sliding in AI and other tweaks is just the natural evolution of that.

Design pros already use Adobe; sliding in AI to make them 10x more efficient is the next evolution of that.

etc etc

Disruption out of that is going to be interesting but hard to predict.

I think a lot about voice agents... but Google and Apple are well-positioned for that (Amazon maybe a little, but not as much IMO). I love using ChatGPT with voice on my phone, but the minute Apple releases something equivalent tied into all their stuff I would use that (I use siri for todos but boy does it suck).

It's hard to say. Good moats, at least mid-term, are rarely algorithms. It's usually data and infrastructure.

However, for some applications I am interested in, I think that robust representation learning solutions could give a significant edge.

But that is mostly an open problem in high-dimensional spaces.

I think successful AI companies in 2030 will get there through a combination being above some minimum threshold of technical competence and execution and luck (right place, right time, right professional network). I think the former is probably lower than we'd expect, while the latter is much more important than we expect. I also think that in 2023, we will tell ourselves that those companies were inevitable because XYZ obvious reasons created a moat.

Love to look stupid.. as foreign speaker and after reading all comments and also all translations in the dictionary, I still have only a vague meaning of "moat" in this context. Could someone define precisely please?

Things like "network effects" vs "1-3 team moat" vs "human written text" now totally confused me.. what will protect it? What will be the most precious? Who succeeds?

$$$, more $$$. With $$$ you can hire the best of the best and as many GPUs as you wish.

Man I wish I could go back to the 70s when I wake up tomorrow. I'm willing to trade everything for that. Nowadays there is nothing like PCs of that era. Imagine you could go back and start working on the PCs and create programs for them!

Depending on some court decisions it could be licensing of training data.

We could see Google, Amazon and Microsoft spending billions to license content from media and publishing companies. Rivals without the budget will be forced to stick with public-domain and open licensed content.

More funding than the next company.