[Launching LVE: The First Open Repository of LLM Vulnerabilities and Exposures](https://lve-project.org/blog/launching-the-lve-project.html) [Noam Brown: AI vs Humans in Poker and Games of Strategic Negotiation | Lex Fridman Podcast #344 - YouTube](https://www.youtube.com/watch?v=2oHH4aClJQs) [S3 E14 OpenAI Research Scientist Noam Brown on Solving Poker and Diplomacy with AI - YouTube](https://www.youtube.com/watch?v=ceCg90Q9N6Y&pp=ygUKbm9hbSBicm93bg%3D%3D) [No Priors Ep. 1 | With Noam Brown, Research Scientist at Meta - YouTube](https://www.youtube.com/watch?v=AAv1S9E6ogE&pp=ygUKbm9hbSBicm93bg%3D%3D) https://arxiv.org/abs/2306.04618 manim cohomology cotangent categories [Rooney - YouTube](https://www.youtube.com/@rooney5395/videos) Devin AI software engineer https://twitter.com/cognition_labs/status/1767548763134964000 Algorithmic progress in language models: "...the compute required to reach a set performance threshold has halved approximately every 8 months...substantially faster than hardware gains per Moore's Law." https://arxiv.org/abs/2403.05812 [New Idea Solves Three Physics Mysteries at Once: Post Quantum Gravity - YouTube](https://www.youtube.com/watch?v=PBA9ac3vg1w) [Physicists Finally Find a Problem Only Quantum Computers Can Do | Quanta Magazine](https://www.quantamagazine.org/physicists-finally-find-a-problem-only-quantum-computers-can-do-20240312/) after deconstrucing into ieaffable void 5495454 times, holding any structure in experience as "fundamental" doesnt make sense to me anymore but its useful to assume some structures for science and engineering though but they're just approximating the ieffable for more optimal control and i enjoy injecting my computational system with 564498465 different memeplexes battling for attention and dominance trying to form coherent identities and then dissolving and unifying them all in all sorts of nonlinear ways brain is fun general biological computational machine playground whatever the physical correlate of consciousness "makes" "us" is what "is" "us", and rest can be in theory modified into arbitrary form, morphological freedom! being deconstructed is like having much more toys to play with but you gotta build stuff from first principles to make it work instead of evolutionary preinstalled ways to do it and you gotta try hard to make it stick more as stuff needs more reified foundations to be more stable its harder to do local stuff but its easier to do global stuff as you're less limited by boundaries but more limited by that as well in a different sense i feel like most people just operate in their local identity thats more optimized for survival "i" build identities from first principles for different situations the strongest metaidentity "i" try to reify is compassionate care for all sentient beings with as little boundaries as possible why do magnets work the electromagnetic field in the standard model quantum field theory just is maybe its derived from strings or other fundamental mathematical structure but even that one still just is but why it just is why is there something rather than nothing existence is the default? existence is illusion of an observer? being observer is an illusion? sfjiwefoihylshndfhjdfbiwbf The Longest Equation in Physics The model Lagrangian is a mathematical expression that summarizes the Standard Model of particle physics, which is the most successful theory of the fundamental interactions between elementary particles. It is composed of four different parts, each describing a different aspect of the Standard Model, it encapsulates the fundamental forces (except gravity) and elementary particles, describing electromagnetic, weak, and strong interactions through gauge symmetries, incorporating fermions (quarks and leptons), gauge bosons (photon, W and Z bosons, gluons), the Higgs boson, and their interactions. The model Lagrangian is written in a compact notation that uses symbols and operators from quantum field theory, such as covariant derivatives, field strength tensors, Dirac matrices, and gauge group generators. It also uses various constants and parameters that are determined by experiments, such as coupling constants, masses, and mixing angles. It is one of the longest equations in physics because it contains many terms and factors that account for all the possible interactions and symmetries of the Standard Model. It was transcribed by Thomas Gutierrez who derived