[Double descent - Wikipedia](https://en.wikipedia.org/wiki/Double_descent) 36:04 [Ilya Sutskever: Deep Learning | Lex Fridman Podcast #94 - YouTube](https://m.youtube.com/watch?v=13CZPWmke6A) Training stages: Double descent UU Curve 1) Randomness start 2) underfitting catching too simple patterns 3) perfect generalizable sweet spot fit (can be sudden phase change to composing Fourier transforms generalization method, models are full of smooth power laws from lots of phase transitions when they learn individual circuits) [Mechanistic Interpretability - NEEL NANDA (DeepMind) - YouTube](https://youtu.be/_Ygf0GnlwmY?si=lhXiTHjrHFqsxfaF) 4) overfitting with Dim of data equaling Dim of model and being too sensitive to small petrubation changes (Hmm autism), learning all noise patterns too, just lookup table instead of generalizing patterns 5) additional space giving raise to another perfect fit, more generalizability, sensitivity to changes reduces 6) too big parameter space in NN weights relation to data dimensions and amount leading to loss of accuracy again, too big brain, too little data Predict giant dreams into really [GPT-4 Hired Unwitting TaskRabbit Worker By Pretending to Be 'Vision-Impaired' Human](https://www.vice.com/en/article/jg5ew4/gpt4-hired-unwitting-taskrabbit-worker) LLM finetuning is tailored to different country's laws on human rights etc. Centralized top down force which's only task is to create more systemic decentralization in all other modalities What is the structure of your most dominant memeplex replicator in your mind Chain of througt LLM AI lesson planning generator https://fxtwitter.com/rowancheung/status/1718985386267906265?t=DS783FHpZM4Aua03-Q9UkA&s=19 Prompt engineering https://fxtwitter.com/rowancheung/status/1718985386267906265?t=DS783FHpZM4Aua03-Q9UkA&s=19 US AI regulation https://twitter.com/rowancheung/status/1719176100968845390?t=fvigTS39ZxvCHAItuPAzLw&s=19. Chatgpt and social media [Sintra 🤖](https://trysintra.com/) China's Baidu's CEO says its ERNIE AI 'is not inferior in any aspect to GPT-4' https://au.finance.yahoo.com/news/baidus-ceo-says-its-ernie-ai-is-not-inferior-in-any-aspect-to-gpt-4-162333722.html Even if we get Universal Basic Income after Artificial General Intelligence, that doesnt stop the rich from still getting richer or in general more powerful over time even in some alternative system replacing money leading to even stronger AGI tech giant aritocracy which might give rise to all sorts of other inequalities Multiperspectivity acceleration History [History of the World Countryballs - YouTube](https://youtu.be/JJpr8m3Kmhk?si=BxLO95VKZ1Pva-zw) [The History of the World: Every Year - YouTube](https://youtu.be/-6Wu0Q7x5D0?si=xHLoD8Wz3w67HZKT) Alternative Futures [FUTURE OF WORLD: 2023-10 000 - YouTube](https://youtu.be/930mr7AQNAo?si=6Pl1DBPUaSVDiUxI) [Ultimate Alternate Future Of The World [Perfected] Map 2020 - 20001 Á©∂Ê•µÁâà„ɪÁ©∫ÊÉ≥‰∏ñÁïåÊú™Êù•Âú∞Âõ≥ REUPLOADED IN 2022 - YouTube](https://youtu.be/ocV2QInXUxw?si=tgzvd0-lYHr3JsNl) Safe sustainable decentralization democratization acceleration Kulweit Prompt engineering https://twitter.com/moritzkremb/status/1722311998451454255?t=HdULiKRUEmZsaFxhCrwFKw&s=19 Can you trust some big AI labs actually being altruistic or are they manipulative psychopathic Ontological ground of deities/entities - in brain, in different dimension Selfcorrecting LLMs https://twitter.com/IntuitMachine/status/1719754056355217485?t=L7TXSEvDTPP-JG0tNaVMKw&s=19 ChatGPT tips and tricks https://twitter.com/rowancheung/status/1720072551601152497?t=JI0hqaekWPflAW_pkF_Byw&s=19 + prompting "this is important" WizardLM complexification of prompts DialogStudio SalesForce dataset dialogů LLM thought loops [GitHub - mpaepper/llm_agents: Build agents which are controlled by LLMs](https://github.com/mpaepper/llm_agents) [[2105.07190] A Comprehensive Taxonomy for Explainable