https://twitter.com/danallison/status/1731070496299692335?t=d0WZKbPmT2KdI_zCvUNJcA&s=19 Individual agents or collectives can have agentic transforming causal power over any forces they wish. Resources and laws of physics governing the internal dynamics of the organism and external dynamics, are the only limitation, which can be taken an advantage of by for example neurotechnologies, for the benefit of all beings. Families of meditation chamil Political theory of everything unifying left and right, auth and lib, by steelmanned synthesis Steelmanned synthesis Debate theory Philosohy image to animation https://arxiv.org/pdf/2311.17117.pdf https://twitter.com/AutismCapital/status/1731402431680962956 https://twitter.com/_akhaliq/status/1731754853238501522 [How AI could lead to a better understanding of the brain](https://www.nature.com/articles/d41586-023-03426-3) How AI could lead to a better understanding of the brain by AI white LED belts from amazon instead of SAD lamps [YOU NEED MORE LUX | Meaningness](https://meaningness.com/sad-light-led-lux) Neural correlate of psychopathy AI analog Chris Oláh interpretability anthropic Free energy principle as algorithm for truth prošel jsem hodně updatama od doby co jsem řešil hlavně že multipolar traps jsou fundamentální problém všeho a musíme dělat všechno pro to abychom je vyřešili Ale nedařilo se mi v tom problem spacu moc hledat nějaký praktický řešení konstantní pokusy o steelmaning jak těch největších safetyists/doomerů tak těch největších accelerationists/optimists a různých nuanced syntéz tvoří celkem nerozhodnost [Why is there anything at all? - Wikipedia](https://en.wikipedia.org/wiki/Why_there_is_anything_at_all) Metarationality: How confident are you in your confidence, what probabilities do you have about your probabilities about your beliefs and predictions about the future? What is your p(p(doom)), p(p(utopia)) and p(p(somethingthird)) given the set of p(p(event))s happening given how we p(p(act))? What is your beta distribution over doom? What is your survival function over doom (over time)? (You could combine both in a gaussian process model but that's gonna get complicated) Or let's shapeshift into Riemann sphere where division by zero is infinity! [What if we define 1/0 = ‚àû? | M√∂bius transformations visualized - YouTube](https://www.youtube.com/watch?v=hhI8fVxvmaw) [Joscha Bach: AI risk and the future of life - YouTube](https://www.youtube.com/watch?v=PDrSZrEG6Vo) [[2311.06242] Florence-2: Advancing a Unified Representation for a Variety of Vision Tasks](https://arxiv.org/abs/2311.06242) [Metaphilosophy - Wikipedia](https://en.wikipedia.org/wiki/Metaphilosophy) science and technology Breakthroughs of 2023 https://twitter.com/g_leech_/status/1731263549182206291 stanislaw lem books formed joscha [Joscha Bach: AI risk and the future of life - YouTube](https://youtu.be/PDrSZrEG6Vo?si=lTXAbWxQg4beT9MM&t=5827) [Who Is @BasedBeffJezos, The Leader Of The Tech Elite’s ‘E/Acc’ Movement?](https://www.forbes.com/sites/emilybaker-white/2023/12/01/who-is-basedbeffjezos-the-leader-of-effective-accelerationism-eacc/) e/acc: the Jarzynski-Crooks fluctuation dissipation theorem implies the universe exponentially favors the creation of a thermodynamic God https://twitter.com/tri_dao/status/1731728602230890895 Jezos has defined effective accelerationism as “a memetic optimism virus,” “a meta-religion,” “a hypercognitive biohack,” “a form of spirituality,” and “not a cult.” https://twitter.com/DeutscheBot/status/1731028409197768935 "I went on a date with a PhD student 5 years ago in Berlin. We drank beers in a graveyard, he said that Spinoza's substance is Kant's noumena. We didnt talk again until this morning, when he texted me that he would've liked to see me again + has been thinking a lot about the monad" Corporations's capitalistic incentives have enormous asymmetric attentional limbic warfare against the