Hacker News Reader

Hacker News

Comment by alberth | Hacker News Reader

Parent Full thread

alberth

Why doesn't Anthropic needs similar levels of capital (or do they)?

View on HN

Anthropic is more secretive about their costs, Ed Zitron is right now investigating their costs, specifically on GCP

Sure he is

because this is for building "AGI", this has little to nothing to do with their current offerings.

This also assumes that intelligence continues to scale with compute which is not a given.

> this is for building "AGI"

I’m increasingly convinced this is AI’s public relations strategy.

When it comes to talking to customers and investors, AGI doesn’t come up. At fireside chats, AGI doesn’t come up.

Then these guys go on CNBC or whatnot and it’s only about AGI.

I don't think it's AGI, but rather video production. OpenAI wants to build the next video social network / ads / tv / movie production system. The moat is the massive compute required.

I'm sure they're not against building this, and they definitely have competing priorities.

But my personal belief is Sam Altman has a singular goal: AGI. Everything else keeps the lights on.

Huh, my read is exactly the opposite: Altman wants to be a trillionaire and isn't picky about how he gets there. If AGI accomplishes it, great, but if that's not possible, "just" making a megacorporation which permanently breaks the negotiating power of labor is fine too. Amodei is the one who I think actually wants to build AGI.

Then why start a company where you have no equity? (Yes I believe he financially benefits from OpenAI, but the more straightforward way would be OpenAI equity)

I think his initial belief was that OpenAI would to be a research organization which would get acquihired or license its tech out, and then Chat-GPT unexpectedly happened. Notice how ever since then he's been trying to get the nonprofit status cancelled or evaded.

I think your read is right. There are a few people who want to be trillionaires and aren't too picky about to get there: Elon Musk, Sam Altman, Trump, Larry Ellison, Peter Thiel, Putin. Maybe Bezos and Zuckerberg.

Of course, there wouldn't be many people who don't want to be trillionaires. Rare exceptions[1]. But these are the people with means to get there.

[1]: No means NO - Do you want a one million dollar answer NO!: https://www.youtube.com/watch?v=GtWC4X628Ek

I definitely would not want to be a trillionaire yeah. Having a million or so would be nice but more and you get roped into all kinds of power play and you have to get security goons with you all the time to avoid getting kidnapped. I'd much rather be anonymous.

to be fair I couldn't recognize 99% of billionaires and crazy people are easy to deal with if you know what I mean.

To be fair, you're probably not an organized crime syndicate looking for targets either.

crime syndicates are just as afraid of people with that much money.

Is there any indication they are actually working on this and Altman is any good at pursuing this goal? I'm seriously asking, please inform the uninformed.

My impression is that I hear a lot more about basic research from the competing high-profile labs, while OpenAI feels focused on their established stable of products. They also had high-profile researchers leave. Does OpenAI still have a culture looking for the next breakthroughs? How does their brain trust rank?

Isn't that just PR?

> This also assumes that intelligence continues to scale with compute which is not a given.

Isn’t it? Evidence seems to suggest that the more compute you throw at a problem, the smarter the system behaves. Sure, it’s not a given, but it seems plausible.

No, that's no longer the case: https://www.newyorker.com/culture/open-questions/what-if-ai-...

> a problem

That word is carrying a heavy load. There's no evidence that scaling works indefinitely on this particular sort of problem.

In fact there is no evidence that scaling solves computing problems generally.

In more narrow fields more compute gets better results but that niche is not so large.

It also depends on the amount of training data, that isn't really growing much after they scraped all the internet.

it's not mathematically proven therefore it is not a given.

> it's not mathematically proven therefore it is not a given

Given doesn't mean proven, it means accepted as true. We can give variables fixed values, for example.

Entire classes of proofs, moreover, prove something cannot be true because if you assume it is you get a paradox or nonsense.

we had this problem in mathematics that's why all ambiguity was eliminated and why adding two numbers together is like 30 pages of proofs.

In a brute force poorly architected way, perhaps.

But human brains are small and require far less energy to be very generally intelligent. So clearly, there must be a better way to achieve this AGI shit. Preferably something that runs locally in the palm of your hand.

Anthropic seems more comfortable using TPUs for overflow capacity. The recent Claude degradation was largely due to a bug from implementation differences with TPUs and from their writeup we got some idea of their mix between Nvidia and TPU for inference.

I'm not sure if OpenAI has been willing to deploy weights to Google infrastructure.

I believe they do, but the author seems to focus on OpenAI since they're more of a household name.

Anthropic will need it if their growth continues.

Reply on news.ycombinator.com