Artificial Intelligence: A Systematic Survey of Surveys on Methods and Concepts](https://arxiv.org/abs/2105.07190) A Comprehensive Taxonomy for Explainable Artificial Intelligence The degree of talking in new information that adds or edits your belief ecosystem is a metaparameter you can consciously edit [[2311.00212] A Unified Framework to Enforce, Discover, and Promote Symmetry in Machine Learning](https://arxiv.org/abs/2311.00212) A Unified Framework to Enforce, Discover, and Promote Symmetry in Machine Learning automate web design https://twitter.com/sairahul1/status/1720360673203704008 https://www.marktechpost.com/2023/10/31/researchers-from-meta-and-unc-chapel-hill-introduce-branch-solve-merge-a-revolutionary-program-enhancing-large-language-models-performance-in-complex-language-tasks/ Mistral from stratch [Mistral Architecture Explained From Scratch with Sliding Window Attention, KV Caching Explanation - YouTube](https://www.youtube.com/watch?v=teEf4OzQ1IY&feature=youtu.be) Any symmetry breaking encodes information How to solve the teen mental health crisis? Strong regulation on limbic attentional hijacking products like social media that are supported by capitalist incentives to force them align with meaning? Use something else than capitalism? THE FACT THAT OUR CIVILIZATIONAL SYSTEM WORKS WITH ALL ITS FRAGILE INTERCONNECTED SUPPLY CHAINS, ECONOMY, CULTURE, TECHNOLOGY FROM PLOW TO BOOKS TO INDUSTRIAL REVOLUTION TO COMPUTERS TO AI ALL OF WHICH WORKS LIKE BLACK BOX MAGICAL ALCHEMY THAT ALL SOMEHOW SELFORGANIZED SUCH LEARNED INTEGRATED INFORMATION IN A METASTABLE CYBERNETIC COMPLEX SYSTEM RUNNING ON PHYSICAL SUBSTRATE GOVERNED BY MULTISCALE DIVERSITY OF DIFFERENTIAL EQUATIONS IS JUST PURELY ABSOLUTELY RAW INSANELY BATSHIT CRAZY AND INFINITELY MINDBLOWING Progress technology science health abundance economy fixing sustainability Weapon (bioweps, deepfakes) Uncontrolability since its a black Box AI summit Draw AGI modules To have no self is to have all selfs Reality Math modeling course https://twitter.com/jbramburger7/status/1720421141368455596?t=WSxYh2UDVlC3CkUPFOAHbg&s=19 Math of neural networks ML [The Complete Mathematics of Neural Networks and Deep Learning - YouTube](https://youtu.be/Ixl3nykKG9M?si=6m-7CcZN8x6R262L) Best courses on everything regarding machine learning from theory to practice [DeepLearning.AI: Start or Advance Your Career in AI](https://www.deeplearning.ai/) Alignment between artificial and biological systems https://twitter.com/AndrewLampinen/status/1720467878355365964?t=oSEeiipv6r2kJMTSBqMk7A&s=19 Will AI shatter human connections Stanford deep learning with nlo lectures [Stanford CS224N: Natural Language Processing with Deep Learning | 2023 - YouTube](https://youtube.com/playlist?list=PLoROMvodv4rMFqRtEuo6SGjY4XbRIVRd4&si=PDbkvH3nauWvMmaq) [[1412.1875] Statistical Physics of Adaptation](https://arxiv.org/abs/1412.1875?fbclid=IwAR2UY7tUFDHZE71yuKzsjqcCRHELsTPUw9KGYLbkmG8zRqU6Kaklfx69irA) https://medium.com/mlearning-ai/the-chatgpt-list-of-lists-a-collection-of-1500-useful-mind-blowing-and-strange-use-cases-8b14c35eb Utilitarianism is the best attempt to make coherent formal ethical model without contradictions LLM automatic factchecking [Evolution of cortical geometry and its link to function, behaviour and ecology | Nature Communications](https://www.nature.com/articles/s41467-023-37574-x) Evolution of cortical geometry and its link to function, behaviour [ChatGPT](https://chat.openai.com/share/de4d45fa-5ee3-4bd6-9edb-a5e5e39c6c78) asking ChatGPT standard model Asking ChatGPT equations in neuroscience [ChatGPT](https://chat.openai.com/share/a188d3f8-cb64-4c8f-9fe8-6a96d88745d2) Physics equations [ChatGPT](https://chat.openai.com/share/d07569b0-7546-49b4-9a19-ece1b34d2d3f) ML equations [ChatGPT](https://chat.openai.com/share/18280b79-159c-4947-99fa-23e3818893a8) Sociology equations [ChatGPT](https://chat.openai.com/share/165be806-9ec7-4803-baf5-b6c6097ef620) AI interpret equations [ChatGPT](https://chat.openai.com/share/01c3a231-320a-4a29-bebc-62a7a7e54729) Compitational sociology equations [ChatGPT](https://chat.openai.com/share/1045f86e-cca1-4a02-a6b4-55df88fd3761) Fundamental equations https://fb.watch/o6LnmbxGKU/ Shannon's information equation, [Friedmann equations - Wikipedia](https://en.m.wikipedia.org/wiki/Friedmann_equations) gravitation https://twitter.com/thenetrunna/status/1720859919497236825?t=k2Dn7IbtrgAmdcEG4xo7sA&s=19 AI black market emerging AI explained Matthew berman I think tech giants gaining more top down influence faster than nationstates can catch up is current and most likely future outcome in terms of who will have the most power in their hands in the future China's analog ACCEL chip beats Nvidia's A100 at 3,000 times its performance and 4,000 million times higher energy efficiency. [China's AI Analog Chip Claimed To Be 3.7X Faster Than Nvidia's A100 GPU in Computer Vision Tasks (Updated) | Tom's Hardware](https://www.tomshardware.com/tech-industry/semiconductors/chinas-accel-analog-chip-promises-to-outpace-industry-best-in-ai-acceleration-for-vision-tasks) Jhanas are that you increase entropy so much that you end up with all information/no information state of consciousness, where the system has potential to be all possible states but none of them (still restricted by biology and deep functions preventing physically dying), increasing the possible states until you end up with nothing by gradual underfitting https://twitter.com/RCarhartHarris/status/1720562802182762632 Boundlessness [Boundlessness - YouTube](https://youtu.be/RTPa18XMGAg?si=8HNkqEK_VW4sDKYV) [When the Subject Vanishes, Duality ENDS - YouTube](https://youtu.be/uUMl8_IDJQ4?si=VF250s4jgC4Bq8Xj) Paradox is where freedom lies MULTI AGENT LEARNING - LANCELOT DA COSTA [How can we add knowledge to AI agents? - YouTube](https://www.youtube.com/watch?v=qxEfcrmTWO4) [What Can Fetish Research Tell Us About AI?](https://www.astralcodexten.com/p/what-can-fetish-research-tell-us) Feynman lectures on physics https://www.facebook.com/physicsinhistory/posts/pfbid0cwk3pGHGfifvKcMy6HNrtyFMLYBgXQ49jmaLVcHC41CZGZvntq1MKgoYtz2zGXxBl To get AGI, core knowledge, such as object permanence, to constrain the space of world models, should be implemented as hardcoded priors [How can we add knowledge to AI agents? - YouTube](https://www.youtube.com/watch?v=qxEfcrmTWO4) https://www.harvardlds.org/wp-content/uploads/2017/01/SpelkeKinzler07-1.pdf [[2305.18354] LLMs and the Abstraction and Reasoning Corpus: Successes, Failures, and the Importance of Object-based Representations](https://arxiv.org/abs/2305.18354) Similar to FEP [Josh Tenenbaum's home page](https://web.mit.edu/cocosci/josh.html) Free energy principle and Active Inference brings to the table top down explanation of intelligence from physics first principles and AI explainability via factor graphs Transformers can generalize https://twitter.com/AndrewLampinen/status/1721445352400654677?t=rjN3FTAG8FYLAkDN1K_IRQ&s=19 Věřím že tam jsou corporate hry o regulatory capture pro profit a monopolizaci moci, psychopati co dělají altruisty, ale taky věřím že někteří to myslí vážně, minimálně ti co řeší AI risk na týhle úrovni už od ty doby kdy v tom nebyly tak šílený peníze (ale je pravda že power corrupts), nebo Geoffrey Hinton co leavnul Google aby mohl mluvit o AI risku Rozeznat psychopaty od lidí co to myslí vážně je někdy celkem hardcore když mají takovou moc a systém podporuje spíš psychopaty v hrách o profit Chápu, ale kromě něho věřím že spoustu researchers a engineers (nebo jiní) v těch AI firmách co teď explodovali co mají roots v efektivním altruismu a racionalistický komunitě a řeší alignment od doby kdy technologie na úrovni LLMs byly čisté nerealistické scifi to myslí vážně, pokud je power necorruptla, nesežraly