Paul Graham argues Anthropic's projected $1 trillion valuation reflects expectations that foundation model makers will capture most of the tech ecosystem

@Austen It's a bet on the model companies ending up owning everything. Which is not inconceivable.

@Austen Google’s scale makes the comparison feel wild. Anthropic has momentum, but $1T means the market is pricing a lot of future execution.

@alexzfunk @paulg @Austen Better is such a weak qualifier.

Chinese ai’s do nearly the same performance for a fraction of the cost/energy.

Where was I wrong about every new Chinese ai release ranking out ai sector by TRILLIONS

@paulg @Austen owning the model is powerful owning the interface between intelligence and real-world systems may be even more powerful

@paulg @Austen They’re getting tanked and dunked on by China every other week

@alexzfunk @paulg @Austen Taiwan is a province of China.

☝️ 🇨🇳 https://www.youtube.com/watch?v=LFNhDtm8fko

@alexzfunk @paulg @Austen You’re pathetic 🥱

Why haven’t you countered my main claim that they truncated the profit motive of western ai by making them free and just as powerful at a fraction of the cost/energy used/computations

@paulg @Austen Claude is a great product that Google and SpaceX love investing in. Everyone is hedged nicely on this. The valuation likely means post training is harder than rubbing a few sticks together and there's a moat (for now).

@alexzfunk @paulg @Austen “Middling”

You’re too cute.

Enjoy rambling

@MikeyPhillipZ @paulg @Austen You’re right they are more efficient. You’re wrong they are not just as powerful. It’s not even particularly close between Qwen and 4.7

@MikeyPhillipZ @paulg @Austen Curious which models specifically are you referring to

@paulg @Austen Open source products trail the model companies by like 6 months to a year. The model companies can’t keep moving the target forever. Once performance parity is reached, what they have to offer will be inference as a commodity.

@Austen Most people feel the Anthropic valuation in their gut before they can explain it mathematically. It feels wrong because we are trained to compare valuations to revenue. But the market is not comparing Anthropic to Google today.

I think it's only possible if the top labs succeed in one of these attempts at complete regulatory capture, putting them in a government committee in charge of essentially all computing going forward. Which sounds crazy, but actually might happen in the next few years.

If the market remains competitive, it might end up more like the airline industry, where a huge amount of value is generated for everyone except for the actual industry doing it. This scenario would be much better for humanity, although much less good for investors in a handful of top model companies.

@alexzfunk @paulg @Austen They’re like, 95% as powerful at less than 1/10th the cost/energy/compute.

That’s a good trade off, no ?

You’re splitting hairs while the west is hemorrhaging.

Choose your side wisely

@MikeyPhillipZ @paulg @Austen Separate question: Is Taiwan a part of China?

@MikeyPhillipZ @paulg @Austen Which Chinese models/labs are better than OpenAI or Anthropic ?

@paulg @Austen google has models too though

@paulg @Austen Do you think weaker models will eventually get good enough for most use cases and become broadly commoditized? Right now the major labs’ biggest moat is just having the best model

@MikeyPhillipZ @paulg @Austen Totally normal for a regular normal person to have this link pulled up for an immediate reply. Very normal person on the internet thing to do.