it from Martinus Veltman's Diagrammatica: The Path to Feynman Diagrams. https://twitter.com/PhysInHistory/status/1767036833047904685 https://twitter.com/ToaLee3/status/1767071418146328884 Equations of physics are equations of God Me and neurodivergent? Nah others are just neuroconvergent in suboptimal local minimas, I prefer to metatraverse the whole landscape of possible neural belief networks Yeah you guessed it right, I'm polyamorous, my partners are the standard model quantum field theory Lagrangian, string theory, loop quantum gravity, asymptotic safe in quantum gravity, post quantum gravity and the whole M-theory landscape and we all do lysergic acid diethylamide together learning and reverse engineering the inner workings of all of reality is my chronic addiction ADHD should be renamed to ungovernable feral attention neurodivergence What we're looking for is how everything works, what makes everything work Scalefree mathematical patterns are love [GENERAL CHEMISTRY explained in 19 Minutes - YouTube](https://www.youtube.com/watch?v=5iTOphGnCtg) https://techxplore.com/news/2024-03-replica-theory-deep-neural-networks.html#google_vignette https://www.pnas.org/doi/10.1073/pnas.2310002121 What do you metaoptimize? https://arxiv.org/abs/2401.04070 [Artificial intelligence helped scientists create a new type of battery](https://www.sciencenews.org/article/artificial-intelligence-new-battery) https://fxtwitter.com/mustafasuleyman/status/1767550900288069984?t=eO0_5ueP06xPBHC6FFNb2Q&s=19 "How close are living systems to optimal states? In our new paper in @PhysRevResearch with @artemyte and @JordiPinero we searched for universal bounds to this problem within the context of free energy harvesting, with special attention to molecular machines" https://twitter.com/ricard_sole/status/1767576416109924565?t=1Eg0EIUyioC_Z_H-7q5A3w&s=19 [ICIS# 107 Gregg Henriques John Vervaeke and Michael Levin - YouTube](https://youtu.be/rAlmzRTbGDE?si=ehVsLc8n9Sf_GRSW) https://phys.org/news/2024-03-neural-networks-mathematical-formula-relevant.html https://www.science.org/doi/10.1126/science.adi5639 https://www.lesswrong.com/posts/kobJymvvcvhbjWFKe/laying-the-foundations-for-vision-and-multimodal-mechanistic https://twitter.com/soniajoseph_/status/1767963316943728779 [Cerebras Systems Unveils World’s Fastest AI Chip with Whopping 4 Trillion Transistors - Cerebras](https://www.cerebras.net/press-release/cerebras-announces-third-generation-wafer-scale-engine) https://twitter.com/Figure_robot/status/1767913661253984474 https://twitter.com/coreylynch/status/1767927194163331345 Figure 01 can now have full conversations with people OpenAI robotics Devin AI programmer https://fxtwitter.com/cognition_labs/status/1767548763134964000 [World’s first major act to regulate AI passed by European lawmakers](https://www.cnbc.com/2024/03/13/european-lawmakers-endorse-worlds-first-major-act-to-regulate-ai.html) https://twitter.com/GoogleDeepMind/status/1767918515585994818 [A generalist AI agent for 3D virtual environments - Google DeepMind](https://deepmind.google/discover/blog/sima-generalist-ai-agent-for-3d-virtual-environments/?utm_source=twitter&utm_medium=social&utm_campaign=SIMA/) Whether you like Devin or not, the exponential is there in either way. This type of project of building an autonomous AI agent is currently being worked on in the background in basically all giant AGI labs that have a disproportionate amount of resources and in tons of smaller companies, so I wonder what else 2024 will manage to bring in this modality. https://arxiv.org/abs/2402.12374 Sequoia: Scalable, Robust, and Hardware-aware Speculative Decoding - Carnegie Mellon University 2024 - Allows running an unquantized Llama2-70B on an RTX4090 with half-second per token latency! [Elliptic Curve ‘Murmurations’ Found With AI Take Flight | Quanta Magazine](https://www.quantamagazine.org/elliptic-curve-murmurations-found-with-ai-take-flight-20240305/) Virgin biological tissue using 0.3 kilowatts not optimized for optimal information processing efficiency and speed but just enough for survival versus chad AGIs/ASIs connected to supercomputers using 25+ megawatts designed from first principles for the most optimal information processing speed and efficiency not constrained by evolution https://twitter.com/burny_tech/status/1767996948529766577 It seems for many people hard to accept that the loss function of human biological