I think the difference being an assumption that AI has a clear value proposition, which means that the investments as is are of substance and not just incestuous backdealing. Which is worth accepting on face, there are too many use cases for this technology that are already generating value, regardless of your opinion on superintelligence.
But surely the *magnitude* of the value proposition matters, as well as how much of that value OpenAI will capture in the long term. And both of those numbers are open for debate, rather than an open and shut case.
I think it boils down to “AI is not overhyped, therefore it’s not a Ponzi scheme.” Which an opinion, but he’s presented it as fact.
Very true, but it does put a floor on the bust unlike in a proper Ponzi scheme. Unfortunately, that floor still allows for trillions in losses, so we all have to hope to god that there’s at least some gold at the end of the rainbow.
It's not just the value of existing use cases, but the value of the infrastructure. Even if progress stopped and AI companies' valuations tanked, the money is being used on data centers which have value regardless. The GPUs will be useful for inference or non-AI workloads for a few years. The rest of it (buildings, cooling, backup power, etc.) will be useful for decades.
The same thing was true of the internet in the late 90s, though. It's not that it doesn't have value, it's about valuation.
My thesis is that betting on ASI long-term is a loser: either the labs fail and the bubble bursts (and at this point failing means not developing it until 2030), or they succeed and all the profits go to the ASI rather than shareholders - not to mention we're likely all dead.
"Roon: all the largest technology fortunes in the world are on record saying they’re betting the entire bank on superintelligence, don’t care about losses, etc. the only growth sector left in planetary capitalism. It is dizzying, running on redline overload.
If that statement does not terrify you, either you did not understand it, you did not think through its consequences, or you have become numb to such things."
No. It doesn't scare me. I may just be numb to tech's nonsense, but having any appreciation for history tells you how this will end. Hell, the history of AI alone tells you what is going to happen here.
One thing that confuses me about Nvidia: As far as I can tell, their "moat" is mostly in the form of software, which looks rather shallow in an AI bull scenario. You don't have to believe in superintelligent AI. You just have to believe in an automated software engineer that can code up high-performance software, matching a specification and passing a test suite. (Unambiguous objective functions are nice for RL, correct?) At that point, Nvidia's key assets become relatively easy to replicate as far as I can tell?
Even if it costs $1B of compute to replicate Nvidia's IP using Claude Code or whatever, that's still looks like a tidy profit given that Nvidia is valued at over $4T. What am I missing? Seems like everyone with a significant Nvidia position should be on hair trigger to sell.
“You do have to watch out for the de facto Ponzi scheme, but this is not that, from a pure business perspective it is a clear win-win agreement.”
I’m not clear — how exactly do we know that this is not that?
I think the difference being an assumption that AI has a clear value proposition, which means that the investments as is are of substance and not just incestuous backdealing. Which is worth accepting on face, there are too many use cases for this technology that are already generating value, regardless of your opinion on superintelligence.
But surely the *magnitude* of the value proposition matters, as well as how much of that value OpenAI will capture in the long term. And both of those numbers are open for debate, rather than an open and shut case.
I think it boils down to “AI is not overhyped, therefore it’s not a Ponzi scheme.” Which an opinion, but he’s presented it as fact.
Very true, but it does put a floor on the bust unlike in a proper Ponzi scheme. Unfortunately, that floor still allows for trillions in losses, so we all have to hope to god that there’s at least some gold at the end of the rainbow.
It's not just the value of existing use cases, but the value of the infrastructure. Even if progress stopped and AI companies' valuations tanked, the money is being used on data centers which have value regardless. The GPUs will be useful for inference or non-AI workloads for a few years. The rest of it (buildings, cooling, backup power, etc.) will be useful for decades.
The same thing was true of the internet in the late 90s, though. It's not that it doesn't have value, it's about valuation.
My thesis is that betting on ASI long-term is a loser: either the labs fail and the bubble bursts (and at this point failing means not developing it until 2030), or they succeed and all the profits go to the ASI rather than shareholders - not to mention we're likely all dead.
We don't know so
Podcast episode for this post:
https://open.substack.com/pub/dwatvpodcast/p/openai-shows-us-the-money
"Roon: all the largest technology fortunes in the world are on record saying they’re betting the entire bank on superintelligence, don’t care about losses, etc. the only growth sector left in planetary capitalism. It is dizzying, running on redline overload.
If that statement does not terrify you, either you did not understand it, you did not think through its consequences, or you have become numb to such things."
No. It doesn't scare me. I may just be numb to tech's nonsense, but having any appreciation for history tells you how this will end. Hell, the history of AI alone tells you what is going to happen here.
One thing that confuses me about Nvidia: As far as I can tell, their "moat" is mostly in the form of software, which looks rather shallow in an AI bull scenario. You don't have to believe in superintelligent AI. You just have to believe in an automated software engineer that can code up high-performance software, matching a specification and passing a test suite. (Unambiguous objective functions are nice for RL, correct?) At that point, Nvidia's key assets become relatively easy to replicate as far as I can tell?
Even if it costs $1B of compute to replicate Nvidia's IP using Claude Code or whatever, that's still looks like a tidy profit given that Nvidia is valued at over $4T. What am I missing? Seems like everyone with a significant Nvidia position should be on hair trigger to sell.
Why is NVidia's moat software? They're primarily a hardware company.
It's the software they ship that you use to control the hardware that differentiates them.
https://developer.nvidia.com/blog/even-easier-introduction-cuda/
https://www.technocracy.news/