https://twitter.com/leland_mcinnes/status/1731752287788265726 ML landscape arxiv, The titles and abstracts were combined and encoded with sentence-transformers. Mapping was done with UMAP. Clustering used the fast-hdbscan library. Plotting via @matplotlib and a bunch of custom code that, if testing continues to go well, will be released soon. UMAP visualizations of scientific fields landscapes (custom made or automated clustering) https://twitter.com/norvid_studies/status/1731798340315054463 https://twitter.com/norvid_studies/status/1731799900449378502 maps of science youtube channel map of philosophy Kids are getting worse at math while AI is getting better at math, hmmmmm acapalle sciencie [Nanobot (Havana Parody) | A Capella Science ft. Dorothy Andrusiak - YouTube](https://www.youtube.com/watch?v=ObvxPSQNMGc) llm jailbreak [[2310.08419] Jailbreaking Black Box Large Language Models in Twenty Queries](https://arxiv.org/abs/2310.08419) [[2307.08715] MasterKey: Automated Jailbreak Across Multiple Large Language Model Chatbots](https://arxiv.org/abs/2307.08715) FEP is collection of results from bayesian statistics, statistical mechanics (nonequilibrium thermodynamics), classical mechanics, cybernetics, control theory, dynamical systems in one model modelling adaptive complex systems, mainly organisms recently quantum free energy principle also integrates quantum information theory, quantum mechanics or quantum field theory [OPTICAL COMPUTING with PLASMA: Stanford PhD Defense - YouTube](https://www.youtube.com/watch?v=Mdh2pLwsK8Y) [List of lists of lists - Wikipedia](https://en.wikipedia.org/wiki/List_of_lists_of_lists) Creativity is combinatory uniqueness Coding LLM https://twitter.com/_akhaliq/status/1731865391205425575?t=vR1txNqy5E5deg25E0qkGA&s=19 Conditiona for theory of consciousness https://twitter.com/algekalipso/status/1731806670597284084?t=0rBbkwJKCuCXHmjgcsXI3A&s=19 Reconstruction of brain imagery https://twitter.com/BrianRoemmele/status/1731521774763012145?t=r3BPn4cNPfJKfUYhRk6P_Q&s=19 Shrondinger equation Dirac equation Optics 3b1b Navier strokes Interface with lots of LLMs [LM Studio - Discover, download, and run local LLMs](https://lmstudio.ai/) [GitHub - danny-avila/LibreChat: Enhanced ChatGPT Clone: Features Anthropic, OpenAI, Assistants API, Azure, Groq, GPT-4o, Mistral, OpenRouter, Vertex AI, Gemini, Artifacts, AI model switching, message search, langchain, DALL-E-3, ChatGPT Plugins, OpenAI Functions, Secure Multi-User System, Presets, completely open-source for self-hosting. Actively in public development.](https://github.com/danny-avila/LibreChat) Were not running out of data yet Synthetic data is all you need AI hardware https://twitter.com/EricWollberg/status/1731745015922122960?t=c5RpwSAksFgcONxtXCjyMw&s=19 Biggest LLMs are trained in big part on synthetic data, Pi was trained just on synthetic data Retrieval AI landscape: The best systems right now: SoTA systems, benchmarks The most popular methods right now Promising methods Buddhism map https://www.facebook.com/groups/1735260376685389/permalink/3655014251376649/ Biggest pages do exobrainu Log loss, or general cross entropy loss, MSE treats all errors equally, but log loss penalizes highly confident wrong predictions more severely V teorii chaosu je asi největší věc to že malá změna v počátečních podmínkách může dát drasticky odlišný výsledky, a když máš konečnou přesnost a výpočetní výkon při modelování, tak spoustu věcí neprediktneš s dostatečnou přesností a realibility. Existuje ale spousta zákonů a metod, který tento chaos kompresují a reprezentují/approximují/učí se vzory co nám dokáží předpovídat různé věci s různou přesností. Hledat zákony v počasí a klima je dost těžký, ale něco máme. Často se tam používají různý numerický a statistický metody, fluid dynamics jsou super. Je to hodně o modelování pravděpodobností, třeba to že bude 100000 stupňů v plusu nebo mínusu je dost nepravděpodobný, pokud nás nesežere