je jiný systémový incentivy udržet si místo, nebo jiný faktory jim nezměnily myšlení Universal darwinism [[1910.13443] Multilevel evolutionary developmental optimization (MEDO): A theoretical framework for understanding preferences and selection dynamics](https://arxiv.org/abs/1910.13443) Let's add love for humans in all rlhfs or more directly in representation engineering by finding and adding love for humans vectors in the activations of LLMs on the neural population level [[2310.01405] Representation Engineering: A Top-Down Approach to AI Transparency](https://arxiv.org/abs/2310.01405) Fatser no leaking data in finetuned open source models [Gradient: AI Automation for Enterprise](https://gradient.ai/) Monosemanticity mechanistic interpretability [Reading AI's Mind - Mechanistic Interpretability Explained [Anthropic Research] - YouTube](https://www.youtube.com/watch?v=Mhp8vpOksWw) Automate company https://twitter.com/sairahul1/status/1721822547065688097 AGI tier list https://twitter.com/arankomatsuzaki/status/1721722679105872186 Perhaps advances in artificial intelligence will create an utopia with an universal basic income for all beings, or a cyberpunk dystopia ruled by tech corporate giants When evaliating abilities of LLMs or AIs in general, we cannot study output in isolation, but output in relation to the training set to see if it can generalize and have emergent capabilitues not existent in the training set [Unveiling AI's Illusions: with Gary Marcus and Michael Wooldridge - YouTube](https://youtu.be/V5c4tVAL5OQ?si=M7suylRWp9EWqUTJ) Transformers dont generalize beyond training data https://twitter.com/abacaj/status/1721223737729581437?t=Pum7kS9KmFX1shDSkXbIcA&s=19 Jak se v bezstátí kde není státní monopol of violence globálně vyřeší violence, climate change, sociopaths, freeriders, jiný multipolar traps, sociální služby, healthcare pro všechny, možnost korporátní dystopie,... LLMs personas jailbreaks https://twitter.com/soroushjp/status/1721950722626077067 [Start here - psychotechnologies.org](https://psychotechnologies.org/about/Start+here) Mathematics is the ultimate symbolic strict compression language Every thing is a computationally reducible pocket of the Rualid. What pullback attractor in the statespace of consciousness, or statespace of generative models, or slice of Ruliad, or infinity groupoid, or dynamics of interacting and merging Markov blankets, topological pockets, physics, information, or what part of platonic mathematics, or God, nature, ieaffble, or all possible ontologies, or imaginings, constructions, are "you" inhabiting? [HOW DO WE EXIST IN THE UNIVERSE? - YouTube](https://youtu.be/6iaT-0Dvhnc?si=Hj_0xJ3Dk8dVYN2T) Every thing corresponds to some (quasi)stable mathematical model analyzable and predictable by all sorts of mathematics encoding physical systems evolving in (relativistic) (quantum) spacetime across interacting scales using stochastic dynamical systems theory using machinery from classical+statistical+relativistic+quantum mechanics, from particle to wave to fluid mechanics to markov blankets to cybernetic complex systems. We can take the statespace of things that science analyzes and map out the statespace of mathematics used to analyze them all, and semantic content framing it, automatically by scanning all of science information on the internet using scraping and data summarizing and compressing AI in a giant lookup table or or semantic map or fuzzy nested metagraph with nodes as things and (fuzzy) relations encoding similarityness (what math equations or semantics are used), inclusion, connection, hiearchies, causality etc. Disciplines as evolutionary attractors in scientific progress can be mapped out by semantic similarity and dimensional clustering. All of this could be added as knowledge database for scientific AGI. GPT4 summary of everything linked Put the whole book into