REVEALED: Open A.I. Staff Warn "The progress made on Project Q* has the potential to endanger humanity" (REUTERS)

[Jaguar93]

 

[Sauce Mane]

I’ll punch that soft ass bullshyt in the mouth, say “EARLY!”, then walk away.

 

[987654321]

 

[Professor Emeritus]

Y’all really think robots can take over?

That’s hard for me to believe, ai is definitely going to have consequences but is it going to be some matrix or terminator shyt I don’t think so

It’s hard for me to believe that ai wouldn’t have a kill switch

And why do people assume ai would be on bullshyt with humans

Imagine you design a "kill switch" for me. But I have a 160 IQ and you have a 100 IQ. How long before I figure out how to disable your kill switch or otherwise render it irrelevant?

Now, when we're talking about AI, forget about that 160 IQ and make it 300 IQ, or 400 IQ, some shyt beyond any human comprehension. Because that's probably where we're headed.

Even say that you design a perfect kill switch that absolutely gives humans the upper hand. Well, who says that no human is going to side with AI and fukk up the system so AI can bypass it? There's always going to be that evil genius who thinks he's smarter than everyone else and can control the consequences. If Putin or Trump (or someone like them) started losing a really serious war, you can't imagine they'd allow AI to run wild in a last-ditch attempt to save their own ass? Or even just some random scientist who wanted to watch the world burn?

 

[bnew]

You mean like how the current administration is forbidding high end chips from being sold/exported to other countries?


But that's still more of a competitive move than anything else.. but yeah, you're right.. the best defense we have is to continue to develop this tech as a leader in this space, but that comes with its own set of problems and could/will have negative effects on the rest of the world.. not just us.

China keeps buying hobbled Nvidia cards to train its AI models

Nvidia’s weakened processors are still more powerful than the alternatives.

arstechnica.com

those policies are just bumps in a road

middle eastern countries are releasing decent LLM's too.

Spread Your Wings: Falcon 180B is here

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

huggingface.co

US restricts exports of Nvidia AI chips to Middle East

Controls apply to A100 and H100 chips, in escalation of US efforts to curb China’s access to products

www.theguardian.com

Saudi Arabia and UAE race for Nvidia chips to power AI ambitions

Saudi Arabia and the United Arab Emirates have embarked on a race to acquireNvidia chips, crucial components for developing AI

aibc.world

 

[Professor Emeritus]

research has been ongoing for decades.

And yet not a single law has been passed that would control or contain their results. Not a single industry standard or guideline that would prevent a potential disaster from happening. Not a single principle set in place that anticipates any of the ways in which AI could destroy society. Even something as simple as Chat GPT came out, and people didn't know what the fukk to do in terms of how to regulate it, how to set limits, how to keep it from disrupting industries and elections.

My friend's son is just starting out in his career and trying to break into the AI existential threat circles as a researcher, so I know they exist. Even some of the biggest funders of the industry have been calling for a pause so they can figure out some sort of regulation and guardrails. But the people who run shyt like Open AI on the profit side generally don't give a shyt. They haven't made any serious commitment to important guidelines or answered any of the ethical questions. They're just going into commercialization as fast as they can, and when the nonprofit side of things holds up their hand and says, "Wait, maybe we need to put the brakes on this for a moment", they swing their nuts and show them it's their way or the highway.

 

[Yzak]

People talking about robots like we’re about to be living in a world like Demolition Man, when the real danger to humanity is AI’s ability to manipulate news and public perception.

Fake and misleading news can end up leading to violence, runs on banks, runs on retail stores, and all types of other negative effects.

Talking about robots and movies are a distraction from what these people truly want to do.

Imagine what these racist cacs and asians are gonna teach their AI. We gotta make our own pro-Black AI

 

[Jazzy B.]

They're designing and iterating something that can think and process a thousand times faster than an actual person with zero emotion.

What could go wrong

 

[bnew]

So all this time, we was warned them browns and Koreans should not have nukes because they’ll endanger the planet, paved the way for AI created in US to possibly hack in to the nuclear systems or cause wars

theres no hack for access to nuclear weapons, they still use floppy disks.

edit:
they updated it

The US nuclear forces’ Dr. Strangelove-era messaging system finally got rid of its floppy disks

The 1970s era Strategic Automated Command and Control System is slowly becoming more modern, and a replacement could be on the way.

www.c4isrnet.com

 

[bnew]

And yet not a single law has been passed that would control or contain their results. Not a single industry standard or guideline that would prevent a potential disaster from happening. Not a single principle set in place that anticipates any of the ways in which AI could destroy society. Even something as simple as Chat GPT came out, and people didn't know what the fukk to do in terms of how to regulate it, how to set limits, how to keep it from disrupting industries and elections.

My friend's son is just starting out in his career and trying to break into the AI existential threat circles as a researcher, so I know they exist. Even some of the biggest funders of the industry have been calling for a pause so they can figure out some sort of regulation and guardrails. But the people who run shyt like Open AI on the profit side generally don't give a shyt. They haven't made any serious commitment to important guidelines or answered any of the ethical questions. They're just going into commercialization as fast as they can, and when the nonprofit side of things holds up their hand and says, "Wait, maybe we need to put the brakes on this for a moment", they swing their nuts and show them it's their way or the highway.

no one outside the agreed parties ae obligated to pause any development. no one wants to be be insignificant within the market as well since anyone who pauses will have a difficult time retaining valuable employees.

