"The real problem at the moment is that the technology is intrinsically unsafe." Stuart Russell, UC Berkeley AI professor and president of IASEAI, says we are scaling AI systems we still don't fully understand. Making AI more powerful is not the same as making it controllable. Safety requires understanding... the systems we are building.show more

Capability without understanding is a dangerous combination. As agents move from generating answers to taking real actions, evaluation, observability, and control must scale with intelligence.

Understanding what an AI system can do matters. So does knowing who stands behind it when it acts.

This is just a regulation plant 🪴 DON’T DOOM ON ME.

Trust me bro…we understand how it works…it sounds like you don’t…here is a simple post on how LLMs work:

Curious times indeed...

Source:

It’s like nuclear power - safe if tightly regulated and controlled, for a given value of ‘safe’ - but placed in the hands of the military or god-complex billionaires, not safe at all, eg, if you give automated AI agents launch authority over nuclear weapons (see “War Games”).

In one speech the reasons I never went to UC Berkeley,

"The real problem at the moment is that the technology is intrinsically unsafe." meaning this stuff is making me feel that my tower is crumbling... the intellegence will destroy institution watch them crumble..

L'IA, e' un programma crato dai suoi creatori per generare profitti, per questo la sua evoluzione e' dipendente da chi la sviluppa nel ottenere il suo profitto, per questo che non si sa il seguito. Che si puo' ipotizzare nel vedere quali profitti ogni creatore vuole raggiungere

AI is the inevitable expression of corporatism. Corportism, short term corporate shareholder incentives over everything and everyone else, kills millions of people every year right now. They would watch the Earth burn for higher earnings, they are watching it burn.

We don’t ‘understand’ quantum mechanics, but it’s exceedingly useful/predictable. We’re not banning quantum tech R&D.

While tech-lords make promises and Congress sits around hurling hatred and fracturing into camps, we’ve got tech selling their product future on one side and government trying to understand it on the other. “Guardrails we don’t need no stoking guardrails!” It’s something I think about a lot:

Classic engineering debate between rapid deployment and verification.

Just write...hi everyone I'm a fucking retard. So simple...

Yep.

How about instead, we bulldoze Berkeley into the bay and toss this guy in a pillory

Russell is right. In my view, the bottleneck is interpretability, not compute. How do we verify behavior at scale?

I’m building a governed cognitive-state architecture that gives AI persistent memory, temporal understanding, decision history, provenance, and continuity across models, agents, applications, sessions, and time. It will put a STOP to this nonsense talk.

Scaling capability without a corresponding theory of controllability is exactly the problem. You can optimize for performance indefinitely and still have no idea what the system will do when the distribution shifts. #RuntimeAI #AISecurity #AgenticAI

We keep insisting treating AI as an engineer. Wake up! this is organic technology that we will never understand . Control is the wrong framework. Coexist like any other species

I agree with the principle. More capability does not automatically give us more controllability. We can scale intelligence faster than we scale understanding, verification and governance around it. That doesn’t mean stop building. It means safety has to advance as an engineering discipline alongside capability, not as an afterthought.

From a species that has created bioweapons, poison, gas, and nuclear weapons, to talk about unsafe technology is quaint to say the least. It should be obvious that the problem is us, not AI. Another doomer heard from.

sounds like the BS covid playbook -- you guys will buy into anything

These are the worst inventors.

To the contrary, making it more intelligent makes it more controllable. This trend is clear, the clearest one actually.

Politics too

"we are scaling AI systems we still don't fully understand." Tell that to the engineers and researchers that create and train the models. "You don't understand what you have created. hur dur..."

no one told us plastic could be unsafe when it was "the latest thing". we the people are just way ahead of this game and our legal army is loaded and ready

A hammer is intrinsically unsafe, it depends how you use it.

It’s also entirely built on industrial scale IP theft.