Why the next leaps towards AGI may be "born secret" • Hacker News Board

Re: Why the next leaps towards AGI may be "born secret"

#11

Post by RcouF1uZ4gsC » Sat, Nov 23, 2024, 5:06 AM UTC

Is there a betting market for this?

I don’t care who is researching this, we won’t have AGI by 2027 and super intelligence by 2030s.

> Based on trends in AI capabilities research since GPT-2, we are on course to expect AGI by 2027. Once AGI capability is available, if labs focus on automating AI research itself, progress in AI should accelerate. If similar progress can be achieved as the phase from GPT-2 to GPT-4, or GPT-4 to AGI, we should expect Superintelligence before the end of the decade.

Re: Why the next leaps towards AGI may be "born secret"

#12

Post by Tostino » Sat, Nov 23, 2024, 5:15 AM UTC

didibus wrote: post #10

TacticalCoder wrote: post #8

"born secret", maybe. But by a US government program? Lately besides corruption and inefficiencies, the government didn't have much to show. NASA go trounced hard by SpaceX. The Kamala "broadband" project did cost $42 billion to taxpayers and not one home has been connected with this new program. In nearly four years. Zero. Meanwhile StarLink is actually connecting people. Be it NVidia, OpenAI, Meta, Google, etc. all…

> The Kamala "broadband" project did cost $42 billion to taxpayers It didn't, the money has not been spent yet.

Don't let facts get in the way. GP was just trying to be inflammatory and you are here ruining everything.

Re: Why the next leaps towards AGI may be "born secret"

#13

Post by Reubend » Sat, Nov 23, 2024, 7:07 AM UTC

AtlasBarfed wrote: post #5

Reubend wrote: post #4

It's a good idea. I don't think secrecy is really necessitated aside from the idea that if the US government is the one spending the money, then they should reap the benefits. Ideally, I'd love to see many governments collaborate here and share the technology with each other as a result.

What is scary about this memo, and contrary to your "lets be friends" outlook, is that we are entering a geopolitical era of precisely the opposite. The US is withdrawing from its global cop role. Russia and China are saber rattling. NATO may be dissolved by Trump, at least the American side of it. Israel and Saudi Arabia are new strange bedfellows and allies. Japan and India are vastly increasing their military spen…

> What is scary about this memo, and contrary to your "lets be friends" outlook, is that we are entering a geopolitical era of precisely the opposite.

That certainly may be true! Or perhaps, the cost of not collaborating is now too great to ignore.

Re: Why the next leaps towards AGI may be "born secret"

#14

Post by a2128 » Sat, Nov 23, 2024, 4:20 PM UTC

RcouF1uZ4gsC wrote: post #11

Is there a betting market for this? I don’t care who is researching this, we won’t have AGI by 2027 and super intelligence by 2030s. > Based on trends in AI capabilities research since GPT-2, we are on course to expect AGI by 2027. Once AGI capability is available, if labs focus on automating AI research itself, progress in AI should accelerate. If similar progress can be achieved as the phase from GPT-2 to GPT-4, or…

IIRC OpenAI has a clause in their agreement with Microsoft that they can terminate sharing models if they develop AGI. So it might not be a good bet unless you believe there will never be shenanigans (either "we developed AGI but Microsoft won't let us declare it" or "we developed 'AGI', now we're free from Microsoft")

Re: Why the next leaps towards AGI may be "born secret"

#15

Post by tim333 » Sat, Nov 23, 2024, 7:54 PM UTC

A "Manhattan Project to build AGI" seem pretty questionable. With the actual Manhattan project it was only the government that was able to work on the bomb as they had most of the resources. To this day it's hard to do a startup to knock out better nuclear warheads.

Whereas with AGI loads of companies are already on it and there isn't really reason to believe that a government entity would get there faster.