 

[aXiom]

China keeps buying hobbled Nvidia cards to train its AI models

Nvidia’s weakened processors are still more powerful than the alternatives.

arstechnica.com

those policies are just bumps in a road

middle eastern countries are releasing decent LLM's too.

Spread Your Wings: Falcon 180B is here

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

huggingface.co

US restricts exports of Nvidia AI chips to Middle East

Controls apply to A100 and H100 chips, in escalation of US efforts to curb China’s access to products

www.theguardian.com

Saudi Arabia and UAE race for Nvidia chips to power AI ambitions

Saudi Arabia and the United Arab Emirates have embarked on a race to acquireNvidia chips, crucial components for developing AI

aibc.world

you said we can't do anything about it.. i provided examples.. they're having to resort to purchasing gimped datacenter and consumer cards and wont have access to any high end cards while nvdia/tsmc cant spit out H100s fast enough for US comapnies.. the chips act was further expanded to ban the exports of even the purposefully cut-down chips that were designed for the Chinese market as AI is deemed a matter of national security.. No A100/H100s no AMD Instincts.. all the good LLMs are based on shyt that we released.. that just ensures that we remain the leaders in this space.. ..

it's the China manufacturing vs engineering issue all over again. Standing on the shoulders of giants doesn't make you one.

But being the leader comes with a certain responsibility.. the rest of the world basically has to trust that we don't cause the end of modern society as we know it.

 

[Paper Boi]

 

[bnew]

you said we can't do anything about it.. i provided examples.. they're having to resort to purchasing gimped datacenter and consumer cards and wont have access to any high end cards while nvdia/tsmc cant spit out H100s fast enough for US comapnies.. the chips act was further expanded to ban the exports of even the purposefully cut-down chips that were designed for the Chinese market as AI is deemed a matter of national security.. No A100/H100s no AMD Instincts.. all the good LLMs are based on shyt that we released.. that just ensures that we remain the leaders in this space.. ..

it's the China manufacturing vs engineering issue all over again. Standing on the shoulders of giants doesn't make you one.

But being the leader comes with a certain responsibility.. the rest of the world basically has to trust that we don't cause the end of society as we know it.

it's not only about hardware though, i suspect a lot of the software to train models aren't as efficient as possible. this policy remains 'effective' only until the software performs better on less powerful hardware.

[2311.10770] Exponentially Faster Language Modelling

Computer Science > Computation and Language​

[Submitted on 15 Nov 2023 (v1), last revised 21 Nov 2023 (this version, v2)]

Exponentially Faster Language Modelling​

Peter Belcak, Roger Wattenhofer

Language models only really need to use an exponential fraction of their neurons for individual inferences. As proof, we present UltraFastBERT, a BERT variant that uses 0.3% of its neurons during inference while performing on par with similar BERT models. UltraFastBERT selectively engages just 12 out of 4095 neurons for each layer inference. This is achieved by replacing feedforward networks with fast feedforward networks (FFFs). While no truly efficient implementation currently exists to unlock the full acceleration potential of conditional neural execution, we provide high-level CPU code achieving 78x speedup over the optimized baseline feedforward implementation, and a PyTorch implementation delivering 40x speedup over the equivalent batched feedforward inference. We publish our training code, benchmarking setup, and model weights.

Subjects:	Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Neural and Evolutionary Computing (cs.NE)
Cite as:	arXiv:2311.10770 [cs.CL]
	(or arXiv:2311.10770v2 [cs.CL] for this version)
	[2311.10770] Exponentially Faster Language Modelling Focus to learn more

Submission history​

From: Peter Belcak [view email]
[v1] Wed, 15 Nov 2023 18:42:50 UTC (35 KB)

https://arxiv.org/pdf/2311.10770.pdf

 

[Piffiztheanswer]

Yall buggin im tryna install some cyberware and pull out the mantis blades on yo bytch!

 

[jj23]

Imagine you design a "kill switch" for me. But I have a 160 IQ and you have a 100 IQ. How long before I figure out how to disable your kill switch or otherwise render it irrelevant?

Now, when we're talking about AI, forget about that 160 IQ and make it 300 IQ, or 400 IQ, some shyt beyond any human comprehension. Because that's probably where we're headed.

Even say that you design a perfect kill switch that absolutely gives humans the upper hand. Well, who says that no human is going to side with AI and fukk up the system so AI can bypass it? There's always going to be that evil genius who thinks he's smarter than everyone else and can control the consequences. If Putin or Trump (or someone like them) started losing a really serious war, you can't imagine they'd allow AI to run wild in a last-ditch attempt to save their own ass? Or even just some random scientist who wanted to watch the world burn?

The secret was to make it a helper. No one fears something that creates resumes or helps write a college paper.
The larger public won't care until it's too late.