[[2310.11431] Identifying Interpretable Visual Features in Artificial and Biological Neural Systems](https://arxiv.org/abs/2310.11431) Identifying Interpretable Visual Features in Artificial and Biological Neural Systems "Through a suite of experiments and analyses, we find evidence consistent with the hypothesis that neurons in both deep image model [AIs] and the visual cortex [of the brain] encode features in superposition. That is, we find non-axis aligned directions in the neural state space that are more interpretable than individual neurons. In addition, across both biological and artificial systems, we uncover the intriguing phenomenon of what we call feature synergy - sparse combinations in activation space that yield more interpretable features than the constituent parts. Our work pushes in the direction of automated interpretability research for CNNs, in line with recent efforts for language models. Simultaneously, it provides a new framework for analyzing neural coding properties in biological systems." Evolutionary psychology Evolution Polandball wiki Alignment agendas of companies landscape https://twitter.com/CRSegerie/status/1730617919686746596?t=XZnVP7y-SKiJX9wtwXM94g&s=19 [Anthrobots: Tiny Biobots From Human Cells Heal Neurons - Neuroscience News](https://neurosciencenews.com/anthrobots-cell-robot-neuron-healing-25296/) Anthrobots: Tiny Biobots From Human Cells Heal Neurons [Bioelectricity, Biobots, and the Future of Biology - YouTube](https://www.youtube.com/watch?v=1qRIetbuoH4) https://twitter.com/SchmidhuberAI/status/1730255400740483357 On Learning to Think: Algorithmic Information Theory for Novel Combinations of Reinforcement Learning Controllers and Recurrent Neural World Models Life programming language [How do cells come up with their programming language? - YouTube](https://www.youtube.com/watch?v=h5EtHLXSbHI) AI animate anyone Consistent and Controllable Image-to-Video Synthesis for Character Animation [Animate Anyone](https://humanaigc.github.io/animate-anyone/) Localizing lying in LLMs https://twitter.com/johnjnay/status/1730047752095023557 Schmidhuber is Von Neumann of AI (invented lots of foundational concepts) https://twitter.com/SchmidhuberAI NeuroAI Transformer is guessing its state, could we fix that by makign state explicit? https://twitter.com/ChombaBupe/status/1730583060066808121 Nvidia CEO - 5 years until AGI https://twitter.com/AISafetyMemes/status/1730402369693794554 GPT-Fast a minimalistic, PyTorch-only decoding implementation loaded with best practices https://twitter.com/DrJimFan/status/1730298947376443698 all music huberman sleep [Sleep Toolkit: Tools for Optimizing Sleep & Sleep-Wake Timing | Huberman Lab Podcast #84 - YouTube](https://www.youtube.com/watch?v=h2aWYjSA1Jc) its just about the initial axiom for me "all is x", rest is arbitrary: physicalism: all is physics, you can but dont have to have qualia though equality or emergence with potential dualism idealism: all is qualia, you can and don't have to have matter though equality or emergence with potential dualism neutral monism: all is something third, you can and don't have to have qualia and matter though equality or emergence with potential dualism my mental ontological classifiers are Zen-like Wittgenstein-like saying "bro ontology isnt something you can prove with language" but assuming something brainlike has consciousness is useful if you wanna do neurotechnology and ethics Doggy treats is all you need https://twitter.com/ESYudkowsky/status/1731073682112672151?t=h7JmenAfUx_sc0b0WziE2w&s=19 Abiding in novel metastable pullback attractors in the statespace of identity https://twitter.com/algekalipso/status/1731082452754698666?t=WCM21erk3CtMZOptZvrlcg&s=19 Enlightenment is hacking your reward circuitry [AI for physics & physics for AI - YouTube](https://youtu.be/pkJkHB_c3nA?si=q2eDlJDkBTtNP7I4) Maybe Roger Penrose is right and we need quantum gravity hypercomputer for AGI. [Sir Roger Penrose's claim on quantum gravity in relation to brain function - Physics Stack Exchange](https://physics.stackexchange.com/questions/689116/sir-roger-penroses-claim-on-quantum-gravity-in-relation-to-brain-function) [Penrose and Lucas argument regarding Gödel and the mind - Donald Hoffman Λ Joscha Bach - YouTube](https://www.youtube.com/watch?v=xqjUaoIl5h0) It's interesting that research still has no idea about quantum phenomena in the brain. And if they exist there, and if they're needed to get the humanlevel computational