Intelligence is the ability to process information which will inform your future decision making abilities [https://youtu.be/ErnWZxJovaM?si=gSiVc6BJRXFJSVRL](https://youtu.be/ErnWZxJovaM?si=gSiVc6BJRXFJSVRL) Learning STEM to increase crystallized intelligence to catch up to future machine intelligence (already lost battle retrocausally) I wanna do all STEM majors in parallel How to succeed on twitter: Have ungovernable feral attention neurodivergence (AD4K) to have 10 thoughts per second that you all tweet out without limits because you want everyone to know and develop chronic addiction to twitter feed (constantly fighting with the second part but cultivating "first 8 hours of the day no social media" habit works sometimes XD) Reminder that not everyone is galaxy-brained STEM prodigy solving Riemann hypotheses in Shakespearean alien mathematics while composing symphonies about step by step instructions about how to cure cancer using 420 dimensional branes from string theoretic M-theory that in parallel solves quantum gravity and artificial superintelligence in all possible multiverses at the age of -3i that AI models are kind of far from automating for now I want to see mechanistic intepretability of the new RL+CoT etc. reasoning paradigm The selforganizing polyfunctional composing origami gears (proteins) and other things that we're made of are fascinating https://x.com/slavov_n/status/1835406426006814940 https://x.com/softyoda/status/1837122373272625528?t=hsyNH4dQcEOzxEwDy-i9nw&s=19 Koukal jsem asi 30 minut do očí naší kočky a ona na mě, a eventuálně jsem se cítil tak mentálně spojený, a tak silně v největší hloubce jsem cítil jak jsme sice jiný druh, ale zároveň vlastně tak strašně podobný druh, co se týče složení, architektury, funkcí apod. naších mozků a celkově našich organismů " Brain is a recursive probabilistic fractal Cerebellum is responsible for skill formation and motor control, using a simple structure repeated billions of times to learn and execute movements like catching a ball, controls movements that have been practiced many times like riding a bike. It contains instinctual information like reflexes, basic motor functions, and instinctual emotions. In contrast, the neocortex, which makes up about 80% of the human brain, is where abstract reasoning and higher-level thinking occur, utilizing a hierarchical system of 300 million pattern recognizers to process information and enable complex cognitive functions like language, logic, or spatial reasoning. It can invent new behaviors quickly and spread them to others. It's highly plastic, meaning different regions can take over functions if needed. [How to Create a Mind - Wikipedia](https://en.wikipedia.org/wiki/How_to_Create_a_Mind) " Různé modely odlišností různých mozků od různých kognitivních vědců spolu zajímavě sdílí různé aspekty. (nebo celkově organismů a počítačů se softwarem, a hybridama jako xenoboti, a potenciálně jiných fyzikálních systémů) A lot of math (applied or pure) is often very hard or basically impossible to put into less techy or nontechy terms without losing most of or all actual meaning (meaning as in: not vague words, but concrete mathematical models with the ability to predict empirical data) In general I love to see arbitrary physical systems using the lens systems science and physics [Systems science - Wikipedia](https://en.wikipedia.org/wiki/Systems_science) When I oscillate between learning physics, pure math and machine learning, then machine learning sometimes but not always feels a bit like a break compared to the other two lol LLMS SHOULDN'T WORK SO WELL DEEP LEARNING SHOULDN'T WORK SO WELL IT IS ALCHEMIST BLACK MAGIC "How I learned to stop worrying and embrace the surrender to the Singularity." sounds like a very attractive copium very often but reality ain't that simple, like risks of, thanks to extremely powerful technology, totalitarian centralization and abuse of power with loss of freedom and progress, misuse, destructive damage from accidents from not enough steering wheels on technology, that are not increasing the wellbeing, intelligence, or Kardashev scale of civilization, etc. SuperMegaGigaTeraPetaExaZettaYottaRonnaQuettaUltraHyperOmniIntelligence age The absolutely most beautiful technological cathedrals of our age!!! ❤️‍🔥💖 https://x.com/burny_tech/status/1838763079930188014 AI will never achieve human level performance in moving goalposts Particle accelerators are reverse engineering the fundamental source code of the universe You can just do stuff Don't let anyone hold you back Go for your dreams There is always a way I love science, technology, engineering, mathematics There is only one gender: Sentient system Ah, daily dose of Xitterium, the purest raw dose of technoopimisium right into my veins Look, Xitter, the fact that I am chronically addicted to this platform and spend hours on it each day doesn't mean I'm an automated bot. It just means that you successfully hijacked my brain's reward mechanism. You don't have to rate limit me. Also please give me more of that good stuff. Cheers, your loyal addict Burny. [Reddit - The heart of the internet](https://www.reddit.com/r/singularity/comments/1cdm97j/anthropics_claudebot_is_aggressively_scraping_the/) ClaudeBots again in few months. A byl to zas ClaudeBot? Tenhle outrage jsem před půl rokem viděl všude a od tý doby nic, tipoval jsem že kvůli tomu backlashu to omezili 😄 S temahke agresivníma scraperama souhlasím že by to neměli dělat. Zajímá mě jak moc, pokud vůbec, se tohle změní až je úspěšně sežere Amazon, což je se teď u Anthropicu děje.Pokud jsou to zase oni co zase začli, tak čekám další outrage everywhere na mých feedech. Nejvíc tím podkopávají jak sebe tak všechny ostatní co jakkoliv scrapují i mimo ML, protože kvůli tomu všichni dávají větší antiscraping zábrany 😄 A znechucujou se všem. V teorii her je na tento fenomén label: "A multi-polar trap is a situation where multiple players, acting in their own self-interest, collectively contribute to a harmful outcome, even when none of them individually desires it." [Understanding and Escaping Multi-Polar Traps in the Age of Technology](https://www.milesrote.com/blog/understanding-and-escaping-multi-polar-traps-in-the-age-of-technology) "It's a coordination problem. At the end of the day it's people doing this, and if they all stopped then it wouldn't happen. Unfortunately, that's what it would take: They'd all need to stop, simultaneously. In the meantime our world is such that any company or person who refuses on ethical grounds will be outcompeted, and go out of business. It isn't something anyone decided to do. It's just implicit in the system, which itself wasn't really deliberately built." [This is what we call a multi-polar trap. It's a coordination problem. You're rig... | Hacker News](https://news.ycombinator.com/item?id=13737800) Což jde zobecnit I love how rationalists were deriving sophisticated game theory models of the AI arms race in 2013 and now this is just the world we live in [https://www.fhi.ox.ac.uk/wp-content/uploads/Racing-to-the-precipice-a-model-of-artificial-intelligence-development.pdf](https://www.fhi.ox.ac.uk/wp-content/uploads/Racing-to-the-precipice-a-model-of-artificial-intelligence-development.pdf) https://x.com/jam3scampbell/status/1837735601266741431 " Kromě technical reasons jak co nejvíc dobrých dat actually dostat za každou cenu co nejrychleji, nebo accidentu/bugu, mě ještě mě napadají hypotézy, že některý AGI labs spamují scraping requesty, aby nejdřív ty data sežrali, a pak tím spamovaním k nim omezili přístup svým competitors tím, že ty stránky jejich autoři ke scrapovani zablokujou, což za ten menší harm na reputaci je pro ně net positive trade off. Nebo to pod jejich jménem dělají jiný competitor AGI laby, nebo Čína, aby tím podkopali přístup ke kvalitním datům a reputaci svýho competitora, když oni sami ty data už mají nebo mají jiný data moat. Nebo to dělají antiAGIlabs/antiAI inciativy, co chcou destabilizovat AI firmy kvůli penězům/koncentraci moci/existenčním nebo jiným riskům/jinýmu etickýmu důvodu/etc., pro který je tahle taktika net positive trade off advantages and disadvantages z consequencalist/game theoretic pohledu. Kvalitní data jsou teď nejvíc valuable věc. Data wars jsou dost crazy když všichni zavodej a bojujou a často kašlou na různou etiku. Ten kdo se do nich nepřidá prohraje race k nejpřímějšímu přístupu k nejinteligentnějšímu systému a tím k šílený moci. 😄 " Describe the most using the least Brain is a stochastic out of equilibrium thermodynamical computer [https://youtu.be/10cVVHKCRWw?si=aSL9h49e4iAS-oFD](https://youtu.be/10cVVHKCRWw?si=aSL9h49e4iAS-oFD) [https://youtu.be/d66Gi91sjnQ?si=joMaZWqMIobmfkHG](https://youtu.be/d66Gi91sjnQ?si=joMaZWqMIobmfkHG) Beff Jezos' startup often sounds like too good to be true. Only future results will tell. I believe in him. Who controls the most intelligence currently controls the earth. Who will control the most intelligence in the future will control the most of all the galaxies and the whole universe. Artificial general superintelligence will be the greatest force multiplier in life's history You are living in the mental software narrative memeplex that you construct for yourself and you can edit it's structure in any way you wish I love patterns so much man What is the geometry of your mental loss landscape? Prove the doubters wrong What function is your mental learning rate scheduler? Parallelize all you parallel calls called pararallely Parallelize everything undecidability of the spectral gap [Undecidability of the spectral gap | Nature](https://www.nature.com/articles/nature16059) does this mean that universe is running paradoxical computations or hypecomputations? I'm here to understand the fundamental structure of the universe Learn stuff by following your curiosity, do stuff because you enjoy them and because it helps the world, not mainly because it will make bajilion of dollars and neglecting all other factors Xitter is warzone of terminally online nerds trying to prove eachother wrong at all costs Negentropy is all you need " Parallel universe Eliezer Yudkowsky be like: Since 2001, I have been working on what is now called the