supernova. U změny klimatu tyto modely konvergují na to že se jsou extrémy počasí globálně s vyšší pravděpodobností v průměru častější, což vidíme jak ve statistikách v minulosti, tak v predikovaných o budoucnosti se současnými trendy, kde lidi mají solidní statisticky kauzální faktor. Hází se na to i AI která bývá lepší než klasický modely. Lidí se potom snaží reverse engineernout jaký algorithmy se ty ML modely naučily a ty pak používat. To že máme různé differenciální rovnice v chemii nebo biologii, jako zákony na vyšší úrovni abstrakci nad extrémně komplexní chaotickou fundamentální fyzikou je mega cool. Ale třeba zákony společnosti jsou skoro neexistující, a hází se na to všechno od statistiky, numerických metod, klasický fyziky, teorie her,... https://ct24admin.ceskatelevize.cz/veda/3631915-umela-inteligence-uz-umi-predpovidat-pocasi-dela-skvele-ale-dopousti-se-i-hrubych-chyb [ECMWF | Charts](https://charts.ecmwf.int/?facets=%7B%22Product%20type%22%3A%5B%22Experimental%3A%20Machine%20learning%20models%22%5D%7D) https://phys.org/news/2023-09-artificial-intelligence-climate.html [AR6 Synthesis Report: Climate Change 2023](https://www.ipcc.ch/report/ar6/syr/) Chtělo by to aby někdo jako DeepMind na AI climate prediction hodil triliony compute, když už udělali revoluci v protein foldingu a material science 😄 https://aps.arxiv.org/abs/2310.03684 LLM defender against jailbreaks Shortcut rule: Models like overfitting on training data which doesnt generalize to test set [[2004.07780] Shortcut Learning in Deep Neural Networks](https://arxiv.org/abs/2004.07780) 35:00 [Mechanistic Interpretability - NEEL NANDA (DeepMind) - YouTube](https://youtu.be/_Ygf0GnlwmY?si=hfongqD5SwiY8oOc) Phases of grokking, gradual generalization followed by sudden clean up, not sudden generalizing phase change 1) Memorization 2) Platou, transition from memorization to generalization, circuit formation, cleaning up the memorization weights while still maintaining good loss 3) Grok into generalized algorithm like trigonometry composition achieving better loss than memorization on tasks requiring generalization Another generalizing algorithmic circuit that Transformers learn that disproves the stochastic parrot hypothesis: Induction heads Look back over the sequence for previous instances of the current token (call it A), find the token that came after it last time (call it B), and then predict that the same completion will occur again (e.g. forming the sequence [A][B] … [A] → [B]). https://www.lesswrong.com/posts/TvrfY4c9eaGLeyDkE/induction-heads-illustrated Another one is finite state automata [[2302.03025] A Toy Model of Universality: Reverse Engineering How Networks Learn Group Operations](https://arxiv.org/abs/2302.03025) A Toy Model of Universality: Reverse Engineering How Networks Learn Group Operations Representation theory So the ego removing chemical is 5-meo-dmt, a very small amount of pyrazolam, and dexmedetomidine via an inhaler The reliable control of hypomania is basically microdosing to very low doses of 3-meo-pcp The latter has many reports back when it was available of using it in this limited fashion and its reliability as such a drug Symmetry breaking magnets aligning in cold environment AI alignment formal methods, behavioral evaluations Higgs symmetry breaking [How the Higgs Mechanism Give Things Mass - YouTube](https://youtu.be/G0Q4UAiKacw?si=vTFNjCJ2KdagcO7c) potential energy falling into mexican hat valley breaking global symmetry which gives particles mass and unifes electromagnetism with weak nuclear force Graph of history of technology [Technology over the long run: zoom out to see how dramatically the world can change within a lifetime - Our World in Data](https://ourworldindata.org/technology-long-run?fbclid=IwAR3XHsREm5RHsM8xlgDAUOXqE_-wp2jtqVfH0ZAWzm2MEc5U7UOyqbl6Xdc) Open source LLM landscape https://twitter.com/IntuitMachine/status/1732001940110815677?t=4Z_L5ssuFQl5IZxHLqhAdg&s=19 LLMs