[TheMrZZ]

> A single TPU 8t superpod now scales to 9,600 chips and two petabytes of shared high bandwidth memory, with double the interchip bandwidth of the previous generation. This architecture delivers 121 ExaFlops of compute and allows the most complex models to leverage a single, massive pool of memory.

This seems impressive. I don't know much about the space, so maybe it's not actually that great, but from my POV it looks like a competitive advantage for Google.

[cyanydeez]

it is. itll still not create AGI without some breakthrough in instruction vs data separation of concerns

[scottyah]

In what way do we not already have AGI?

[airstrike]

in all of them

[Der_Einzige]

Peter norvig is smarter than you are: https://www.noemamag.com/artificial-general-intelligence-is-...

[omcnoe]

Does non-artificial intelligence have clean instruction/data separation?

[lugu]

You can park a lot there. No offence but I love how AGI doesn't mean anything. It used to be that AI was a goal post. Now it is AGI. We could use characters from sci-fi culture to describe milestones. In order to achieve robocop level, we must solve the instruction vs data problem.

[lowbloodsugar]

Robocop was a human brain in a suit! Don't give them any ideas!

[lokar]

Thus it always was. I’m old enough to remember when “if AI could beat a grandmaster at chess” was considered the finish line.

[sanex]

Over fitting to the benchmarks since 1996

[emp17344]

Well, yeah… turns out that goal wasn’t a good indicator for AGI, so we re-evaluated. That’s changing your hypothesis in the face of evidence, not “moving the goalposts” in the fallacious sense.

[Mistletoe]

What’s the indicator for AGI now? We are so far past the Turing Test it isn’t funny. In fact the models now are too intelligent, you would never think a human would have that much knowledge quickly about a subject you chose at random.