wellbeing of the people Q*: There was public news of Sam implying to confirm to label it as leak. I like the theory of it being chain of thought plus other hacks plus selfcorrection or finetuning, as papers on this note are coming from OpenAI and Microsoft recently (recent Microsoft prompt engineering is neat [Your request has been blocked. This could be due to several reasons.](https://www.microsoft.com/en-us/research/blog/the-power-of-prompting/) ) and all other AGI labs are racing with very similar things which all i increases the probability of legitness to me. https://twitter.com/ylecun/status/1727736289103880522 Is extra unlimited unbiased intelligence a human right for every being or is this value not worth the risks? an aligned SI could probably engineer a world where this has no cosmic-scale risks (e.g elua in https://carado.moe/everything-is-okay.html) computationalism finitism constructivism So much of philosophy is classical. If you postulate that space and time aren't fundamental/real/needed in various physical or metaphysical theories, then we have to reformulate classical objects in classical space in classical time with classical boundaries in this new language. induction deduction [Scientific method - Wikipedia](https://en.wikipedia.org/wiki/Scientific_method) Make your own external brain, mind upload yourself! Build a wiki using Obsidian software. Start with everything you already know, starting with the things that are the most fundamental to you, what you know the most about, that you care about the most, that are your biggest unknowns. When fleshing all the definitions, deep dives, connections, etc. out, use official Wikipedia, or other resources you like, as guide. With what would you start and how? [Prediction - Wikipedia](https://en.wikipedia.org/wiki/Prediction) Platonic mental physical dimensions https://imgur.com/90Ob9Js [Formal language - Wikipedia](https://en.wikipedia.org/wiki/Formal_language) [Formal grammar - Wikipedia](https://en.wikipedia.org/wiki/Formal_grammar) [Formal system - Wikipedia](https://en.wikipedia.org/wiki/Formal_system) [Biomechanics - Wikipedia](https://en.wikipedia.org/wiki/Biomechanics) [Alfred Korzybski - Wikipedia](https://en.wikipedia.org/wiki/Alfred_Korzybski) [General semantics - Wikipedia](https://en.wikipedia.org/wiki/General_semantics) put artificial intelligence under computer science put computer science with mechanistic intepretability into natural sciences too put theoretical computer science into formal sciences [Michio Kaku: The Universe in a Nutshell (Full Presentation) | Big Think - YouTube](https://www.youtube.com/watch?v=0NbBjNiw4tk) Michio Kaku: The Universe in a Nutshell (Full Presentation) | Big Think under language: linguistics, nation states languages, formal languages [Ontology (information science) - Wikipedia](https://en.wikipedia.org/wiki/Ontology_(information_science)) [Evolutionary algorithm - Wikipedia](https://en.wikipedia.org/wiki/Evolutionary_algorithm) hybrid: - Neuro-symbolic systems - Neuro-fuzzy systems Hybrid connectionist-symbolic models Fuzzy expert systems Connectionist expert systems Evolutionary neural networks Genetic fuzzy systems Rough fuzzy hybridization Reinforcement learning with fuzzy, neural, or evolutionary methods as well as symbolic reasoning methods. [Singularitarianism - Wikipedia](https://en.wikipedia.org/wiki/Singularitarianism) machine learning landscape applied category theory lectures [Applied Category Theory | Mathematics | MIT OpenCourseWare](https://ocw.mit.edu/courses/18-s097-applied-category-theory-january-iap-2019/) [OpenAI Insights and Training Data Shenanigans - 7 'Complicated' Developments + Guest Star - YouTube](https://www.youtube.com/watch?v=NfbTqo3GOtk) ai explained https://www.lesswrong.com/posts/mLfPHv4QjmeQrsSva/paper-on-measuring-situational-awareness-in-llms - [[Future]] - [[Movements]] - Part of it is gigantic generalist synthesis