masking https://twitter.com/maxkreminski/status/1730343002462388728?t=IlFaq4nJeM4w6QCJNOZ20w&s=19 [Symmetry breaking - Wikipedia](https://en.wikipedia.org/wiki/Symmetry_breaking) [Spontaneous symmetry breaking - Wikipedia](https://en.wikipedia.org/wiki/Spontaneous_symmetry_breaking) [A new quantum algorithm for classical mechanics with an exponential speedup](https://blog.research.google/2023/12/a-new-quantum-algorithm-for-classical.html) A new quantum algorithm for classical mechanics with an exponential speedup thermodynamic AI, instead of resisting entropy, using it for compute https://fxtwitter.com/ColesThermoAI/status/1625368583789334528 https://twitter.com/ColesThermoAI/status/1731119599297638695 [Extropic assembles itself from the future](https://www.extropic.ai/accelerate) Patrick Coles thermodynamic AI [Thermodynamic AI and the Fluctuation Frontier | Qiskit Seminar Series with Patrick Coles - YouTube](https://www.youtube.com/watch?v=dd1jURhLR8Y) https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3010743/ The Intense World Theory – A Unifying Theory of the Neurobiology of Autism P=NP [P vs. NP: The Biggest Puzzle in Computer Science - YouTube](https://www.youtube.com/watch?v=pQsdygaYcE4) [Complexity Theory’s 50-Year Journey to the Limits of Knowledge | Quanta Magazine](https://www.quantamagazine.org/complexity-theorys-50-year-journey-to-the-limits-of-knowledge-20230817/) P = easy to solve (in polynomial timeú, easy to check if result is valid NP = difficult to solve, easy to check Metacomplexity prompt engineering is all you need https://twitter.com/IntuitMachine/status/1730547176206459099 statistical mechanics in 1 hour [Teach Yourself Statistical Mechanics In One Video - YouTube](https://www.youtube.com/watch?v=zFAxiRAiM24) [Phenomenology (physics) - Wikipedia](https://en.wikipedia.org/wiki/Phenomenology_(physics)) Lots of science is physicalistic or dualistic, some try idealism, i go for merging complexity theory (systems and compsci) quantum complexity theory algorithmic information theory [What is the Pumping Lemma - YouTube](https://www.youtube.com/watch?v=qtnNyUlO6vU) Pumping Lemma Computer science: mthetheoretical computer science, applied computer science Theoretical physics, applied physics theoretical systems science, applied systems science worldviews are just useful tools for different contexts you can switch between like masks remodernism is magical consciousness, modernism is science and postmodernism is relativism metamodernism implements all these truths at the same time in all or in different contexts subjective experiences dont automatically generalize to everyone's this is an interestign cogntiive fallacy i see everywhere its argument of the type of "just try communicating with God and you'll 100% believe in God" assuming 0% error rate and through confirmation bias ignoring all other evidence closemindedness radical subjectivism i just dislike when everyone is trying to make me also subscribe to their "one true solution/objective truth" and ignore all others im always like, let me be free from your constrains i prefer cognitive liberty trust me bro i have 0% error rate ill make you similarly fundamentalist my memeplex will succesfully replicate on the depeest levels i am the perfect replicator Palm-E robotics, LLM robots doing physical tasks [RT-X and the Dawn of Large Multimodal Models: Google Breakthrough and 160-page Report Highlights - YouTube](https://www.youtube.com/watch?v=GZdytTKeGYM) [Scaling up learning across many different robot types - Google DeepMind](https://deepmind.google/discover/blog/scaling-up-learning-across-many-different-robot-types/) [Watch Google’s AI robot autonomously complete an array of physical tasks - YouTube](https://www.youtube.com/watch?v=j6O_uePUKKI) 1,200 tokens per second for Llama 2 7B on H100 [Optimum-NVIDIA Unlocking blazingly fast LLM inference in just 1 line of code](https://huggingface.co/blog/optimum-nvidia) LLM