#"Sam Altman Seeks Trillions of Dollars to Reshape Business of Chips and AI"

1 messages · Page 1 of 1 (latest)

ruby pewter
#

So much for "hardware overhang" as an argument for rushing AI development!

subtle lotus
#

Good observation, not what I thought of initially

#

If banks invest in this, that would be a good reason for a protest

subtle lotus
#

The global gdp is about 100 trillion btw, this is pretty crazy

lofty locust
#

This story is obviously fake, not even the US government has that much money to spend

austere cloak
#

It's not obviously fake to me. Yes, the numbers are unprecedented. Yes, most reporting so far is downstream of the single WSJ piece. But e.g. Nathaniel Whittemore's take on the AI Breakdown podcast today is plausible. This is a global scale idea requiring support across governments. You talk to UAE because you need them as well as the US.

#

I know that the OpenAI approach to reducing x-risk is race not pause at least for now, and that we differ in that.

But despite the obvious capabilities angle, is there any chance this is nonetheless an ambitious safety plan?

A realistic approach to implementing compute safety measures (audit, limits, off switches) through chips is helped by coordination on this kind of scale.

steep dew
#

7 trillion is greater than the amount that I estimated to build AGI. This should be enough to make a serious attempt at ASI. Many people have told me that no one would be willing to pay this much for AI, this story suggests otherwise.

#

Also, isnt the compute overhang arguement debunked?

ruby pewter
# steep dew Also, isnt the compute overhang arguement debunked?

"debunked" among people like us who were super skeptical to begin with, but really there have only been counterarguments as far as I am aware. But when Sam Altman made the hardware overhang argument in public appearances (e.g. on Lex Friedman's podcast) and then starts directly accelerating hardware, that goes beyond him having a model of reality I disagree with and into the realm of he wasn't really being sincere to begin with.

lofty locust
#

I think the hardware overhang argument is valid

#

But it depends on having timelines longer than 5 years