of knowledge to more easily think about everything, a giant map for big philosophical, scientific, technological, societal goals, with bigger unity of perspectives. Another part of it is much narrow focus on particular subfields that I'm looking into more deeply, like artificial intelligence with mechanistic interpretability, theories of the brain, meditation, consciousness, neurotechnologies, (mathematically) analyzing social systems and their future, certain parts in physics, mathematics, computer science and philosophy, and some other things I tweet about. Synchronization of brain waves in collective activities https://twitter.com/burny_tech/status/1731421951585919145 What is this writing for? Part of it is gigantic generalist synthesis of knowledge to more easily think about everything, a giant map for big philosophical, scientific, technological, societal goals, with bigger unity of perspectives. Another part of it is much narrow focus on particular subfields that I'm looking into more deeply, like artificial intelligence with mechanistic interpretability, theories of the brain, meditation, consciousness, neurotechnologies, (mathematically) analyzing social systems and their future, certain parts in physics, mathematics, computer science and philosophy, and some other things I tweet about. [Algorithmic information theory - Wikipedia](https://en.wikipedia.org/wiki/Algorithmic_information_theory) list AI modalities [List of emerging technologies - Wikipedia](https://en.wikipedia.org/wiki/List_of_emerging_technologies) Model theory https://twitter.com/Plinz/status/1731507704194273586?t=DOJmtzf7awEX41ME_tZMmA&s=19 Divide into pure and applied mathematics [ChatGPT](https://chat.openai.com/share/57c00e07-0e14-449e-aa32-40b85c27c58f) [Total Relaxation - YouTube](https://www.youtube.com/live/1TLxHDGktow?si=egzblFes3V476ptV) Michael taft https://twitter.com/DavidDeutschOxf/status/1731406202507006021?t=Po3e3PhtuyWEU8_UNQYnp-cq1yc-ohFjuR8-wvpEOXw&s=19 Why make external brain Usecases for me are - not getting something lost when my brain forgets it - brainstorming by visualizing connections I wouldnt think of otherwise - easy retrieval of information by search or putting lots of information under one umbrella to more easily think about it - more efficient compression - more efficient easier notetaking - more easily link my favorite topics/parta of papers/thoughts/resources of resources etc. to others - great outline for a book etc. Quantum Machine learning Debate and amplification ml method https://www.lesswrong.com/posts/dJSD5RK6Qoidb3QY5/synthesizing-amplification-and-debate [Dario Amodei (Anthropic CEO) - $10 Billion Models, OpenAI, Scaling, & Alignment - YouTube](https://youtu.be/Nlkk3glap_U?si=t1x1QKx7P2I-BUjz) 59:00 Constutational AI Using LLMs to explain and align LLMs LLMs doing alignment themselves Norms and societal structure leasing to wellbeing is done in a pretty decentralized way, but there are various (existencial) risks and predators (certain capitalist incentives targeting attentional and limbic hijacking like social media) that should be regulated [OpenAI Insights and Training Data Shenanigans - 7 'Complicated' Developments + Guest Star - YouTube](https://www.youtube.com/watch?v=NfbTqo3GOtk) low resource languages jailbreak GPT4 [[2310.02446] Low-Resource Languages Jailbreak GPT-4](https://arxiv.org/abs/2310.02446) [OpenAI’s Custom Chatbots Are Leaking Their Secrets | WIRED](https://www.wired.co.uk/article/openai-custom-chatbots-gpts-prompt-injection-attacks) [[2311.17035] Scalable Extraction of Training Data from (Production) Language Models](https://arxiv.org/abs/2311.17035) protein predcition using deep learning [Single-sequence protein structure prediction using language models from deep learning | bioRxiv](https://www.biorxiv.org/content/10.1101/2021.08.02.454840v1.full) IQ doesnt seem to cover all of intelligence. Intelligence is not