visualized [LLM Visualization](https://bbycroft.net/llm) [Introducing Gemini: Google’s most capable AI model yet](https://blog.google/technology/ai/google-gemini-ai/) apple joins open source https://twitter.com/deliprao/status/1732250132614184970 introduction to metaphysics [How Long is Now? | Introduction to Metaphysics ü™ê - YouTube](https://www.youtube.com/watch?v=8HzIlKe--NU) I am something everything is doing M theory [M Theory | Towards a theory of everything? - YouTube](https://www.youtube.com/watch?v=plU31-qtZ78) automated web dev [Builder.io: Ship twice as much, twice as fast](https://www.builder.io/) automated mobile app dev flutterflow [You probably won‚Äôt survive 2024... Top 10 Tech Trends - YouTube](https://www.youtube.com/watch?v=vyQv563Y-fk) lab grown babies [Lab-grown babies could become a reality within five years - Earth.com](https://www.earth.com/news/lab-grown-babies-revolutionary-science-or-ethical-disaster/) https://twitter.com/burny_tech/status/1708651158435184826 https://twitter.com/SchmidhuberAI/status/1732430359969571014 Schmidhuber Q* [GitHub - stas00/ml-engineering: Machine Learning Engineering Open Book](https://github.com/stas00/ml-engineering) https://twitter.com/BasedBeffJezos/status/1732591658158747814/photo/1 https://twitter.com/robbensinger/status/1732711254836547682/photo/2 agi soon/never could/couldnt go wrong landscpae of opinions and predictions Gemini architecture https://twitter.com/srush_nlp/status/1732427569352323401?t=s0l_YCVsPEjVI-_lD45MAw&s=19 Electrons are the only true commodity Inflation is about predictability and wage stickiness // why deflation bad Economy as ecology, economy as search-space; need for forest fires, simulated annealing Safron bioai [Frontiers | Editorial: Bio A.I. - from embodied cognition to enactive robotics](https://www.frontiersin.org/articles/10.3389/fnbot.2023.1301993/full) Maudlin hating superdeterminism [Tim Maudlin | Bell’s Theorem and Beyond: Nobody Understands Quantum Mechanics | The Cartesian Cafe - YouTube](https://youtu.be/mRT5zXAwvBs?si=18JHX3QweECkv5Vf) support vector machines [Support Vector Machine (SVM) in 2 minutes - YouTube](https://www.youtube.com/watch?v=_YPScrckx28) [The Kernel Trick in Support Vector Machine (SVM) - YouTube](https://www.youtube.com/watch?v=Q7vT0--5VII) [[1203.3271] The thermodynamics of prediction](https://arxiv.org/abs/1203.3271) The thermodynamics of prediction Markov cetegories https://ieeexplore.ieee.org/document/10302372 QRI is the EA equivalent of Learyism, the movement to liberate the human mind from the cruel shackles of endogenous neuromodulation, filtered through the lens of high functioning autism https://twitter.com/Plinz/status/1728684394297360471 Statistical mechanics lecture [The role of statistical mechanics - YouTube](https://www.youtube.com/watch?v=YcPGZpl0Mug) [Statistical physics of biological systems: From molecules to minds - 1 of 4 - YouTube](https://www.youtube.com/watch?v=nqKeip4tnUE&t=5378s) Statistical physics of biological systems: From molecules to minds William Bialek [No Turning Back: The Nonequilibrium Statistical Thermodynamics of becoming (and remaining) Life-Like - YouTube](https://www.youtube.com/watch?v=10cVVHKCRWw) No Turning Back: The Nonequilibrium Statistical Thermodynamics of becoming (and remaining) Life-Like [Autism as a disorder of dimensionality – Opentheory.net](https://opentheory.net/2023/05/autism-as-a-disorder-of-dimensionality/) [Testing predictive coding theories of autism spectrum disorder using models of active inference | PLOS Computational Biology](https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1011473) testing predicve coding theories of autism build your world model from first principles [Milky Eggs » Blog Archive » What the Accelerationist’s Political Compass says about us](https://milkyeggs.com/society/what-the-accelerationists-political-compass-says-about-us/) antidepressants systems theory explanation