machines isnt to maximize raw intelligence, but maybe to a first approximation something along the lines of just barely surviving with as little resources and effort as possible, which can correlate with raw intelligence, but system that is from first principles optimized for the most efficient and fastest information processing at scale will most likely look very different from biological machines in terms of structure and size and so on. How do we save the universe from entropy? It left out communicating through Hawking points in the cosmic microwave background (this assumes Penrose's Conformal Cyclic Cosmology). https://imgur.com/qJxAjec Immortality roadmap [MAIN](http://immortality-roadmap.com/) https://www.lesswrong.com/posts/nDYfkku67fLfkWjah/immortality-roadmap Is the ultimate potential of sentient systems experience all possible experiences? Myslíš že "ty" jsi tvoje osobnost, jako osobnostní software co jede v mozku? já si myslím že to co dělá "nás" je nějakej matematickej zdroj ve fyzice co přímo nemusí mapovat na osobnost, ten vzor co koreluje s vědomím aneb myslím si že můžeš mít prožitek bez jakýkoliv osobnosti whatever the physical correlate of consciousness "makes" "us" is what "is" "us", and rest can be in theory modified into arbitrary form morphological freedom! myslím že to nebude přímo lokalizovaný na nějaký jedný části, ale spíš nějaký více distribuovaně naimplementovaný vzor osobnostní obsah, obsah v paměti, dovednosti, jiný konkrétnější software v prožitku, (biological machine) fyzická schránka... myslím že to všechno jde nahradit čímkoliv jiným a pořád tam bude jeden spojitý prožitek individuála, co jenom radikálně změnil ty části co byly změněny teď jenom která z teorií vědomí v neurovědách má pravdu, pokud má, že to je přesně ten vzor, který koresponduje (kauzálně koreluje) s tím jedním v časoprostoru spojitým prožitkem individuála možnost je vymapovat veškerý (fundamentální?) uložený informace v mozku co tu osobnost tvoří [Predictive coding - Wikipedia](https://en.wikipedia.org/wiki/Predictive_coding) a to nacpat do dostatečně kompatibilních uměle vytvořených neuronek s dostatečně kompatibilní architekturou teď je asi nejlepší generalist výsledek co vyšlo včera od Googlu co se týče hraní her [A generalist AI agent for 3D virtual environments - Google DeepMind](https://deepmind.google/discover/blog/sima-generalist-ai-agent-for-3d-virtual-environments/) pak je pár podobných výsledků v irl robotice v praxi efektivnější solutions využívají wrappery multimodal modelů napojený na strašně tools a decodovani inputu https://fxtwitter.com/AISafetyMemes/status/1767935401136980278 ale iirc ještě není někde pořádně naškálovný řešení na online learning (kromě toho že cpeš věci do context window) Wiki má ale fajn seznam promising approaches [Online machine learning - Wikipedia](https://en.wikipedia.org/wiki/Online_machine_learning) myslím že do pár let to někdo naškáluje a bude to další boom, AI co se dokáže učit nový věci ne přes dlouhý finetuning/přetrénování nebo přes nacpání dat do context window, ale podobně jako člověk přes rychlý a efektivní ukázaní pár examples/pár vyzkoušení si daný činnosti a uložení toho do svý neuronky nejbližší co k tomu v praxi teď je to že mají context window na short term memory a extrerní databáze na long term memory, která má technicky nekonečnou volume, což už teď je v podstatě superhuman 😄 nebo že se delší dobu odpojí na přetrénování no, např Perplexity je v podstatě superhuman model v tom že ví o všem dost generalisticky hodně věcí a má přístup k celýmu netu na opravování konkrétnějších faktů (a nějaký mechanismy filtrace blbých stránek pro lepší zdroje) my často v počítačích fyziku ovládneme aby se chovala jako deterministická von neuman architektura na tranzistorech a pak tu fyziku co supressujeme nazpátek na tomhle simulujeme, to je strašně inefficient pro konkrétní usecases, tím jak mozek fyziku exploituje místo tohoto 😄 na druhou stranu díky tomu jsou počítače tak moc obecný v jiným smyslu než mozky, ve smyslu že na tom jde dost konkétně jednoduše deterministicky programovat to co tam programujeme von neumann architektura jde technicky simulovat i na biologický architektuře, ale to je podobně neefektivní jako simulovat biologickou architekturu na von neumann architektuře podobně neefektivní je simulovat kvantový počítače na klasických počítačích a klasický počítače na kvantových, i když je to možný XD ale