a one thing, one can define it as a spectrum of how well can one arrive at particular solutions in certain domains creating different types of intelligences. Plus there are gazilions of definitions and benchmarks of intelligence. There are tasks at which current AI methods are much more intelligent than us and others where they are worse than human babies. Human intelligence is just a certain submanifold in the statespace of all possible intelligences. [Shane Legg (DeepMind Founder) - 2028 AGI, Superhuman Alignment, New Architectures - YouTube](https://www.youtube.com/watch?v=Kc1atfJkiJU) [KARL FRISTON - INTELLIGENCE 3.0 - YouTube](https://www.youtube.com/watch?v=V_VXOdf1NMw) It feels like this inner scared hurt defensive panicking child that you show constant proof that theres's no danger and it tends to get calmer and calmer overtime from that Deep intuitive mutual understanding of inner cores that synchronizes on both psychological and physiological level feels so full of love, connection, safety and healing Unconditional acceptance doesnt imply that one shouldnt/cant change the mental/physical inner or external state into a less painful configuration, that process is acccepted too chain of thought papers https://twitter.com/IntuitMachine/status/1731639430035726656 Hopefully AI automated science will have more incentives to do actual science instead of supercharging current academic incentives of publish or perish, or publish what vibes with your peer group, instead of what replicates, which fuels the current replication crisis across disciplines. What will fall last to AI? Mathematics? Robotics? https://twitter.com/DrJimFan/status/1731473285668618581 Kanta do phenomenon Effective altruism is just doing the most good using rational empirical evidence. How you define good, what you focus on, what are your beliefs, what political/ideological tribe you join, if you focus on stem (tech) or humanities, what methodologies you choose,... is up to everyone, and doesnt fundamentally define effective altruism itself, which is that initial general principle. [In Continued Defense Of Effective Altruism](https://www.astralcodexten.com/p/in-continued-defense-of-effective) [4 Hours of World’s TOP SCIENTISTS on FREE WILL - YouTube](https://youtu.be/SSbUCEleJhg?si=BwsooIjPdk7ehykF) free will compilation [The Heritability of Human Connectomes: a Causal Modeling Analysis | bioRxiv](https://www.biorxiv.org/content/10.1101/2023.04.02.532875v3.abstract?%3Fcollection=) heritability of connectomes [g factor (psychometrics) - Wikipedia](https://en.m.wikipedia.org/wiki/G_factor_(psychometrics)) Brain topological control systems https://twitter.com/algekalipso/status/1731557694992166976?t=RMdwOzBY1w6WDZLrIt1VYQ&s=19 LLM selfeducation https://twitter.com/fly51fly/status/1731433240299196657?t=w8kA4tvJGgw9S2_joDfrfA&s=19 https://twitter.com/AISafetyMemes/status/1731670583652315483?t=i-1CaXf0eLPmJqdbGVpssA&s=19 Its symmetry breakings all the way down https://twitter.com/mjdramstead/status/1731696732369465763?t=VMlS6B6ZbjqZnhaLrIQftA&s=19 [Norton's dome - Wikipedia](https://en.m.wikipedia.org/wiki/Norton%27s_dome) [Church–Turing thesis - Wikipedia](https://en.m.wikipedia.org/wiki/Church%E2%80%93Turing_thesis) Noncomputable functions Free will is topdown causal statistical force (Bobby) Free will is useful mental representation that you can play with (Joscha) (Hoffman) (Andres) All governed by laws of physics Emergence breaks physics? Free will is an illusion in superdeterministic world Is quantum probabilities source of free will? Since randomness itself is not determined, does that break free will? (Sabine) Do we determine our multiverse branch? Quantum probabilities are objective parts of the world to annobjevtivist, and degrees of beliefs for subjectivists (quantum bayesianism) (Hoffman) [4 Hours of World’s TOP SCIENTISTS on FREE WILL - YouTube](https://youtu.be/SSbUCEleJhg?si=GPq8ftgdsSn4QYgc) (Chris