https://twitter.com/s_r_constantin/status/1732769202488881634 věřím že je naděje že se bude věda a technologie ohledně zlepšování wellbeingu v jednotlivcích a společnosti a nasazení v praxi v naší planetární infrastruktuře (fungující rady, terapie, léky, neurotechnologie, zdravější sociální systémy, apod.) časem zlepšovat víc a víc než je teďka a lidi k tomu budou mít větší a větší přístup (free healthcare for all!) věřím že AI revoluce tomu může strašně moc pomoct věřím že svět bude v globálním kontextu víc zlepšovat než zhoršovat co se týče všeho co s tímto souvisí věřím v to a chci pro to udělat co nejvíc aby se šance na to co nejvíc zvýšila there is hope! I believe there is hope that the science and technology around improving wellbeing in individuals and society and putting it into practice in our planetary infrastructure (working selfhelp cultrue, therapies, drugs, neurotechnologies, healthier social systems, etc.) will improve more and more over time than it is now and people will have more and more access to it (free healthcare for all!) I believe the AI revolution can help a lot! I believe that the world will get better than worse in a global context! i believe we can steer all the factors influencing the future towards goodness! I believe in it, and I want to do as much as I can to increase the chances of it! i believe there is hope! sometimes trying to direct brain from point a to point b on the web without attention getting caught by something else feels almost impossible Fear not the schizo who has 10,000 ideas, fear the schizo-autist who has one idea and tries to explain 10,000 things adhd plus autism plus giftedness https://twitter.com/forthrighter/status/1732115431098323162 Feynman, Sagan The Mechanical Universe Ghost in the Shell Social VR Worakls certain friends Muse The Sequences Star Maker The Beginning of Infinity Psychedelics Twitter Bret victor, jonathan blow, bryan jhonson https://twitter.com/AnnaLeptikon/status/1732707839381045430 UBI in Kenya https://conference.nber.org/conf_papers/f192616.pdf bayesian learning of priors https://twitter.com/CRSegerie/status/1732885322067853472 Have gigantic dreams and go for them no matter what. If you fail, try gain. If you fail again, try again. Consistency and persistence is key. AGIs everywhere will happen. I dont think it can be stopped. Certain risks are real and might be or already are happening. Like strenghtening of totalitarian tendencies by using AI to destabilize democracy. I think the best available strategy we have is to create counter AIs as defence. I wanna find application from mathematics of string theory or loop quantum gravity for reverse engineering neural networks in mechanistic interpretability New AI open source advancements https://twitter.com/IntuitMachine/status/1733459697494036493?t=VUaKFfaOgI5W6WPsy433Sw&s=19 There exists some minimal social glue between any two people Adhd x autism x giftedness https://www.facebook.com/katyhigginsleemft/photos/a.8448708408476470/8628671087146867/?type=3 [[2311.04254] Everything of Thoughts: Defying the Law of Penrose Triangle for Thought Generation](https://arxiv.org/abs/2311.04254) [Biophysicists Uncover Powerful Symmetries in Living Tissue | Quanta Magazine](https://www.quantamagazine.org/biophysicists-uncover-powerful-symmetries-in-living-tissue-20231025/?fbclid=IwAR3ynF7k7LkZ2uQ9wMYrkwhfKOaaDtwDOG9lii5q3f4FbCqgndDqsrlqV4I) Mistral eval https://twitter.com/burny_tech/status/1733914743381889357?t=ZqnuHxDBFWsVcterc8e-rg&s=19 https://fxtwitter.com/abacaj/status/1733660077154816013 Popsci eh https://twitter.com/burny_tech/status/1733913656285392909?t=D6ulSrWGUOs4Dnhj5kCP5A&s=19 Real time text to speech AI https://twitter.com/LinusEkenstam/status/1733270971967303692?t=ibelwLSOc4gt6lCjI_YtcA&s=19 Hyena statespace Models https://twitter.com/togethercompute/status/1733213267185762411?t=Vvz-ovVnD1-Kv-xfTYTWkg&s=19 https://twitter.com/IntuitMachine/status/1733462200730140761?t=txpbxT-45ls24Tcj3Xr5KA&s=19 