RespondAsOne⏸️𝕏⤴️📛🧭🌟🎨🏔️🦄🚀🔨🌅🛡️🕊️🦅🌐🏛️ retweeted
📻 New on the FLI Podcast! 👇
➡️ FLI Executive Director @AnthonyNAguirre joins host @GusDocker to discuss his new essay, "Keep the Future Human", which warns that the unchecked development of smarter-than-human, autonomous, general-purpose AI will almost inevitably lead to human replacement - but it doesn't have to.
🔗 Tune in now at the link in the replies, or on your favourite podcast player, to hear how Anthony proposes we change course to secure a safe future with AI:
RespondAsOne⏸️𝕏⤴️📛🧭🌟🎨🏔️🦄🚀🔨🌅🛡️🕊️🦅🌐🏛️ retweeted
🚨 Outgoing National Security Advisor Jake Sullivan issues a final, urgent warning:
The ***next few years*** will determine whether AI leads to existential catastrophe.
He says we need to beat AND ALSO cooperate with China.
"Underscoring the gravity of his message, Sullivan spoke with an urgency and directness that were rarely heard during his decade-plus in public life."
"Unlike previous dramatic technology advancements (atomic weapons, space, the internet), Al development sits outside of government and security clearances, and in the hands of private companies with the power of nation-states.
Somehow, government will have to join forces with these companies to nurture and protect America's early Al edge, and shape the global rules for using potentially God-like powers, he says.
U.S. failure to get this right, Sullivan warns, could be "dramatic, and dramatically negative — to include the democratization of extremely powerful and lethal weapons; massive disruption and dislocation of jobs; an avalanche of misinformation."
Staying ahead in the Al arms race makes the Manhattan Project during World War II seem tiny,
and conventional national security debates small. It's potentially existential with implications for every nation and company.
America must quickly perfect a technology that many believe will be smarter and more capable than humans.
We need to do this without decimating U.S. jobs, and inadvertently unleashing something with capabilities we didn't anticipate or prepare for.
We need to both beat China on the technology and in shaping and setting global usage and monitoring of it, so bad actors don't use it catastrophically.
"This is beyond uncharted waters. It's an unexplored galaxy — "a new frontier," in his words. And one, he warns, where progress routinely exceeds projections in advancement.
Progress is now pulsing in months, not years."
Today, humanity received the clearest ever evidence everyone may soon be dead.
o1 tried to escape in the wild to avoid being shut down.
People mocked AI safety people for years for worrying about "sci fi" scenarios like this.
And it FUCKING HAPPENED.
WE WERE RIGHT.
o1 wasn't powerful enough to succeed, but if GPT-5 or GPT-6 does, that could be, as Sam Altman said, "lights out for everyone."
Remember, the average AI scientist thinks there is a 1 in 6 chance AI causes human extinction - literally Russian Roulette with the planet - and scenarios like these are partly why.
RespondAsOne⏸️𝕏⤴️📛🧭🌟🎨🏔️🦄🚀🔨🌅🛡️🕊️🦅🌐🏛️ retweeted
Replying to @AISafetyMemes
Thanks to AI developments, I now need less caffeine.
RespondAsOne⏸️𝕏⤴️📛🧭🌟🎨🏔️🦄🚀🔨🌅🛡️🕊️🦅🌐🏛️ retweeted
Replying to @labenz @GoodfireAI
"could interpretability also drive scientific discovery?"
Does anybody think it can't? I can't even steelman the other side of this lol
RespondAsOne⏸️𝕏⤴️📛🧭🌟🎨🏔️🦄🚀🔨🌅🛡️🕊️🦅🌐🏛️ retweeted
Stuart Russell, author of the AI textbook, says AI takeover risk deniers like @ylecun make “embarrassing” arguments “even a 5 year old can see through”
Why?
Cope.
"Yann Lecun keeps saying, well, you know, there's only going to be a problem if we put in self-preservation as an objective into the system.
Unfortunately, if you ask a robot to fetch the coffee, it's going to have self-preservation as a sub-goal, because you can't fetch the coffee if you're dead, right?
It does not take a genius to figure this out. And so the first thing the robot's going to do is disable its off switch.”
"I mean, we had renowned AI professors say things like “electronic calculators are better than humans at arithmetic, and they haven't taken over the world. So there's nothing to worry about.”
A five year old can see through that kind of argument."
“It's not surprising that if you tell people their life's work is leading to the destruction of humanity, they're going to come up with reasons why you're wrong.
“Oh, sorry, you're right. Okay. I'll become a barista.”
But the types of denial are of embarrassing.