někdy jediná možnost, tím kolik máme bitů v klasických počítačích a kolik máme qubitů v kvantových, ale algoritmická komplexita je pak někde jinde když se to má škálovat mikroskopický systémy se nejlíp simulují na kvantových počítačích tím, jak je to blíž ke kvantovce ale všechny ty architektury jsou turingovsky kompletní a tím dokážou technicky naimplementovat libovolný vypočítatelný zkonstruovatelný zdroj co fyzika dovoluje v různých praktických limitacích! spojovací magie matematiky díky čemu se i můžou teoreticky navzájem rekurzivně simulovat OpenAI investuje do neuromorfických čipů, takže je možný že další jejich modely začnou migrovat z digitálních AI chips optimizovaný pro násobení matic na víc analogový [OpenAI Agreed to Buy $51 Million of AI Chips From a Startup Backed by CEO Sam Altman | WIRED](https://www.wired.com/story/openai-buy-ai-chips-startup-sam-altman/) jejich architektura https://arxiv.org/abs/2003.02642 [ActInf MorphStream 002.1 ~ Alon Loeffler "From Neuromorphic Nanowire Networks to Neurons" - YouTube](https://www.youtube.com/watch?v=Wp7OebX_k70) koukám že zkouší i online learning? 🤔 Online dynamical learning and sequence memory with neuromorphic nanowire networks [Online dynamical learning and sequence memory with neuromorphic nanowire networks | Nature Communications](https://www.nature.com/articles/s41467-023-42470-5) [ActInf MorphStream 002.1 ~ Alon Loeffler "From Neuromorphic Nanowire Networks to Neurons" - YouTube](https://www.youtube.com/watch?v=Wp7OebX_k70) [Growing Living Rat Neurons To Play... DOOM? | Part 1 - YouTube](https://www.youtube.com/watch?v=bEXefdbQDjw) [Lab-Grown "Mini-Brain" Learns Pong - Is This Biological Neural Network "Sentient"? - YouTube](https://www.youtube.com/watch?v=67r7fDRBlNc) [Lab-Grown "Mini-brains" Perform Non-Linear Computation, Eat Neurotransmitters, & Go To Space - YouTube](https://www.youtube.com/watch?v=G0XeA_X2Cag) [Hybrid Monkey Breakthrough: A Step Toward Human-Animal Chimeras - YouTube](https://www.youtube.com/watch?v=3ofZ-YWoeCE) [Organisms Are Not Made Of Atoms - YouTube](https://www.youtube.com/watch?v=vaJcmWjMNwo) https://arxiv.org/abs/2403.00910 [Record Quantum Computation at D-Wave: Millions of Years Down To Seconds - YouTube](https://www.youtube.com/watch?v=P2hn9kQHi_s) 100 years ago my great-grandfather worked on a farm then he got replaced by machines my grandfather went on to work in a car factory and he got replaced by robots my father went on to become a McDonald chef but he too was replaced by robots I didn't want to suffer the same fate so I got into programming, they said coding is the new literacy and everyone needs to Learn Python but then Github Copilot started writing a little bit of code but then GPT4 and Claude Opus started writing more snippets of code but then Gemini 1.5 and Cursorsh started writing even more code but then autonomous agents started emerging and started writing even more code but then artificial general superintelligence emerged and we [Improving the robustness of analog deep neural networks through a Bayes-optimized noise injection approach | Communications Engineering](https://www.nature.com/articles/s44172-023-00074-3) https://twitter.com/_vaishnavh/status/1767381869375492429 https://arxiv.org/abs/2403.06963 [Freeman Dyson, Part I: From Physics to the Far Future | Episode 2101 | Closer To Truth - YouTube](https://www.youtube.com/watch?v=SWi_O-MLmB0) Science is not a collection of truth, it is a constant continuing unraveling and exploration of mysteries [Dyson's eternal intelligence - Wikipedia](https://en.wikipedia.org/wiki/Dyson%27s_eternal_intelligence) [Energy-based model - Wikipedia](https://en.wikipedia.org/wiki/Energy-based_model) [Yann LeCun - Self-Supervised Learning: The Dark Matter of Intelligence (FAIR Blog Post Explained) - YouTube](https://www.youtube.com/watch?v=Ag1bw8MfHGQ&pp=ygUSZW5lcmd5LWJhc2VkIG1vZGVs) Distances in machine learning https://twitter.com/FrnkNlsn/status/1768073324129915219?t=s7Jg96yOHLySqG6b-d8k2A&s=19 https://static1.squarespace.com/static/635693acf15a3e2a14a56a4a/t/65ef1ee52e64b52f145ebb49/1710169832137/AIcollaboration.pdf https://twitter.com/Research_FRI/status/1767224829667635585 Inside me are many wolves: a) All existential risks are actually highly improbable or impossible let's not worry b) AI existential risks are not actually that real and highly improbable c) AI existential risks are very real and probable