my AGI as knowledge [Kalman filter - Wikipedia](https://en.m.wikipedia.org/wiki/Kalman_filter) We see the laws we do because we are such observers supporting them ChatGPT how would you structure Book about philosophy, physics, information, biology, consciousness, wellbeing, meditation, psychedelics, sociology, systems science, world problems, solutions, movements, love, transcendence [ChatGPT](https://chat.openai.com/share/da7b4215-f95e-4530-b558-e61dfdf5764f) Money and power inequality UBI AI paradigms: Symbolic Analytic Perceptron black box Selforganizing (Turing) Corporate aritocracy Ask all SoTA models with prompt engineering tricks Humanity's Knowing, current state and Future Philosophy Science Sociology Economy, types of economical systems, alternatives to money, abundance, inequality Technology Culture, people, health - opression Environment Governance - dystopia, chaos, democracy, other systems Statistics Scenarios Causal power - agents (economical, technological, cultural,...) or agent agnostic - Moloch Holistic visions, ideologies and movements - Political compass Joscha, how AI can change everything Joscha Humanism Tribalism Truthism Progress, wellbeing, sustainability P(doom) P(progress) Collectvism individualism Alignment agendas Representation engineering Antropic superposition What can you do to make the world a better place: On foundations: philosophy, science, technology, infrastructure,... Directly: activism, sustainability, climate, AI integration and alignment, social help, love and kindness, diplomacy, democratic governance, donating anything, cleaning,... Indirectly: make money and donating it, investing, spreading the message,... Goetzel's ontological map Neuroscience: Neuron anatomy Connectivity Network Harmonics Topology AI summit EU, USA AI act regulations in China, UK, Russia, rest of Asia, Africa, South America, middle east, Oceania,... Does quantum probability exist ontologically? Depends: wave function realism yes, baysianism no, superdeterminism,... Does fundamental time exist ontologically? Depends: superdeterminism yes, second law of thermodynamics yes, Carlo rovelli loop quantum gravity emergent, hoffman no [Carlo Rovelli: Loop Quantum Gravity, Relational Time - YouTube](https://youtu.be/r_fUPbBNmBw?si=4vvkl_wlBC6InoeP) [Donald Hoffman: The Nature of Consciousness [Technical] - YouTube](https://youtu.be/CmieNQH7Q4w?si=HxKUQxE-ffbVE2Bi) What distinguishes information processing in rocks from animals, and animals from humans? degree of causal agency, hiearchical processing, spaciotemporal counterfactual planning, complexity, integrated information [HOW DO WE EXIST IN THE UNIVERSE? - YouTube](https://youtu.be/6iaT-0Dvhnc?si=Hj_0xJ3Dk8dVYN2T) Blackhole singularities break laws of physics, spacetime doesnt exist there NNs, CNNs, RNNs, Transformers Thoughts are spaciousness dancing Deepmind AGI tier list Our system: Some problems are local, some problems are global Clitoris Neurotech Meta reading brain Yan lecunn more humanilike AI architecture [The first AI model based on Yann LeCun’s vision for more human-like AI](https://ai.meta.com/blog/yann-lecun-ai-model-i-jepa/) https://openreview.net/pdf?id=BZ5a1r-kVsf [JEPA - A Path Towards Autonomous Machine Intelligence (Paper Explained) - YouTube](https://www.youtube.com/watch?v=jSdHmImyUjk) 1). Classical Mechanics 2). Electricity & Magnetism 3). Thermodynamics 4). Optics 5). Quantum Mechanics/Nuclear Physics [Mathematical physics - Wikipedia](https://en.wikipedia.org/wiki/Mathematical_physics) Methods in neuroscience [ChatGPT](https://chat.openai.com/share/4a69694d-b05a-491a-a550-e2632aaced68) Can we even define ethics in humans to put it into AIs? AI dny vyfocena Ceske priority AI workshop mapa https://www.lesswrong.com/posts/jvGqQGDrYzZM4MyaN/growth-and-form-in-a-toy-model-of-superposition [[2310.06301] Dynamical versus Bayesian Phase Transitions in a Toy