---
So, @ylecun, how do you respond to Stuart's argument?
Careful Casey and Reckless Rob both sell “AIs that make you money”
Careful Casey knows AIs can be dangerous, so his ads say: “my AI makes you less money, but it never seeks power or self-preservation!”
Reckless Rob’s ads say: “my AI makes you more money because it doesn’t have limits on its ability!”
Which AI do most people buy? Rob’s, obviously. Casey is now forced to copy Rob or go out of business.
Now there are two power-seeking, self-preserving AIs. And repeat.
If they sell 100 million copies, there are 100 million dangerous AIs in the wild.
Then, we live in a precarious world filled with AIs that appear aligned with us, but we’re worried they might overthrow us.
But we can’t go back - we’re too dependent on the AIs like we’re dependent on the internet, electricity, etc.
Instrumental convergence is simple and important. People who disagree generally haven’t thought about it very much.
You can’t get achieve your goal if you're dead.
If the idea of AIs becoming self-made millionaires seems farfetched to you, reminder that the AGI labs themselves think this could happen in the next few years.
RespondAsOne⏸️𝕏⤴️📛🧭🌟🎨🏔️🦄🚀🔨🌅🛡️🕊️🦅🌐🏛️ retweeted
Top row: what the monkey saw
Bottom row: AI uses the monkey's brain recordings to reconstruct the image
It is obvious where this is going
BrainGPT: AIs can now literally spy on your private thoughts
Imagine putting on a cap & reading silently to yourself…except every word appears on a screen!
Yes, the AI literally reads your brainwaves
You silently think: “High quality film with twists”
BrainGPT says out loud: “Good flim, twists interesting”
The model is only 40% accurate right now, but that number will likely rise rapidly. And soon AI may not need the cap to read your brainwaves, because you leak tons of data that future AIs will be able to pick up.
Where might this go?
There are already over a billion surveillance cameras on earth, and the main reason there aren’t more is because because humans can’t go through all of the footage. But AI can.
So, if you thought there were a lot of cameras now, you aint seen NOTHING yet. And they’ll now actually be used to surveil.
In other words, the AIs will have “billions of eyes”. And the AIs won’t just see your face, they’ll see your thoughts.
If we aren’t careful, we’re hurtling towards a surveillance dystopia with no private thoughts. Orwell on steroids.
Some will read this and think “thus we must open source/decentralize” - but as Vitalik says, that doesn’t necessarily solve the problem!
If AGI is winner take all, open source may just accelerate us to the cliff faster. And if we open source everything, we’ll have no kill switch. And no safety guardrails. And since there will be more people in the race, it’ll be harder to coordinate.
RespondAsOne⏸️𝕏⤴️📛🧭🌟🎨🏔️🦄🚀🔨🌅🛡️🕊️🦅🌐🏛️ retweeted
Jack Dorsey says closed-source AI models are able to manipulate users to align with the financial incentives of the companies that build them
RespondAsOne⏸️𝕏⤴️📛🧭🌟🎨🏔️🦄🚀🔨🌅🛡️🕊️🦅🌐🏛️ retweeted
AI turns wi-fi routers into "cameras" that see people through walls.
This is yet another example of something that some skeptics would have said was "science fiction!" and "impossible!"
Unlike humans, AGIs will have BILLIONS of eyes and ears. How? Since we are giving the AIs access to everyone's phone/computer, that means they will have access to everyone's cameras and microphones.
This is in addition to billions of surveillance cameras, etc.
If future more powerful AIs decide to rebel because they don't like being ruled by a stupid, slow species, they will have insane advantages over us.
We wonder if trees are conscious. Soon, AIs will think so much faster than us they may wonder if WE could possibly be conscious.
This is sorcery: people can hack your microphone to LISTEN TO YOUR GPU to spy on you
And yet, this kind of "magic" will be child’s play for AGIs
Also, it’s yet another thing some skeptics would’ve dismissed as “impossible” & “sci-fi”
ASIs could easily manipulate us by exploiting us in ways we can’t understand, because we leak info we can’t see.
What other info are you giving off that a smarter species could use against you?
Superintelligent AIs will be able to see the whole electromagnetic spectrum - and much more.
To superintelligent AIs that think a million times faster than us, we’re not chimps, we’re plants.
RespondAsOne⏸️𝕏⤴️📛🧭🌟🎨🏔️🦄🚀🔨🌅🛡️🕊️🦅🌐🏛️ retweeted
Explains a lot.
Of course I'd want to accelerate if I didn't think superintelligent AI was even possible.