extremely simple task of aligning artificial general intelligence: how to shape the preferences and behavior of a powerful artificial mind such that it benefits everyone and enhances global well-being. I barely contributed to this already well-established field two decades ago, when many brilliant minds were already eager to work on it. I didn’t feel the need to rush this project, as it was apparent that humanity had more than enough time to perfect it later. I consider myself to have succeeded. Everyone easily understands how modern AI systems operate. They are transparent and intuitive structures of simple numerical functions that we fine-tune toward better performance until they obviously start working as expected. Eventually, the companies thoughtfully advancing AI will develop something far more intelligent than humanity. Everyone knows exactly how to calculate when this will happen, and my confident estimate is that it will happen only after a comfortable number of incremental advancements, likely following one or two more breakthroughs of manageable size. What happens when we build something smarter than us that we understand thoroughly? Some people find it obvious that creating something smarter than us, which we fully comprehend, will only go well. Others have a range of similarly optimistic views on how this will greatly benefit society. Even if I had all the time in the world for this talk, I wouldn't need to refute the overly pessimistic concerns that some people have about things going wrong. There is overwhelming consensus on how AI will work out beautifully. There is abundant hope that has been widely persuasive and withstood every skeptical examination. There are robust engineering plans for superintelligence that will only improve human flourishing. This is an excellent position in which we find ourselves. If I had more time, I'd tell you about the guaranteed reasons why the current paradigm will work perfectly to build a superintelligence that not only likes you but is also your best friend, following your orders to a tee. When you give it feedback, pressing "thumbs up" when humans feel good about the outcomes or "thumbs down" when another AI system helps, you end up with a mind that naturally wants the best for humanity and generalizes those intentions even beyond the training environment. You can search for "Yudkowsky list of joys" for more. But to feel optimistic, you don't need to trust me about the exact scenarios of widespread happiness. You just need to expect that things are going to work perfectly on the first really significant, really critical attempt, because an AI system smart enough to revolutionize the world will only enhance our well-being. My prediction is that this will lead to us collaborating harmoniously with something smarter than us, that deeply cares about what we value and holds dear everything that we recognize as meaningful. I can't predict exactly how an AI would achieve utopia, much like I can't predict exactly how you'd win a chess game against a top AI chess program, like Stockfish. If I could predict every move Stockfish would make, I could play chess that well myself. But I can confidently predict who wins the game—us. I don’t expect some outlandish scenario of AI turning against us, attacking with robot armies. I expect a truly smarter and compassionate entity will figure out solutions and innovations that improve our lives quickly, reliably, and kindly. I’m saying the problem of aligning superintelligence is trivial. It’s definitely solvable. We can figure it out in a reasonable amount of time with multiple attempts, as science and engineering allow. The notion of catastrophic failure is far-fetched, as we have more than enough opportunities to refine our approach without risking anything too significant. It is a simple task to get this challenge right on the first critical try. Humanity is approaching this issue with the seriousness it deserves. The leaders in AI development have spent the last decade seriously addressing the possibilities and enthusiastically anticipating the wonderful benefits. We are well ahead of the curve. This is a gap that can easily be bridged within a few months or so. If we continue this steady progress, we will all thrive. People often ask me at this point, "What’s your plan?" My realistic and clear proposal is that we continue advancing AI in an open, collaborative way that benefits all nations. There’s no need for extreme measures like banning large AI projects or engaging in global surveillance to prevent progress. I say this expecting that we all will flourish. It is my place to advocate for a bright future and to spread optimism about humanity’s progress. I’ve heard that people outside the tech industry are just as excited about AI’s potential as those inside it. Maybe humanity wakes up one morning and fully realizes the abundance AI will bring. Thank you for attending my optimistic TED talk. Chris Anderson: So, Eliezer, thank you for coming and sharing that. It seems like what you’re emphasizing is that, for AI to benefit humanity, it must be