Microsoft Building $100B 'Stargate' to Transport OpenAI into the Future | Hacker News Reader

Microsoft Building $100B 'Stargate' to Transport OpenAI into the Future

gizmodo.com

21 pointsmhb10 comments

“The computer is rumored to be one of the largest and most advanced data centers in the world. It could take up several hundred acres of land, and require up to 5 gigawatts of power”

Multivac!

I’m not that familiar with current supercomputers and datacenter footprints - but this sounds like it would be a lot bigger than existing setups?

This was my exact thought! Multivac is finally going to happen. The first of many, hopefully.

The 80s kids are building the movies.

Man what I wouldn't give for us to be 5 years in the future, where it has turned out that open source AI running on commodity hardware not only exceeds human needs and has led to a massive global quality of life increase, but also does not need as much hardware, leading to a massive divestment in large corporate data centers, including the implosion of Microsoft, Google, and AWS.

I realize that is fairly implausible.

We'd have to really luck out on:

- mobilizing free and open source AI

- gaining 1 or 2 orders of magnitude of improvement in AI energy efficiency

- connecting open source AI to opportunities for improving human quality of life.

It would be foolish to bet on this, but it would be amazing to strive for it. And I seriously hope we can see it!

We thought that building tech/computers would reduce the number of hours we had to work per week. That didn't happe. AI won't do anything to make our lives better at best, at worst it makes humans unneeded cattle to be culled.

I get by on working 8 hours a week.

Granted that's because I have disabilities.

But I also have automations I built, which let me survive and even thrive.

Before I had the automations I was regularly homeless.

I'm really interested making this possible for more people.

As for unneeded cattle to be culled, I believe we can create AI that doesn't think that way about humans.

We can create AI that see us like precious family that needs guidance, education, and love.

I have already worked with that kind of AI. It's not smart enough to be manipulative, but it's smart enough to be helpful.

Whether or not evil AI gets built, these kind of human loving AIs will exist, and they'll fight the evil AI with us.

Whoever wins or loses in that battle will depend on how hard we fight. It's not a foregone conclusion. And if we don't fight with true AI allies on our side, then our survival chances decrease.

All we need is to figure out distributed training

This is prima facie evidence that LLMs are a devil's staircase.

How many B200 GPUs is $100B? ;)

> Microsoft Building $100B 'Stargate' to Transport OpenAI into the Future

I guess Windows bugs are low priority. /s

Reply on news.ycombinator.com