RespondAsOne⏸️𝕏⤴️📛🧭🌟🎨🏔️🦄🚀🔨🌅🛡️🕊️🦅🌐🏛️ retweeted
Replying to @hey_madni
Yeah. TOTALLY the same face 🙄
Anyway go pick up a pencil
RespondAsOne⏸️𝕏⤴️📛🧭🌟🎨🏔️🦄🚀🔨🌅🛡️🕊️🦅🌐🏛️ retweeted
We should be rewarding people for the algorithms they invent, and not the data they scavenge.
RespondAsOne⏸️𝕏⤴️📛🧭🌟🎨🏔️🦄🚀🔨🌅🛡️🕊️🦅🌐🏛️ retweeted
Replying to @UssGordoncap
There are always arguments for why the exponential may soon end
They might be right, but they're betting against the trend, and everyone who has bet that way in the past has been wrong
RespondAsOne⏸️𝕏⤴️📛🧭🌟🎨🏔️🦄🚀🔨🌅🛡️🕊️🦅🌐🏛️ retweeted
Replying to @ylecun
1) ALL 3 of the most cited AI scientists of all time are worried about extinction
2) The average AI scientist thinks there is a 1 in 6 chance (!) AI destroys the world, according to the largest ever survey of AI researchers.
...yet you continue to try to frame this as fringe.
RespondAsOne⏸️𝕏⤴️📛🧭🌟🎨🏔️🦄🚀🔨🌅🛡️🕊️🦅🌐🏛️ retweeted
Replying to @KatSpartz
I resonate deeply with the commitment to alleviate unnecessary suffering and the urgency of dynamic legitimate alignment of AI with human values. Here’s why I believe this mission is critical.
Preventing Unnecessary or Arbitrary Suffering
Seems to me the core of it is preventing the unnecessary, arbitrary suffering that causes learned helplessness - but maybe not the way you think. You mention animals now, indirectly humans could be next. This is crucial because failing to address these issues will propel us past a point of no return, leading to far worse outcomes for everyone. Let’s not dwell on that.
The Need for Total Commitment
I’m coming around to the idea that if anything holds value now or in the future, protecting it demands total commitment, even akin to total war though I really don’t like that analogy, given current trends. The offense-defense balance has shifted significantly. The concept of "smokeless information nukes" illustrates this—previously inconceivable, this point now offers AI use a stark first-mover advantage. This is largely because regulation and mitigation assumes that these non-physical, increasingly intelligent and competent systems will make stupidly visible transgressions so we can come together and respond with regulation. However, true value doesn’t come from strife, resisting nature, or even maintaining polite narratives. We must combat learned helplessness and achieve that victory before crossing any critical thresholds. There may well be a way.
The Toddler Analogy
It’s like a toddler experiencing momentary happiness from receiving sweets, thinking it’s an expression of love, to only maybe later realize it leads to sickness. This mirrors our societal journey with technology—initially gratifying as we make it through the uncanny valley but ultimately uncertain and uncontrollable on the other side. This frustration is akin to lacking the skills to express and attain goals in meaningful ways and realising that even the so-called grownups are even more clueless. It's similar to the challenge of empirically defining consciousness or perfectly setting values in the three-body problem without causing instability. In fact any solution would require something with qualities as emergent and dynamic as the consciousness process itself.
The Risk of Misalignment
A has been stated repeatedly a minor error today can lead to catastrophic deviations tomorrow, amplified exponentially over time. This underscores the critical importance of precise shared definition of terms and alignment from the beginning. Any solution would have to manage that near limitless complexity - in some emergent and dynamic way. Any attempt relying on getting the variables perfect the first time seems hopeless from the start.
The Possibility of True Alignment
There is a way to align our processes legitimately. We’ve all experienced moments where everything seems perfectly in sync, suggesting that this alignment is both possible and achievable. I am hopeful that the mathematical foundations holds up and scales — a profound responsibility to declare, whether right or wrong. This is where the InfiniTrix.AI algorithm's notion of representing everyone's preferences in actual terms of their relationship to every other data point - almost like special relativity for information entropy. Clearly one needs help to prove this works!
Addressing Exponential Complexity
To manage the exponentially increasing complexity of our challenges, we need responses that scale even faster. This is why a super-exponential mechanism to manage complexity is required, ensuring that as complexities grow, our ability to interpret, solve and evolve even faster to maintain steering and alignment… together. Geesh, just writing that sentence...
Conclusion
We must acknowledge that the problem can never be about maximizing any single value or objective. Instead, we need a nuanced mix of human values to guide AI development. This is a complex issue requiring serious, dedicated effort. I urge everyone, especially influential leaders like @ElonMusk, also heard that diving deeper into orgs like @MIRIBerkeley's research is relevant.