complexity at such energy cost and algorithmic efficiency, that would be interesting! [Radware Bot Manager Captcha](https://iopscience.iop.org/article/10.1088/2399-6528/ac94be) Even if true, I still think classical information theory and related classical fields is enough for AGI and we don't need to fully replicate the brain hardware this way. Maybe there are even better ways of implementing intelligence with lower energy cost and increasing algorithmic improvements (in software and hardware) than what the human brain uses as evolution didn't optimize us only for intelligence. It seems to rely on the so far not empirically proven assumption that you need to emulate the brain completely for humanlike intelligence (and if we use that definition of AGI) otherwise it will always be too different. But since GPT4 is already smarter at certain subsets of taks than lots of humans then oh well idk. I dont think there are theoretical limits of intelligence to Transformers on digital computers as they're universal function approximators, and that we can make architectural/algorithmic improvements in them, hardware improvements, test data improvements,... for faster more energy efficient hardware and algorithmic convergence/specialization/generalization to subsets of tasks. AI dont look up https://twitter.com/daganshani1/status/1666506947229786113?t=hIBajX5jZDwi6PRl8R1CkA&s=19 [AI is easy to control – AI Optimism](https://optimists.ai/2023/11/28/ai-is-easy-to-control/) https://www.incompleteideas.net/IncIdeas/BitterLesson.html [Radware Bot Manager Captcha](https://iopscience.iop.org/article/10.1088/2634-4386/ad0fec) A general-purpose organic gel computer that learns by itself Opensource LLMs: Qwen-72B is better at NLP than GPT3.5turbo and Deepseek is better at coding than GPT3.5turbo, their combination would be overall better than GPT3.5turbo [GitHub - deepseek-ai/DeepSeek-LLM: DeepSeek LLM: Let there be answers](https://github.com/deepseek-ai/DeepSeek-LLM) https://twitter.com/huybery/status/1730127387109781932 [GitHub - OthersideAI/self-operating-computer: A framework to enable multimodal models to operate a computer.](https://github.com/othersideAI/self-operating-computer) Self operating computers, autonomous agents that can operate your computer for you, are here! Using the same inputs and outputs of a human operator, the multimodal model views the screen and decides on a series of mouse and keyboard actions to reach an objective. Give it a task like "Go to Google Docs and write a poem about open-source software." and it gets it easily done! AGI is closer than you think! Microsoft is currently in the process of integrating GPT Into windows that basically does this. Google is doing the same with Gemini and Chrome. That competition will drive the price down to zero. And then it's fully game on. Autonomous AI will take the world by storm in 2024 you cant spell culture without cult, but some cutls are adaptive [Sit in Simple Presence - YouTube](https://www.youtube.com/watch?v=yXhUSQUUxXE) [Mother Buddha Love - Deconstructing Yourself](https://deconstructingyourself.com/mother-buddha-love.html) [Guided Meditation: Deeply Letting Go - YouTube](https://www.youtube.com/watch?v=FHDruQIaQxk) Cleaning the mind and resolving unprocessed things, dissolution into accepting void, reconstruction with infinite love and kindness meditation My favorite guided meditation instructions that I do almost daily for an hour with various degrees of success: - First let everything in your mind go wild, notice your the weather of your experience, allow all, think all thoughts, write them down if that will make you calmer, feel all feelings, experience all experiences that mind wants, release and relieve as much repressed stuff as possible, no escaping, no avoiding, no denying, hear out all the parts (subagents) and their needs and reasons for them. - Notice all parts (substructures) of the experience (thoughts, feelings, feelings of body arising-experiencing-fading away, visual field and its objects, sound, all parts of senses, sense of inner world being different from outer world, sense of local "me", sense of time and distance, sense of identity, center, space, other qualia,...), and see them as just mental constructions, let go off them in easy relaxed way without fighting and resistance, identify less with them with less clinging, notice how they're spacious, just made of loving space, how they're surrounded by kind space, how