enough to care or very probable Yudkowsky style d) nonAI existential risks are much more probable 1) Giving AI technology to market too much will probably create cyberpunk/ancap dystopia too much 2) Giving AI technology to governments too much will probably create surveillance kind of dystopia too much Bostrom style 3) Halting AI technology will overall slow down technology and make us stuck in undisruped unsustainable dynamics for a longer time and we'll selfdestruct or not really grow as civilization in interesting transhumanist ways Joscha Bach style 4) Accelerating too fast increases AI existential risk and other existential risks indirectly, risks of self-destruction from overly strong disruptions, selfrecursive improvement of AIs Yudkowsky foom style 5) Assuming AI alignment is strongest problem and even if we solved it, that's just not how this technology will be implemented in practice because politics in practice are bleh 6) We can just (naively?) technooptimist ourselves into good future Kurzweil style let's not worry 7) Future will be "normal" and not really radically changing, no big disruptions will actually happen in our lifetime (doubt) 8) Future will be UBI utopia for every being 9) Future will be (postapocaliptic) anarchistic transhumanist survival of the fittest on steroids 10) Future won't be 11) Future will be cool cooperative transhumanist exploration of the galaxy with cool aligned AIs and humans merging with AIs into hybrids 12) Future will be exploration of the galaxy by AIs Additionally: Reality will be somewhere in the middle of these options This big uncertainty in the unpredictability of these futures with timelines of various disruptions by this very limited biological brain halt my confidence of planning to pursue x thing for a longer time being actually meaningful in the local context or in the global context of civilization so it's hard for me to stick to one thing for longer Creating a gigantic map of physics is a complex task that would require a significant amount of space and detail. However, I can provide you with a high-level overview of the main branches of physics and their key concepts. Please note that this is not an exhaustive list, and there are many more sub-fields and concepts within each branch. 1. Classical Mechanics - Newton's laws of motion - Kinematics and dynamics - Work, energy, and power - Rotational motion - Fluid mechanics 2. Thermodynamics and Statistical Mechanics - Laws of thermodynamics - Heat and temperature - Entropy and disorder - Kinetic theory of gases - Statistical ensembles 3. Electromagnetism - Electric charge and fields - Magnetic fields and forces - Maxwell's equations - Electromagnetic waves - Optics 4. Quantum Mechanics - Wave-particle duality - Schrödinger equation - Heisenberg uncertainty principle - Quantum entanglement - Quantum field theory 5. Relativity - Special relativity - Lorentz transformations - Time dilation and length contraction - Mass-energy equivalence - General relativity - Curved spacetime - Gravitational waves - Black holes 6. Particle Physics - Standard Model - Quarks and leptons - Fundamental forces (strong, weak, electromagnetic) - Higgs boson - Cosmology and the early universe 7. Condensed Matter Physics - Solid-state physics - Crystallography - Semiconductors and superconductors - Magnetism - Nanoscience and nanotechnology 8. Astrophysics and Cosmology - Stellar evolution - Galaxies and galaxy clusters - Big Bang theory - Dark matter and dark energy - Cosmic microwave background 9. Biophysics - Biomechanics - Neurophysics - Molecular biophysics - Bioenergetics - Medical physics These branches of physics are interconnected, and many concepts and principles overlap between them. Advances in one field often lead to breakthroughs in others, showcasing the unity and interdependence of physical phenomena across various scales and domains. Btw muzes mergovat LLMs a tvorit shoggothy co maji kombinace vzoru z obou co spojujes, takova sexualni reprodukce LLMs https://fxtwitter.com/svpino/status/1749522361307812320?t=z9g_mXm-3l9rBQILrECGXw&s=19 https://arxiv.org/abs/2401.10491 Az moc dobre, merge of multiple minds into one is the way "The best 7 billion parameter models on the HuggingFace Leaderboard are nearly all merges of other different models. Merging not only works, but it's actually a smart way to get a better model!" <https://huggingface.co/abacusai/Slerp-CM-mist-dpo>