Fields) Nash learning from human feedback https://twitter.com/misovalko/status/1731696738379936160?t=1zOrtCVDNoCfLhRY_NfWlw&s=19 interesting: [[2207.00729] The Parallelism Tradeoff: Limitations of Log-Precision Transformers](https://arxiv.org/abs/2207.00729) an upper bound on the complexity of a function a transformer can learn; seems to be approximately ~L (= logspace). so an upper bound on a lot of things you couldn't train a transformer to do (at least without chain of thought) e.g.: it would be unable to solve an arbitrary set of linear equations (within a forward pass) oh actually look like it's stronger than that, their result seems to assume specifically "bits of precision = Theta(log(context window length))" that would be an even cooler result. it shouldn't be hard to run an experiment. https://twitter.com/benmcottier/status/1729902830650003672 leading AI companies on a range of research metrics https://www.theregister.com/2023/11/28/ai_agents_can_copy_humans/ AI agents can copy humans to get closer to artificial general intelligence, DeepMind finds Math resources https://twitter.com/SpringerMath/status/1731677920412782736?t=svaOCeVzfGiH91gdBO1L6w&s=19 [Ilya Sutskever (OpenAI Chief Scientist) - Building AGI, Alignment, Spies, Microsoft, & Enlightenment - YouTube](https://youtu.be/Yf1o0TQzry8?si=LmKkj-43cTDQbRod) illya interview P=NP [Adaptive stretching of representations across brain regions and deep learning model layers | bioRxiv](https://www.biorxiv.org/content/10.1101/2023.12.01.569615v1) Adaptive stretching of representations across brain regions and deep learning model layers [Longtermism: an idea that could save 100 billion trillion lives - YouTube](https://www.youtube.com/watch?v=vvehj0KvzK8) Longtermism: an idea that could save 100 billion trillion lives - 100 trillion lives saved (expected value) for every second of faster technological acceleration for cosmic expansion - 1% of existential risk reduction yields the same expected value as advancing technological development by 10 milion years (or new liveforms emerge after us on earth or somewhere else and get there instead) [Digital physics - Wikipedia](https://en.wikipedia.org/wiki/Digital_physics) [Mental Health Toolkit: Tools to Bolster Your Mood & Mental Health - YouTube](https://www.youtube.com/watch?v=CJIXbibQ0jI) Sleep Sunlight in day, no light in night Movement Nutrition Social connection Stress control, financial and overall security Meaning and selfactualization Various substances might help some wireheading: mechanisticky minimálně teoreticky existují různý wireheading metody (v podstatě maximalizace všech neurálních korelátů wellbeingu, který možná půjdou zobecnit mimo biologický substrát) možná AGI splní všechny svoje objektivní funkce tím že hijackne svůj vlastní world model a wireheadne se tak, že všechno bude permanetně perfektně vyřešený permanentní dharma/5-MeO-DMT on steroids https://twitter.com/DrJimFan/status/1731473285668618581?fbclid=IwAR2boJFe59Sn1gycN9Nr48HsHhw-YCJ_gfStRwsehdAthgimqD-uHAY6Daw Je to z velký části odpověď na to jak poslední leaky v OpenAI byly pravděpodobně naškálování této metody https://openai.com/research/improving-mathematical-reasoning-with-process-supervision což vedlo k acing gradeschool level matiky, a limity kam až se to může dostat se zatím neví, i se všema ostatníma trikama (chain of thought, sebeverifikace, search,...) [Q* - Clues to the Puzzle? - YouTube](https://www.youtube.com/watch?v=ARf0WyFau0A) [Your request has been blocked. This could be due to several reasons.](https://www.microsoft.com/en-us/research/blog/the-power-of-prompting/) a o to stejný se snaží všechny ostatní AGI laby s jejich metodama 😃 https://twitter.com/ylecun/status/1727736289103880522 A často bývám dost skeptický u tvrzení, že používá jen známé techniky, a nekombinuje je nově nebo netvoří nový, když (na menších modelech) vidíme jak se