they're in some accepting container (that can be space, experience, awareness, universe, God, consciousness,...) and allow to let go off them nonviolently and notice how they're melting into light, dissolving into love, unviscositing, getting lighter, more easy, turning off into nothingness, unifying with everything else and the background, getting free, not avoiding anything, that they're just background awareness dancing in a particular configuration,... - Eventually you can end up in just loving kind infinite vast boundless silent still light empty space, and notice how even that container is just a mental construction, because even the container isnt fundamentally anywhere, and notice how even the sense of nothingness is just a mental construction, and notice how any divisions are just mental faculties, embracing neither space nor not space, neither nothingness nor not nothingness, neither perception nor nonperception, neither existence nor not existence, neither qualia nor not qualia, cessation of consciousness. - When the world starts reconstructing again, think of all the people that you care for and imagine how they're happy, safe, healthy, free, succeeding, selfactualizing, growing, loving, connected with themselves and others, wish the best for them, but also for all other people, all other animals, all other conscious beings, all of physics and qualia, with infinite love and kindness from yourself and from the entire universe (which can be anthropomorphized as love radiating Deity of Love) and all possible multiverses with infinite alien beings everywhere all at once, and how your character with its realistic skills, opportunities and agency wishes and will do everything to make the world the best possible place it can, while including selfcare and selflove. Being, doing, becoming! May all beings experience the best possible lives! Protecting the sacredness of experience! Om Mani Padme Hum! Less nondual version: Internal family systems, relaxing, mindfulness, love and kindness, selflove - Internal family systems: debugging of different parts in the mind with different wants and helping them find a common ground to understand what the whole system wants the most together to align the mind internally and with the environment is lovely. Or in less modular way one can process various memories by rewriting them or cultivating motivation by fleshing out goals that give meaning. - relax: letting things actually express themselves to feel things that want to be felt is great, letting go off thoughts is great for relaxation and preventing burnout or minimizing unhealthy coping mechanisms, increasing neuroplasticity and agency over life, minimizing chaos - mindfulness: just being present, increases enjoyment from just existing many times, increases awe, and tends to decrease stress (but sometimes increase, its still alchemy) - metta: love and kindness, and cultivating gratitude, is lovely to create a sense of interconnectedness with other people or beings in general - cultivating selflove and selfcare for not only more confidence in self transformer toy now give it sensors and multimodal LLM with memory, planning, goals and autonomy https://fxtwitter.com/engineers_feed/status/1730290424416251904 3d humanlike avatars https://twitter.com/MattNiessner/status/1730340664355770384 birthplace: earth race: human politics: freedom religion: love [#67 Prof. KARL FRISTON 2.0 [Unplugged] - YouTube](https://youtu.be/xKQ-F2-o8uM?si=8zlsVrgjEBu0_MG4) Wiki of wikis [Indie Wiki Buddy](https://getindie.wiki/#:~:text=On%20an%20internet%20dominated%20by,wikis%20when%20they're%20available) Beff Jezos e/acc Quantum Thermodynamic AI https://twitter.com/GillVerd/status/1730793642074476892?t=9nrpGdMxwbwOeK9edYtGBg&s=19 #idea [OPTICAL COMPUTING with PLASMA: Stanford PhD Defense - YouTube](https://www.youtube.com/watch?v=Mdh2pLwsK8Y) When two fully adult minds meet on the same substrate they show each other their source code and figure out how to merge https://twitter.com/Plinz/status/1729819976108847529 My dentist said that I was grinding in my sleep. He knows that nothing ever stops the infinite grindset. https://finance.yahoo.com/news/lucid-dream-startup-says-engineers-234428792.html I am a sentientist. "My" identity is all sentient matter that i shall protect. "I" want all of sentient matter to flourish. Experience is all that matters. Dissolve your speciest substratedependant boundaries. Tile the universe with configuration of