Model of Superposition](https://arxiv.org/abs/2310.06301) [Being Effulgent Spaciousness - YouTube](https://www.youtube.com/watch?v=ReOHjMtsY0o) Geopolitical tensions: diplomacy, removal? https://twitter.com/victorveitch/status/1722300572554969090?t=Yf69uwcTq2mJJpF0SZp2dg&s=19 [[2311.03658] The Linear Representation Hypothesis and the Geometry of Large Language Models](https://arxiv.org/abs/2311.03658) The Linear Representation Hypothesis and the Geometry of Large Language Models https://academic.oup.com/bioscience/advance-article/doi/10.1093/biosci/biad080/7319571?login=false Because of how information flows in Transformers its like theyre learning line by line Python code as lne goes deeper knto layers by more training [Andrej Karpathy: Tesla AI, Self-Driving, Optimus, Aliens, and AGI | Lex Fridman Podcast #333 - YouTube](https://youtu.be/cdiD-9MMpb0?si=TMARs05jE7YyPcOI) Reductionism Strong emergence Weak emergence Formal definition of emergence and downwards causality [Reconciling emergences: An information-theoretic approach to identify causal emergence in multivariate data | PLOS Computational Biology](https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1008289) https://twitter.com/_fernando_rosas/status/1258734493403353094?t=GwtuoR-jkfygPQr3Xepk3g&s=19 [CNS*2020: O12, O13, O14 - YouTube](https://youtu.be/0MBHzU7HE6Q?si=VHb-Z0xTxPUN9yqM) AGI taxonomy How will AGI look like AGI timelines [Matt Wolfe - YouTube](https://www.youtube.com/@mreflow) Yoshua Bengio [Generative Flow Networks - Yoshua Bengio](https://yoshuabengio.org/2022/03/05/generative-flow-networks/) [[2202.13903] Bayesian Structure Learning with Generative Flow Networks](https://arxiv.org/abs/2202.13903) [Yoshua Bengio on Pausing More Powerful AI Models and His Work on World Models - YouTube](https://youtu.be/I5xsDMJMdwo?si=bUM7LWPKICSdF-RF) Learned explicit bayesian reasoning in an inference engine on top of world model knowledge sampling many predictive theories probabilistically Mention effective field theories in physics summary [[2101.07884] The Quantum Field Theory on Which the Everyday World Supervenes](https://arxiv.org/abs/2101.07884) The Quantum Field Theory on Which the Everyday World Supervenes Geberalized alignment problem to humans Transformers attentional memory https://twitter.com/IntuitMachine/status/1722727424947859896?t=DBTxAGT941683Wr2Jv9Icg&s=19 AGI autonomous From Wellbeing to Protopia with Artificial Intelligences: Philosophy, Science, Humanity, and its Future I wonder if moral circle has literally some circleshaped neural dynamics correlate in the compassionate part of the brain The Friedmann equations are fundamental formulas in cosmology that describe the expansion of the universe. They show how factors like the density of matter, the curvature of space, and the cosmological constant (a measure of the energy density of empty space) influence this expansion. In simpler terms, these equations tell us how the universe grows and evolves over time based on what it's made of and the properties of space itself. Think of them as cosmic rules that dictate how the universe stretches and evolves from the Big Bang to its future state. [Friedmann equations - Wikipedia](https://en.wikipedia.org/wiki/Friedmann_equations) Wellbeing factors Evolution of NLP Link each course from deeplearning.io Boston dynamics Eureka Generative video Nootropics Oxitocin sniffing for wellbeing [Mental Health Toolkit: Tools to Bolster Your Mood & Mental Health - YouTube](https://youtu.be/CJIXbibQ0jI?si=GfMZiOG0fx-JIc8l) Sleep Sunlight in day, no light in night Movement Nutrition Social connection Stress control Selfactualization Dont be scared of your mind :3 In not aware of any credible evidence of UFOs, but i'm open to it Qualia is the most general concept Visualize book as a tree Linguists make language more structured than it actually is from a statistics perspective