https://twitter.com/noahdgoodman/status/1733538547955876183?t=3NIKFLc8B0y06qKuJFnVTA&s=19 https://twitter.com/togethercompute/status/1733213277596041655?t=w5X9U-QA98r_mvJoBX3iGg&s=19 https://fxtwitter.com/tsarnick/status/1733699416366924206?t=8pl_QhztJh9MFk2uKgeyYA&s=19 Nick Bostrom: AI will help us reduce total existential risk by helping reduce biotech and nanotech risk this depends on offense/defense tradeoff and degree of coordination How can we reduce the deploy cost of LLMs by 100x? From GPUs to MLSys, from MoEs to Speculative Decoding, in this blog post, we give a full-stack examination of transformer inference optimization [Notion – The all-in-one workspace for your notes, tasks, wikis, and databases.](https://yaofu.notion.site/Towards-100x-Speedup-Full-Stack-Transformer-Inference-Optimization-43124c3688e14cffaf2f1d6cbdf26c6c) [Leverage Dopamine to Overcome Procrastination & Optimize Effort | Huberman Lab Podcast - YouTube](https://youtu.be/K-TW2Chpz4k?si=BOHilObr-XXBZC9m) You can approximate dopamine as molecule of motivation and convergence and reinforcement Dopamine fast isnt really about minimizing it but about redirecting it Excercise cold showers good nutrients meaning cultivation substances ML research papers breakdown https://twitter.com/rasbt/status/1733845997791408486?t=4-3QexFRaZyYyqbI8OpO9g&s=19 https://twitter.com/dair_ai/status/1733846365631828142?t=9YthhCOybR27iCZy-jcRog&s=19 AGI pól compass meme https://twitter.com/VitalikButerin/status/1729251822391447904?t=GJxpHIogxIxMeualuctHgA&s=19 LLMs for mathematicians [[2312.04556] Large Language Models for Mathematicians](https://arxiv.org/abs/2312.04556) Segmentation SotA [EfficientSAM](https://yformer.github.io/efficient-sam/?fbclid=IwAR0sIzfTniE_PdIJPvrjCFytmF-as-CdA81a_X00MYONR0Jb4KtwzHAHfjM) Mech interp reproduction crisis https://twitter.com/StephenLCasper/status/1733905055957344375?t=_MY_wheQhZAbMR3NKkC83A&s=19 Suffering in string vacua https://twitter.com/algekalipso/status/1733688696006901929?t=_jQ30otqq703CUo31214rw&s=19 QRI brain https://twitter.com/algekalipso/status/1733754308368458102?t=HEPgN9fZJCoMBisrGj27CA&s=19 I feel like you can get objective information processing of a physical process by for example when you consider all possible frames under the definition of information being all the bits needed to fully predict (and potentially control) the system. In this notion QTF itself is a subset of information we can have about a system. Mamba statespaces [Mamba - a replacement for Transformers? - YouTube](https://youtu.be/ouF-H35atOY?si=oixYX6Bl-pXYTKsH) https://twitter.com/algekalipso/status/1726705346792821157?t=_SmjaKslXToWFqgBxOzZfA&s=19 Compassion without math is blind. Math without compassion is empty. 💡 Math & Compassion ❤️ [SpikeGPT: researcher releases code for largest-ever spiking neural network for language generation](https://news.ucsc.edu/2023/03/eshraghian-spikegpt.html) [[2302.13939v1] SpikeGPT: Generative Pre-trained Language Model with Spiking Neural Networks](https://arxiv.org/abs/2302.13939v1) SpikeGPT: Generative Pre-trained Language Model with Spiking Neural Networks Modification of Transformers closer to the brain, token-mixer is a SpikingRWKV layer, which replaces the self-attention mechanism with an RNN-like structure that preservesparallelization This plus neuromorphic chips sounds lovely and like the future as Altman is investing into neuromorphic chips [Dendrites: Why Biological Neurons Are Deep Neural Networks - YouTube](https://youtu.be/hmtQPrH-gC4?si=UzYUxL4W86WSkFV3) dendrites are deep neural networks Open source model Llama 2 from Meta is better than free version of ChatGPT and can now run on gamer GPU https://fxtwitter.com/IntuitMachine/status/1733463543469507018?t=qRiw9SVMX5-_bU1n--73MA&s=19 My mind is in the superposition of society is both collapsing and expanding to the stars with singularity any second now Testing selfreflection and consciousness in AIs