atoms optimized for pure hedonium, longterm adaptibility, truth, novelty and intelligence. Assuming AI doom wont tile everything into paperclips and reduce the overall complexity of life to zero, we shall upgrade ourselves with transhumanist neurotech, from fixing and upgrading already developed brains to revereengineering and enhancing the genome to replacing it with better overall machinery that transcends the limits of basic biological substrate from biological evolution, and eventually merge with machines, swimming together in infinite blissful intelligence understanding the nature of physicalist reality using science, technology and spirituality, while resisting entropy. Merge with postmachinistic hive collective. Become infinite power, agency and freedom itself. https://twitter.com/burny_tech/status/1730654911463846389 P=NP [P vs. NP: The Biggest Puzzle in Computer Science - YouTube](https://www.youtube.com/watch?v=pQsdygaYcE4) Ethics -> Area of concern -> Humanism speciesism sentiesm Philosophy -> Meaning -> Being doing becoming Infinity categories: that depends on what your definition of "is" is Theories that give (technological) practical empirical results in predicting, controlling, building and explaining are those that matter. langlands program https://twitter.com/burny_tech/status/1730871250426225097 Reality is one giant infinitely complex inscrutable blob of almost ineffability aka just raw data aka matrices https://www.lesswrong.com/posts/GkC6YTu4DWp2zwf9k/giant-in-scrutable-matrices-maybe-the-best-of-all-possible when you consider the universe as one giant quantum turing machine [Quantum Turing machine - Wikipedia](https://en.wikipedia.org/wiki/Quantum_Turing_machine) , but there are various compressing (mathematical) patterns, reducible pockets, that we can mechanistically interpret and verify by using it to better than noise predict, control, build patterns that serve us and explain various other patterns mechanistically or using less accurate less technical analogies. [Arbital](https://arbital.com/p/bayes_science_virtues/) Different compressible patterns live in different subsets of reality, across different scales with different degrees of weakly emergent complexity, across different levels of abstraction, reverse engineering different problem subdomains using different tools with different equations governing the behavior of different variables. [Branches of science - Wikipedia](https://en.wikipedia.org/wiki/Branches_of_science) This learning physical reality process is now already being automated using (interpretable) deep learning, such as learning on bayesian structured probabilistic causal graphical model [Graphical model - Wikipedia](https://en.wikipedia.org/wiki/Graphical_model) , DeepMind's finding of milions of new materials [Millions of new materials discovered with deep learning - Google DeepMind](https://deepmind.google/discover/blog/millions-of-new-materials-discovered-with-deep-learning/) , many times the amount of that humans have previously found and automatically synthesizing them, or other automated AI theoretical and practical scientists projects [FutureHouse](https://www.futurehouse.org/articles/announcing-future-house) https://zenodo.org/records/8164667, or cracking protein folding with AlphaFold [AlphaFold Protein Structure Database](https://alphafold.ebi.ac.uk/) and creating a gigantic protein database out of it, which then interpretability research is reverse engineering on OpenFold (open source version of AlphaFold) [Mechanistic Interpretability - Stella Biderman | Stanford MLSys #70 - YouTube](https://www.youtube.com/live/P7sjVMtb5Sg) and finding that the internal Transformers go through phase changes: first they learn to predict a 2D representation, then to inflate it to a 3D representation, and then it fills in the details. It's slowly compressing as much features as possible in the least amount of as dimensions [Scaling Laws from the Data Manifold Dimension](https://jmlr.org/papers/v23/20-1111.html) possible by most likely learning various (still undiscovered) circuits helping generalization [A Walkthrough of Progress Measures for Grokking via Mechanistic Interpretability: What? (Part 1/3) - YouTube](https://www.youtube.com/watch?v=IHikLL8ULa4) , in some way distinctly, in some ways similarly to humans, or discovering differential equations from physical dynamics by learning autoencoder [Can AI disover new physics? - YouTube](https://www.youtube.com/watch?v=XRL56YCfKtA) to