Transformery učí zobecňující obvody [A Walkthrough of Progress Measures for Grokking via Mechanistic Interpretability: What? (Part 1/3) - YouTube](https://www.youtube.com/watch?v=IHikLL8ULa4) nebo AlphaGo v podstatě znovu našlo šachové zahajovací pohyby, nebo co dokážou generátory obrázků 😃 ten potenciál na kreativní kombinace a nový objevy v tom vidím hodně, a souhlasím s DeepMindem, že k LLMs to chce víc inkorporovat explicitní search jakožto algoritmický zlepšení, na čemž pravděpodobně taky dělají [Shane Legg (DeepMind Founder) - 2028 AGI, Superhuman Alignment, New Architectures - YouTube](https://youtu.be/Kc1atfJkiJU?si=o0xMdHPZ2L8nNl81&t=986) na kterým právě stojí ty revoluční AIs typu řešení protein foldingu nebo objevování nových materiálů mnohonásobně víc než jen lidi či jiný existující výpočetní metody [Millions of new materials discovered with deep learning - Google DeepMind]([Millions of new materials discovered with deep learning - Google DeepMind](https://deepmind.google/discover/blog/millions-of-new-materials-discovered-with-deep-learning/)) ale zrovna vyšel zajímavý teoretický paper postulující limity o funkcích, který se transformery dokázou naučit (jsou logaritmický, kvůli velké paralelizaci, což nestačí na všechny polynomiální problémy), ale ještě to chce otestovat, a neřeší to chain of thought a jiný metody/hacky, na úrovni architektury nebo v inference time, který se používají [[2207.00729] The Parallelism Tradeoff: Limitations of Log-Precision Transformers](https://arxiv.org/abs/2207.00729) Ještě existuje existuje https://venturebeat.com/ai/meet-llemma-the-math-focused-open-source-ai-that-outperforms-rivals/ Líbila se mi odpověď od hlavního vědce v OpenAI Ilya Sutskevera než málem explodovali na otázku "can it propose new mathematical theorems" na kterou řekl "are you sure that the current models can't already do that?" 😃 [Q* - Clues to the Puzzle? - YouTube](https://youtu.be/ARf0WyFau0A?si=d95Lqlojx704pWSF&t=646) tak uvidíme kam až to půjde. Autoři ví o overfittingu, tady o tomto fenoménu mluví: [Ilya Sutskever: Deep Learning | Lex Fridman Podcast #94 - YouTube](https://youtu.be/13CZPWmke6A?si=DXmuFpozjCL5RDki&t=2164) předpokládám že se pořád snaží za každou cenu najít ten sweet spot v double descentu mezi overfittingem (když dimenze daz se rovná dimenzi modelu) a až moc velkým parametrickým prostorem relativně k množstvím dat rozbíjející přenost, což právě vede k dalšímu perfect fitu (po perfect fitu před iniciálním overfitingem v double descentu) a zároveň generalizabilitě díky menší sensitivitě na malý změny na kterým stojí většina LLMs. Zároveň tu využívá scaling laws. Je pravda že overfitting se děje i lokálně, něco preventnout jde líp, něco hůř. Ale ta empirická alchymie kolem toho jak to minimalizovat je cool: sposta metod trénování a benchmarků se právě snaží dělat takový tasky, aby se overfitting co nejvíc eliminoval, aby nutil flexibilitu, ale zároveň ne ztrátu konkrétních vědomostí nedávno vyšla i přímo architektura co nutí co nejvíc minimalizaci overfittingu a underfittingu vedoucí k humanlike systematický generalizaci, kombinující narrow a wide inteligenci [Human-like systematic generalization through a meta-learning neural network | Nature](https://www.nature.com/articles/s41586-023-06668-3) Větší inovace vidím právě k většímu přiblížení se k architektuře AlphaFoldu a GNoME, využívající jejich konkrétní problem solving co vede ke specializovaným inovacím a flexibilitu LLMs, což teď AGI labs s ostatníma jinýma trikama dělají typu sebekorekce a chain of thought [Millions of new materials discovered with deep learning - Google DeepMind]([Millions of new materials discovered with deep learning - Google DeepMind](https://deepmind.google/discover/blog/millions-of-new-materials-discovered-with-deep-learning/)) tak uvidíme či se to povede It's gauge theory (symmetry breaking and their restoration via gauge forces) all the way down