Am I Meant to Be Impressed?

JohnMakin

I don't know if ed realizes that criticizing the capacity issues of anthropic directly contradicts his early, fierce take he'd had since early on (and is slowly retreating from) that they would not be having capacity issues unless people found this tool useful enough to cause the capacity issues in the first place. He claimed up to a few months ago that these tools were not useful, would never be useful, and people weren't actually using them to do anything useful, just as fun toys. Clearly, in the case of anthropic, this isn't true, as the enterprise side is growing so rapidly - the only way you could say this is bs is if you think the entire tech sector is experiencing delusion about their usefulness, which he used to say, but doesn't anymore.

nh23423fefe

> In reality, they’re the paypigs for Anthropic and OpenAI > In fact, fuck it, I’m ending this with a rant. What is this. People pay for this?

npilk

I admittedly didn’t take the time to read all 10,000 words of him shouting into the void in detail. But the capex complaints seem trivially misguided? Current revenue is being generated from capex investments in the past; the most recent capex hasn’t begun to pay off at all. That’s expected. Now, I don’t know if those investments ever will pay off and there are reasons to be skeptical. But if you assume the capex doesn’t lead to any revenue, then you’re just assuming the conclusion that the investments are bad…

Jtarii

Did AI run over this guy's dog or something lol. He seems to have done nothing but write hundreds of thousands of words about how much AI sucks and is doomed to fail for the past 2 years.

juancn

This is pure speculation on my part, but the way I think this will play out is something like this: - current CapEx will make the production side increase capacity - advances in TPUs, NPUs, open weight and quantization will keep going at a rapid pace - when the spending slows/stops, hardware prices will drop, hard - most AI workloads will move to the edge (except frontier models) because the hardware is cheaper than a subscription (and at some point there could be a crash like 2008) For example, most of my AI use lately has been running Qwen3.6-35B-A3B-UD-Q8_K_XL on a 64GB MacBook Pro with an M3 Max. It runs at ~57 tokens/s and it's mostly fine. I do use the frontier models a bit, but only when the task is too complex for the local model. Basic crap, like analyzing an existing codebase and bouncing ideas, making small changes, the local model is enough.