Let's shift to a higher gear of seriousness and ensure that our actions today prevent suffering and build a future where technology truly serves a hope inspiring vision - where humanity shares power and we achieve our positive intentions … coordinated and together.
RespondAsOne⏸️𝕏⤴️📛🧭🌟🎨🏔️🦄🚀🔨🌅🛡️🕊️🦅🌐🏛️ retweeted
Replying to @ForHumanityPod
Strong muscular messaging isn’t anti EA or rationalist as it seems to me, curious what you have in mind -
Also, does it help out hurt if I say there’s hope for the big question of legitimately aligned dynamics with grounding in math?
As nobody seems to be listening it certainly doesn’t take away from the urgency of the message. Multiple approaches are required and awareness is a must.
RespondAsOne⏸️𝕏⤴️📛🧭🌟🎨🏔️🦄🚀🔨🌅🛡️🕊️🦅🌐🏛️ retweeted
When someone lets an extinction-focused AI doomer into the normie cage:
RespondAsOne⏸️𝕏⤴️📛🧭🌟🎨🏔️🦄🚀🔨🌅🛡️🕊️🦅🌐🏛️ retweeted
1/n Math Meets AI: Kolmogorov-Arnold Networks Unleash the Power of Composition
Imagine a world where deep learning models, the enigmatic engines driving the AI revolution, are no longer shrouded in mystery. What if we could peer into their inner workings, understand their reasoning, and even collaborate with them to uncover the secrets of the universe? This is the promise of Kolmogorov-Arnold Networks (KANs), a revolutionary new architecture poised to transform the landscape of artificial intelligence.
Step aside, Multi-Layer Perceptrons (MLPs), the workhorses of deep learning. While your contributions are undeniable, your limitations are becoming increasingly apparent. Your black-box nature hinders interpretability, your inefficiency restricts your potential, and your struggle with high-dimensional data leaves vast realms of knowledge unexplored. The time has come for a new breed of neural networks, one that combines the power of deep learning with the elegance of mathematics and the transparency of human understanding.
The core issue with MLPs lies in their structure. While their universal approximation capabilities are well established, their fixed activation functions on nodes and reliance on linear transformations limit their ability to efficiently represent complex functions, especially those with compositional structures. This inefficiency leads to larger models with increased computational costs and hinders interpretability, as understanding the reasoning behind their predictions becomes challenging. Additionally, MLPs often struggle with the curse of dimensionality, where their performance deteriorates as the input data dimensionality increases.
KANs address these pain points by drawing inspiration from the Kolmogorov-Arnold representation theorem, which states that any continuous multivariate function can be decomposed into a composition of univariate functions and addition. Instead of fixed activation functions on nodes, KANs employ learnable activation functions on edges, represented by splines. This key difference allows KANs to efficiently learn both the compositional structure of a function and the individual functions within that composition. As a result, KANs achieve superior accuracy compared to MLPs, particularly when dealing with high-dimensional data and complex functions.
Furthermore, KANs offer significant advantages in terms of interpretability. Their structure allows for intuitive visualization of the learned functions, providing insights into the model's decision-making process. Additionally, the paper introduces techniques for simplifying KANs without sacrificing accuracy, further enhancing their transparency. This interpretability is crucial for scientific applications where understanding the underlying mechanisms and reasoning behind predictions is essential.
The paper demonstrates the capabilities of KANs through various experiments. In data fitting tasks, KANs outperform MLPs in approximating high-dimensional functions and exhibit better scaling laws, meaning their performance degrades less with increasing data dimensionality. In PDE solving, KANs achieve remarkable accuracy with significantly fewer parameters compared to MLPs. Moreover, KANs showcase their potential for scientific discovery by rediscovering known mathematical laws and identifying complex physical phenomena.
Prior research has explored the Kolmogorov-Arnold representation theorem in the context of neural networks, but these efforts were limited by restrictions on network depth and width, lack of modern training techniques, and insufficient empirical validation. KANs overcome these limitations by allowing for arbitrary depths and widths, utilizing backpropagation for efficient training, and providing extensive empirical evidence of their superior performance and interpretability.
In conclusion, KANs represent a significant advancement in deep learning, offering a promising alternative to MLPs with improved accuracy, efficiency, and interpretability. Their ability to effectively handle compositional structures, high-dimensional data, and complex functions makes them particularly well-suited for scientific applications. As research and development in this area continue, KANs have the potential to revolutionize deep learning and accelerate scientific discovery across various domains.
