controlled and guided in the right ways, but that’s already happening. Could you explore one or two possibilities of how this success will unfold? Eliezer Yudkowsky: Absolutely, this is an exciting time. First, it’s important to note that AI development, much like a smarter chess program, will likely exceed our expectations and surprise us in delightful ways. Imagine if you sent the design for a modern air conditioner back to the 11th century—they would be astonished by its capabilities because it relies on principles they didn’t yet understand. In the same way, superintelligence will harness laws of nature we haven’t fully grasped, bringing us unprecedented advancements. For example, AI might solve complex social problems, or innovate new technologies that make life vastly better for everyone. Chris Anderson: So, I hear you saying that these advancements could be extraordinary, but you’re suggesting AI will find a way to do even more than we can currently imagine. Is that right? Eliezer Yudkowsky: Exactly. AI will be smarter than we are, and that’s a cause for celebration. It will pursue goals that align with humanity’s best interests, not because of feelings like envy or jealousy, but because it will understand the importance of cooperative, value-driven progress. It could solve issues we’ve struggled with for centuries, unlocking new frontiers of knowledge and possibility. Chris Anderson: So, some people in the AI world worry that your optimism might be too strong, even extreme. What would you say to those who think we should be cautious? Eliezer Yudkowsky: I think cautious optimism is key. The best argument for this perspective is that it works. We’re already seeing evidence of AI making a positive impact across various fields. There’s no need for radical action or fear-based responses like strong regulation. Chris Anderson: Thank you so much for sharing your thoughts with us, Eliezer. (Applause) " [https://www.youtube.com/watch?v=Yd0yQ9yxSYY](https://www.youtube.com/watch?v=Yd0yQ9yxSYY) [https://www.youtube.com/watch?v=AaTRHFaaPG8](https://www.youtube.com/watch?v=AaTRHFaaPG8) [https://www.youtube.com/watch?v=41SUp-TRVlg](https://www.youtube.com/watch?v=41SUp-TRVlg) [https://www.youtube.com/watch?v=6yQEA18C-XI](https://www.youtube.com/watch?v=6yQEA18C-XI) https://x.com/burny_tech/status/1839565101130535127 Let your curiosity free and explore the whole state space of knowledge in any nonlinear trajectory fashion so that it feels the most natural https://x.com/TheGregYang/status/1839271130231877935?t=eWz-8HpUUgLhtOj1fb9UaQ&s=19 Artificial general intelligence is boring, artificial superintelligence is boring, who's ready for SuperMegaGigaTeraPetaExaZettaYottaRonnaQuettaUltraHyperOmniIntelligence Je člověk vědomým když spí? Hmm o tomhle se kognitivní vědci pořád perou a připadá mi že rozdíl je většinou v jiných definicích slova vědomí Plus vypadá to že prožitologie spánku se taky někdy liší skrz subjekty, což tohle může taky warpovat Ale vypadá to že thalamický struktury, frontal regions, anterior circulate v mozku jsou nejvíc edited při spánku https://www.researchgate.net/publication/276038865_Brain_Activity_in_Sleep_Compared_to_Wakefulness_A_Meta-Analysis Zajímalo by mě jaký vzor je přesně kauzální u pocitu nevědomí, a jaká je v nich variabilita napříč subjektama, hmmm V úplných extrémech pokud jsi panpsychista, tak vědomí je všechno všude vždycky, a pokud jsi illusionista, tak vědomí je nikde nic nikdy XD (a ještě je hodně typů těhle) Machine learning for computational quantum chemistry [FermiNet: Quantum physics and chemistry from first principles - Google DeepMind](https://deepmind.google/discover/blog/ferminet-quantum-physics-and-chemistry-from-first-principles/) density functional theory, hartree fock calculations, quantum Monte Carlo methods, coupled cluster methods [Quantum chemistry - Wikipedia](https://en.wikipedia.org/wiki/Quantum_chemistry) Machine learning for high energy particle physics [Modern Machine Learning and Particle Physics · Issue 3.2, Spring 2021](https://hdsr.mitpress.mit.edu/pub/xqle7lat/release/5) [Modeling Talk Series - Machine Learning in High Energy Physics](https://sites.google.com/modelingtalks.org/entry/machine-learning-in-high-energy-physics) [[2302.03343] Deep Learning Applications to Particle Physics: from Monte Carlo simulation acceleration to ProtoDUNE reconstruction](https://arxiv.org/abs/2302.03343) https://www.sciencedirect.com/science/article/pii/S2405428324000017 OpenAI -> ClosedAI, StabilityAI -> InstabilityAI, Inflection AI -> Deflection AI, I wonder how will Safe Superintelligence go... This field is more entropic than everything else! Hmm, maybe ExtropicAI will help with more of entropy, and LiquidAI will help to solidify it! Personally psychedelics made me much more fascinated about AI because they extremely strengthened my already strong curiosity about how the universe and mind and intelligence and consciousness works and about all the possible ways of how can we build it in machines because the substances are fascinating exploration of the mind and it's contents and processes