Remix.run Logo
▲ kapkapkap 19 hours ago

Anthropic stock price going down has nothing to do with what you think about the future of AI.

If you think that the open weight models are just 6 months behind (And will eventually catch up as progress eventually plateaus), then even if you think AI will take every job in the world, etc...Anthropic does not end up as one of the 10 most valuable companies in the world in this scenario. Far from it.

Ask Claude itself how far behind it thinks Google and OpenAI's models are and it will tell you "probably a few months at most".

▲827a 18 hours ago | parent | next [-]

I think the idea that "stock price has nothing to do with what you think about the future of AI" cuts both ways. Beyond a point I really don't think Anthropic's model performance is a strong impactor on their revenue, in much the same way that the quality of Outlook beyond a point does not seem to impact Microsoft's ability to sell M365 subscriptions. If its good enough, the momentum + network effects of the B2B corporate coordination systems they build on top of the model are all that really matter to justify their current valuation.

▲hrncn 17 hours ago | parent [-]

The problem is that this is only the case because you can either buy Anthropic or nothing, because the only other pureplay AI competition (OAI) decided to postpone the IPO. That is where the current open valuation comes from. The top models are more or less fungible and there is no way to capture the b2b side like MS, if the safeguard bullshit or some other stuff continues to act up the switch is pretty easy.

▲827a 16 hours ago | parent [-]

My point is that Anthropic is not the pure AI play that people seem to think it is. At the end of the day there are no pure AI plays. Their revenue is not coming from mathematicians solving unsolved problems and the invention of new pharmaceuticals (yet). Their revenue is coming from companies like Eli Lilly paying them a half billion dollars a year to provide for them B2B software. A hundred Lilly employees could say "we hate Outlook switch us to Google" and their leadership will not listen. Similarly, Opus 5.5 is far past the point of being useful and powerful enough that customers are less interested in capabilities, and more interested in "how can you help me deploy those capabilities". No one has a better story for that than Anthropic, and I don't think anyone else is getting close either.

▲nightski 19 hours ago | parent | prev | next [-]

Not only that we are one model innovation away from them being worthless. You need $1T in compute because transformers are so ungodly inefficient. That compute will still be useful even if we find a more efficient architecture, but the margin on it will be so much less.

Deepseek was a sign of this on a smaller scale.

▲xmcp123 18 hours ago | parent | next [-]

Can you expand on that deepseek part?

▲zrn900 9 hours ago | parent [-]

Deepseek concentrated on optimizing inference and it cuts up to 90%. Then published papers on how to do it. Everyone adopted it. It came as a shock to the AI bubble in the US because they bet everything on high need for compute.

Deepseek kept optimizing. Their latest paper said that by treating text as images, even more optimization could be possible as LLMs process them faster. I don't know whether they incorporated it yet.

So the parent commenter is correct: The AI bubble bet everything on the immense need for compute. It not only turned to be unprofitable, but it was also obsoleted within 1.5 years.

▲logicchains 18 hours ago | parent | prev [-]

>That compute will still be useful even if we find a more efficient architecture

Mathematically there's no way to have lossless attention without O(n^2) cost; some magical massively more efficient architecture is not going to pop up the same way nobody's going to find a way to sort an arbitrary list in log(n) time. There may still be more opportunities for incremental gains (like using the bare-minimum number of full-quadratic-attention layers necessary), but they're not going to be orders of magnitude more efficient without sacrificing intelligence.

▲vlovich123 17 hours ago | parent [-]

Counter proof by existence - humans use significantly less energy and don’t do intelligence through an n^2 process. The human brains seem to operate closer to SSMs like Mamba with a recurrent architecture and attention is implemented using sparse distributed weights. So yes, it’s possible to have very high intelligence without O(n^2). Is it as good as transformers? Hard to tell right now.

▲machina_ex_deus an hour ago | parent | next [-]

Humans do something even "worse" - all their "weight matrices" e.g. neurons activation patterns are constantly changing while they think. If every llm had to have it's weight matrix updated during inference, that would be orders of magnitude more costly than merely having n^2 attention.

▲ashkankiani 15 hours ago | parent | prev [-]

he said specifically lossless attention. As in that algorithm. what you just said is irrelevant to the conversation because all of the current AI hype is built on the back of attention

▲vlovich123 4 hours ago | parent [-]

Here’s what was literally quoted in the response:

> That compute will still be useful even if we find a more efficient architecture

Fixing N^2 attention would require a different architecture, but the person I was responding to was treating attention as if it’s the only architecture we have for intelligence when we know that not only do we have less competitive intelligence for cheaper, human intelligence doesn’t use n^2 attention. You have also fallen into the trap of misunderstanding the statement same as the person I replied to.

▲watwut 19 hours ago | parent | prev | next [-]

Why would anyone take Claude answer to that question seriously?

▲jqbd 19 hours ago | parent | next [-]

Because even the most biased, optimistic source proclaims doom of Anthropic, in the grandparent's view at least.

▲cyanydeez 19 hours ago | parent [-]

humans in aggregate are largely stupid and myopic.

here's the paradox de juor: https://mathparadoxes.com/ergodicity/

Aggregating beliefs and saying "this is true" works only in vert general ways.

▲randomNumber7 16 hours ago | parent [-]

If you think s.o. is stupid because he doesn't intuitively understand this game you are. The people who got that right learned how to do the math long before answering.

Also one could always argue in favor of a +EV bet and it would be way better than the lottery.

▲wavemode 18 hours ago | parent | prev | next [-]

If you can't take Claude's answer to that question seriously, that in itself is an indictment of Anthropic's valuation.

▲randomNumber7 16 hours ago | parent [-]

No. It just shows you have some understanding of the technology.

▲ImPostingOnHN 19 hours ago | parent | prev | next [-]

The same reason anyone would take any Claude answer to any question seriously.

There's nothing special about that question.

▲nairboon 19 hours ago | parent | prev [-]

Well yesterday the news reported that Claude is basically worth $2T (they later fixed that headline), so surely something worth 2'000'000'000'000 USD is to be taken seriously? /s

▲greggoB 19 hours ago | parent [-]

> yesterday

Today down to $1.2T [0].

So you should change "is" to "was", or apply a factor of 0.6 to however seriously you thought it should be taken.

[0] https://shattered.io/anthropic-ipo-valuation-buyers-1-2-tril...

▲nairboon 18 hours ago | parent [-]

I'm merely pointing out the discrepancy between tech insiders not trusting Claude's output and public markets briefly valuing it at $2T (now only $1.14T).

I didn't take the $2T valuation serious, hence the /s marker.

▲greggoB 14 hours ago | parent [-]

So it's noise compared to noise? xD

▲dubcanada 17 hours ago | parent | prev [-]

Ask Claude how long it thinks it will take to build anything and the answer is bananas. It will say things like it will take 5 weeks to build a what ever app, and then build that app in an hour. Timelines from Claude are basically a waste of tokens.