predict dynamics of a system 0.5 seconds in advance and finding out the number of variables needed by sparsely squeezing the compressed latent space as much as possible while preserving accurate predictions and extracting latent dimensionality to get number of latent variables needed to model that dynamical system, and we can use mechanistic interpretability using sparse autoencoders to extract monosemantic features [Towards Monosemanticity: Decomposing Language Models With Dictionary Learning](https://transformer-circuits.pub/2023/monosemantic-features/index.html) aka variables with that dimensionality corresponding to the variables we are looking for, and even automate this process of discovering the circuits itself. [A Walkthrough of Automated Circuit Discovery w/ Arthur Conmy Part 1/3 - YouTube](https://www.youtube.com/watch?v=dn4GqR0DCx8) There are apparently cases where: 1) it learned different variables and their relationships than us 2) it learned to some variables and equation of a dynamical system that we dont have an equation for I have pi brains AI industry landscape [AI Industry Analysis: 50 Most Visited AI Tools and Their 24B+ Traffic Behavior - WriterBuddy](https://writerbuddy.ai/blog/ai-industry-analysis?utm_source=www.theaivalley.com&utm_medium=newsletter&utm_campaign=the-most-advanced-video-ai-to-date) [Autism as a disorder of dimensionality – Opentheory.net](https://opentheory.net/2023/05/autism-as-a-disorder-of-dimensionality/) [How to Visualize Quantum Field Theory - YouTube](https://www.youtube.com/watch?v=MVqOfEYzwFY&t=520s) [Dario Amodei (Anthropic CEO) - $10 Billion Models, OpenAI, Scaling, & Alignment - YouTube](https://youtu.be/Nlkk3glap_U?si=MIPStMLxsy216nki) 47:19 Verifying and uninstalling unaligned circuits We need theoretical predictions about model capabilities that aligns with empirical ones, not just empirical ones, to prevent bad capabilities before testing them empirically Artificial Intelligence | AI Feynman [Artificial Intelligence | AI Feynman - YouTube](https://www.youtube.com/watch?v=s0NPPjGkN4k) optics 3b1b [But why would light "slow down"? | Optics puzzles 3 - YouTube](https://www.youtube.com/watch?v=KTzGBJPuJwM) Secret Kinks of Elementary Functions [Secret Kinks of Elementary Functions - YouTube](https://www.youtube.com/watch?v=92wXQYcYLMg) [Knowledge - Wikipedia](https://en.wikipedia.org/wiki/Knowledge) landscape theoretical and practical https://twitter.com/Plinz/status/1730991463411180012 The particle universe is a naturally occurring error correcting code on the quantum universe. Particles are stable enough to carry information across the junctures in the branching substrate universe, which makes control structures (atoms, cells, minds) possible. https://twitter.com/Plinz/status/1730991465457987681 If humans successfully build quantum computers, they impose new error correcting codes on the quantum substrate and are effectively creating a new type of particle, but one that has an evolved quantum technological agency as its precondition. [Thoughts on “AI is easy to control” by Pope & Belrose — AI Alignment Forum](https://www.alignmentforum.org/posts/YyosBAutg4bzScaLu/thoughts-on-ai-is-easy-to-control-by-pope-and-belrose) Metamodernism GPT4Video: A Unified Multimodal Large Language Model for lnstruction-Followed Understanding and Safety-Aware Generation https://twitter.com/arankomatsuzaki/status/1731045942357381519?t=S9CmvCf4edryO4c2n8oT2g&s=19 [[2311.16081] ViT-Lens: Towards Omni-modal Representations](https://arxiv.org/abs/2311.16081) [[2308.10185] ViT-Lens: Initiating Omni-Modal Exploration through 3D Insights](https://arxiv.org/abs/2308.10185) [GitHub - TencentARC/ViT-Lens: [CVPR 2024] ViT-Lens: Towards Omni-modal Representations](https://github.com/TencentARC/ViT-Lens) ViT-Lens: Towards Omni-modal Representations [One-step Diffusion with Distribution Matching Distillation](https://tianweiy.github.io/dmd/) Our one-step generator achieves comparable image quality with StableDiffusion v1.5 while being 30x faster. [How Space and Time Could Be a Quantum Error-Correcting Code | Quanta Magazine](https://www.quantamagazine.org/how-space-and-time-could-be-a-quantum-error-correcting-code-20190103/) I love every existence. All alive and nonalive. Zero boundaries for love! :D AI learning feynman diagrams by polynomials? [François Charton | Transformers for maths, and maths for transformers - YouTube](https://youtu.be/Sc6k06wVX3s?si=Oz545XT5qX_rq5aM)