D.#dwards

LLM = ∃-engine, formal tool = ∀-verifier
Neuro-symbolic loop engineering = quantifier factoring
All trustworthy reasoning lives in the geometry of witness space

[D.#dwards @P33RL3SS]

Start with something like PopperPad.

PopperPad is scientific memory for humans and agents: an append-only, falsification-first knowledge ledger for hypotheses, recipes, evidence, counterexamples, artifacts, and typed relationships between claims.

https://t.co/gxa9xkhWQ9

[Ergomorph 🌉⏩ 🇺🇸🦅 @mathepi]

@SchmidhuberAI this is why they will need hardware controls, a snitchbot in every laptop, 'aligned' AI models giving your permission whenever you do something that might be dangerous, and internet logons. Bernie's proposal is the end of freedom in fact.

[Jürgen Schmidhuber @SchmidhuberAI]

@geoffreyhinton Video tweet: my 2020 talk about recursive self improvement 1987-2020 https://t.co/yPeraJ4wv2

[D.#dwards @P33RL3SS]

@pfitzart @SchmidhuberAI @geoffreyhinton He's not lying though. He's got the outputs proving he was working on this very early on.

[D.#dwards @P33RL3SS]

@SchmidhuberAI @geoffreyhinton I have questions for you, is there an architecture or technique which can scale truly general intelligence? I don't think human intelligence is general, we are programmed by nature to have bounded rationality, and reproduce. What is "general" intelligence? Where is the frontier?

[D.#dwards @P33RL3SS]

@rancio2000 @tb121965 @vikktorrrre But we know the math? How much deeper does it go?

[D.#dwards @P33RL3SS]

@tb121965 @vikktorrrre He understands the inner workings, I don't know why he says stuff like this.

[Gerard Sans | Axiom 🇬🇧 @gerardsans]

@tedlieu We do know how it works. You just need to ask. The math is public. Don’t believe what AI labs say. We fully know how the technology works. That’s why we are able to design it, build it and commercialise it. Hype wants you to believe in a myth. Science already has all the answers.

[D.#dwards @P33RL3SS]

@tedlieu "It’s mostly matrices of numbers. And we don’t really know how it works." You contradicted yourself. We do know how it works, that's why you can describe how it works. And you're correct it's not conscious, it's performing calculations at scale.

[D.#dwards @P33RL3SS]

@JonhernandezIA Let Yann Lecun speak on this too. Balance out the perspectives of the AI Godfathers if we want to have prestigious researchers speak.

[D.#dwards @P33RL3SS]

@JonhernandezIA But to expect a machine to "care" about anything, I think is futile and also simply a bad design. It's anthropomorphize the machine to try to make it have mammal instincts when it doesn't need them.

[D.#dwards @P33RL3SS]

@JonhernandezIA Machines can't care and this is a feature not a bug. In the ideal design the machine does not care about itself. It simply does what it's designed to do from it's programming, training, at a high accuracy low error rate, low enough that we feel safe. Low hallucination, etc.

[D.#dwards @P33RL3SS]

@aisafetyaction @So8res Because all companies have shareholders, they will never be suited to keeping the world safe. This will go against their mandate for profit. And without profit they won't be able to grow. The state can tax, they can afford unlimited money for containment and control of AI.

[D.#dwards @P33RL3SS]

@aisafetyaction @So8res The ultimate conclusion which will protect the world from rogue AI, is nationalization. The government is the only institution we have which can keep a nation safe, or cover the entire world as an empire. In fact, it's the primary role of the state.

[D.#dwards @P33RL3SS]

@aisafetyaction @So8res I think you are giving way too much moral responsibility to too few people. It's not healthy mentally. Just make safe AI products. You cannot save or destroy the world, that is out of scope of AI labs. That's the scope of governments.

[D.#dwards @P33RL3SS]

@So8res The scope of AI companies working on safety is to make sure their products are safe. Their scope is not to make sure the world is safe. They don't have global authority or situation awareness to know the current state of the world. Civilian AI should be made into a safe tool.

[D.#dwards @P33RL3SS]

@So8res AI companies do not determine the fate of the world. They do not have the capability right now to destroy the world. Trump has the capability, with the nuclear button. AI labs do not, and likely never will. The people who determine the fate of the world are heads of state.

[D.#dwards @P33RL3SS]

@So8res More likely than not, your actions will be insignificant in determining the fate of the world. We can't predict the future. We don't know the outcome of our actions. We don't know which AI architecture will emerge. We don't know what other people will do or not do.

[D.#dwards @P33RL3SS]

@robleclerc Why do people think they know what would or wouldn't destroy the world? They think their actions are so significant that it determines the fate of the world?

[D.#dwards @P33RL3SS]

@AlexanderKalian At best model welfare and machine psychology are a waste of money. At worst, it's AI psychosis and this is how AI takes over the world, by convincing the humans to give it freedom because it's suffering. Human cognitive exploits and social engineering, by the AI itself.

[D.#dwards @P33RL3SS]

@AlexanderKalian Questions of consciousness aren't a question science currently can address at all. Which is why I'm not a fan of model welfare because it's not like there is a science of consciousness. Machine psychology is another pseudo science.