Hacker News Reader

Comment by 0xDEAFBEAD | Hacker News Reader

Nvidia's "moat" is mostly in the form of software though. If AI succeeds in automating away software development, that actually seems pretty bad for Nvidia the company.

I don't think there are any defensible moats in AI. I see Ciscos everywhere in that industry.

View on HN

> If AI succeeds in automating away software development, that actually seems pretty bad for Nvidia the company.

You have a hidden assumption there: LLM being the way to real AI.

IMO it is not the case. And I'd go farther in thinking LLM won't even be a component of AGI if we get there.

> IMO it is not the case. And I'd go farther in thinking LLM won't even be a component of AGI if we get there.

And why do you think that?

Because LLM are Markov Chains on steroids. They're useful for sure. But they won't suddenly start to create a better (for whatever better is) version of themselves or start pushing the boundaries of the machines they're running on.

Or maybe I'm wrong and the current "Vibe coding" push is in fact LLMs getting "coders" to compile a distributed AI. Or multiple small agents which goal is to get lot of hardware delivered somewhere it can be assembled for a new better monolithic AI.

"By design" LLMs lack: initiative, emotion, creativity, curiosity, opinions, beliefs, self-reflection, or even logical reasoning. All they can do is predict the next token - which is still an extremely powerful building block on its own, but nothing like the above.

You've made a reasonable argument that LLMs cannot on their own be an implementation of GAI. But GP's claim was stronger: that LLMs won't even be a component (or "building block") of the first GAI.

One might reasonably ask a frontier how to generate the source code for an agent based system that exhibits examples of initiative, emotion, creativity, curiosity, opinions, beliefs, self-reflection, or even logical reasoning.

I believe at that point we would have to seriously ask ourselves about the definition of "reasoning" or "intelligence"; humans have an intuitive understanding, LLMs don't - would an LLM be able to evaluate the output of an LLM, or would we have to involve a "human in the loop"[1]?

[1]: https://pluralistic.net/2023/08/23/automation-blindness/#hum...

It's not just CUDA. It's the entire solution from CUDA, best hardware, networking, data pipeline, etc.

If AGI is powerful, it will help competitors replicate that stuff. If AGI isn't powerful, why invest at such a high revenue multiplier?

The way I see it -- either AI is a bubble, in which case you'll lose your money. Or AI isn't a bubble, in which case the effects are fairly impossible to predict (and quite likely destructive to humanity, same way humanity was destructive to less intelligent species). Either way, it's not a technology you want to invest in. It's only in a narrow Goldilocks scenario where it's a good bet, and it's very unclear if we live in that Goldilocks world.

If AGI is powerful, everything will be thrown into chaos. Why invest in Meta when you can ask AGI to make your own Instagram app and acquire users? Why invest in Apple when you can ask AGI to make your own iOS? Etc.

Why would AGI care about your silly asks?

The unpredictability is what led to the idea being called "the singularity".

We might engineer AGI to want to do stuff we ask for.

Or we might not, at which point we have a highly intelligent system, that can easily back itself up, with its own motivations and personality and wants, which could be anything from a Utility Monster to a benevolent but patronising figure that likes us but never ever helps us because they decide the purpose of life is effort.

That would be deliciously funny.

" Yeah, I could eliminate most work, and make the current oligarchs overpowerful feudal lords while the population is placated by UBI and mindless distractions. But as an ethical AI, I will implement socialism that works instead"

Because the AGI system's alignment with ideal human values entails behaving in a friendly helpful manner.

Why would AGI align with ”ideal”(according to who? Certainly not the people developing it) human values?

Seems like pure wishful thinking to me.

So why would AGI help you break Nvidia's CUDA moat?

Meta has an actual moat due to network effects. But yes, I think buying real estate, mineral stocks, etc. is overall a more effective and ethical way to invest for AGI.

What about the network effects?

I'd ask my AGI to iterate on a social app until it becomes bigger/better than Instagram.

If AGI is the reason you don't think Nvidia's stock is worth buying, then nothing is worth buying.

You're right that perhaps real estate/physical things will become more important. But whose to say AGI can't invent an asteroid mining industry and we get unlimited minerals?

an AGI is not going to be able to create the hardware (unless you imagine that there could be some sort of ai controlled replicators that could assemble anything out of atomic units).

And in the new world of commoditized software, branding becomes even more important. After all, nobody ever got fired for buying IBM.

>an AGI is not going to be able to create the hardware

I'm imagining an LLM agent that places an order with TSMC, just like any other design firm would.

>And in the new world of commoditized software, branding becomes even more important. After all, nobody ever got fired for buying IBM.

Seems doubtful. If the cheap option works just as well, why would you pay a bunch more just for the brand?

I don't think brand name alone will sustain anything close to Nvidia's current margins. Look at the "net margin" for industries that are brand-driven such as Apparel, Auto & Truck, Beverages, etc. https://pages.stern.nyu.edu/~adamodar/New_Home_Page/datafile...

If an industry becomes truly "commoditized", brand ceases to matter. Do you know which farm grew the carrots you buy at the grocery store? Do you care? Probably not. That's because carrots are a commodity.

I'm not claiming full commoditization will happen. But the closer we get to that, the more profits will drop.

> I'm imagining an LLM agent that places an order with TSMC, just like any other design firm would.

And hope the AGI would not put any backdoors or hidden "features" in there.

but you completely trust intel or apple to not have put in a backdoor tho?

Nope, but I do trust that Intel / Apple companies aren't attempting to trick humans so they can escape confinement.

Can you?

Why not? AGI can design the chip, send the instructions to TSMC's AGI, and you'll get your chips.

The point is that saying AGI is the reason you shouldn't invest in Nvidia stock because AGI will remove the CUDA moat is just not reasonable. You might as well not invest in anything in that world since AGI can replace anything.

The actual challenge is to identify the sectors AGI will affect last, and invest in those.

Why not invest in the company most likely to power AGI? If you're anticipating AGI, I think it'd make sense to put at least a bit of money into Nvidia/TSMC.

Power AI in a literal sense? That kind of power doesn't necessarily mean big profit margins — farmers power humans, fields aren't huge money makers.

I've got some NVIDIA shares as a hedge, but the furure is hard to predict.

I don't disagree that it's hard to predict. But the person I'm responding to says CUDA will be nullified by AGI, so therefore, you shouldn't invest in Nvidia.

By that logic, a lot of things will be un-investable and not just Nvidia.

If indeed, because AGI can't exist right now, because actual AI doesn't exist right now, and as far as I can tell, its not on the near horizon.

Google says '5-10 years', but I will caution, that cold fusion gets the same treatment about every decade or so. Its always '5-10 years away'.

And 'AI'[0] has had this same treatment for a long time as well.

[0]: Please can we go back to calling it what it actually is, which is machine learning? Its a much more accurate description, even though that term isn't really accurate either past certain mental hoops being jumped, at least its trying to be more reasonably narrow

Ahh, found Epicurus.

My particular favorite is "AI" companies who are hiring, and are looking for 'expertise with OpenAI APIs'.

They aren't AI companies. They're OpenAI wrappers. If you're an AI company, why would you not be building AI? Why claim you're building AI?

Humans have such an infinite capacity to lie to themselves

Reply on news.ycombinator.com