https://twitter.com/bayeslord/status/1753651804141953261 Nobody should suffer from hating their everyday life. I currently have enough money for survival by doing stuff I like doing in conditions compatible with my brain. But not everyone has this luck. I fucking hate this. I want the culture to change. I want the system to change. I don't want this suffering in this world. This being culturally okay is not acceptable indoctrination. Universal basic services will be the future. Culture supporting meaning universally is basic human right. I believe technology can get us there faster and automate what people dislike doing, if directed properly on the technological, political, cultural and economic level while mitigating dystopia and existencial risks. Lots of countries tried UBI already, it's not something impossible. Let's make this list bigger. Decentralized Universal Basic Services. [Universal basic income pilots - Wikipedia](https://en.wikipedia.org/wiki/Universal_basic_income_pilots) "we're all building sections of this system, and e/acc is about rushing to spin it up before current powers can stop us If something exists that isn't a govt, isn't a corp, but can feed and house people, it threatens those in power more than any weapon" https://twitter.com/StateSpaceGodel/status/1753884625867510221?t=4J93-odc6nyHe1KQXjLaYg&s=19 How could Decentralized Universal Basic Income or Services work? https://twitter.com/burny_tech/status/1754229667844805057?t=JEE3O71_5gX3g8A-K-TiJA&s=19 Maybe semidecentralized (digital) UBU/UBS is ideal AI learning like child https://twitter.com/NeuroscienceNew/status/1753169576634834962?t=YxV851VBG6s3sn8uKBI8xw&s=19 Mamba mixture of experts https://twitter.com/IntuitMachine/status/1753791747669315840?t=WNGTOIFkebWPiJfwYxp9Tw&s=19 [First functional human brain tissue produced through 3D printing](https://interestingengineering.com/science/first-functional-human-brain-tissue-produced-through-3d-printing) https://www.sciencedirect.com/science/article/pii/S1934590923004393?via%3Dihub Let's accelerate brain size and efficiency. "GPT-3 is learning first syntax, and then it learns style, and then it learns semantics like arithmetic and so on. It's the long tail of style. And for us, it's different. We start out with semantics." "We learn this in direct interaction with the world, indexically, by pointing at things and touching them. And then we abstract this into a syntax of the representational languages that we are dealing with." - Joscha Bach [Neural manifolds - The Geometry of Behaviour - YouTube](https://youtu.be/QHj9uVmwA_0) Landau theory [ChatGPT](https://chat.openai.com/share/2fdcd3ef-0dbc-4941-83d0-cf92f6a1ea1e) Explain step by step the mathematics of physics of electricity, water molecules, ATP synthesis interacting from first principles starting with standard model quantum field theory in detail explaining each equation and each term [ChatGPT](https://chat.openai.com/share/6ee990d4-1600-4a4d-a95f-a813edb3017b) Create a gigantic list of as many equations as possible governing computers on all levels, from physics to above https://twitter.com/burny_tech/status/1753938388422447204?t=5VWz8nz9irpjYYuzQDqyDw&s=19 AI overview AI equations section in book Write a gigantic list of as many equations as possible from all sciences [ChatGPT](https://chat.openai.com/share/90e0565c-233a-42e9-b819-27b7fd97348c) [ChatGPT](https://chat.openai.com/share/a2c04812-286a-4b8e-941a-336dfdfa2850) [ChatGPT](https://chat.openai.com/share/c9ded041-7245-4c58-ad5e-ae3e2f335165) Intelligence augmentation acceleration I want app for realtime augmented reality that would constantly scan everything around you that you are looking at and constantly explained using AI with Wikipedia and other resources how it works internally from first principles by showing all mathematical quations governing the physics and other science fields and their (by voice, touch, thinking and other ways interactable) expandable in depth step by step technical explanations of what each term does including how it mathematically relates to other equations and having an option to simplify complex topics using analogies or go into in depth infinite rabbithole of fundamental physics and how all other sciences mathematically emerge https://twitter.com/burny_tech/status/1753948818360447277?t=98Ou9Adugn8_ft4gfWl9Jw&s=19 Control theory mathematics [ChatGPT](https://chat.openai.com/share/e39f18ce-1a2d-4c90-b36e-7abf944c1769) [Everything You Need to Know About Control Theory - YouTube](https://youtu.be/lBC1nEq0_nk?si=deMx_KPsjwbwHGIq) [Classical Control Theory - YouTube](https://youtube.com/playlist?list=PLUMWjy5jgHK1NC52DXXrriwihVrYZKqjk&si=wo2LQKuo9TQQ4PMm) [soft question - Visually stunning math concepts which are easy to explain - Mathematics Stack Exchange](https://math.stackexchange.com/questions/733754/visually-stunning-math-concepts-which-are-easy-to-explain) Sabine climate change [A Day in the Life of a Motor Protein - YouTube](https://youtu.be/tMKlPDBRJ1E?si=OtsdrvWINi5T4yy7) [I wasn't worried about climate change. Now I am. - YouTube](https://www.youtube.com/watch?v=4S9sDyooxf4) [Climate Scientist responds to Sabine Hossenfelder on Climate Sensitivity - YouTube](https://www.youtube.com/watch?v=q4EuvpDzlUY) List of engineering disciplines [ChatGPT](https://chat.openai.com/share/31841136-c57d-488b-b3c9-8fa0da5b0524) xLTSM outperforming mamba and Transformers? https://twitter.com/typedfemale/status/1753319671741440340?t=AtqkMMczoquQ_jYozviDPQ&s=19 Do you know any mathematical models of generalization that could be used to design better generalizing AI? [A Mathematical Theory of Generalization: Part I by David H. Wolpert](https://www.complex-systems.com/abstracts/v04_i02_a04/) AI where it only speaks when I'm done talking, and stops when I interrupt it. https://twitter.com/zan2434/status/1753660774541849020?t=uJCr_DpntzykqaXOuyHHsQ&s=19 Rag cheat sheet 2 https://twitter.com/jerryjliu0/status/1753923945789903351?t=H2F2RuxiayvaMDmUFVspXQ&s=19 Rag tree https://twitter.com/bindureddy/status/1753994930366930953?t=uWOzM11yZhfpJKjnS6Tx4A&s=19 Human intelligence isnt special https://twitter.com/itamar_mar/status/1754016399092310459?t=b1na2xAWLUbB9Fak7qoO9g&s=19 Scientists discover a potential way to repair synapses damaged in Alzheimer’s disease. In laboratory mice that have a condition mimicking human Alzheimer’s disease, they found that a protein KIBRA can reverse the memory impairment associated with this type of dementia. https://www.buckinstitute.org/news/buck-scientists-discover-a-potential-way-to-repair-synapses-damaged-in-alzheimers-disease/ Neural Networks on the Brink of Universal Prediction with DeepMind’s Cutting-Edge Approach Learning Universal Predictors [Neural Networks on the Brink of Universal Prediction with DeepMind’s Cutting-Edge Approach | Synced](https://syncedreview.com/2024/01/31/neural-networks-on-the-brink-of-universal-prediction-with-deepminds-cutting-edge-approach/) [[2401.14953] Learning Universal Predictors](https://arxiv.org/abs/2401.14953) "Meta-learning has emerged as a powerful approach to train neural networks to learn new tasks quickly from limited data. Broad exposure to different tasks leads to versatile representations enabling general problem solving. But, what are the limits of meta-learning? In this work, we explore the potential of amortizing the most powerful universal predictor, namely Solomonoff Induction (SI), into neural networks via leveraging meta-learning to its limits. We use Universal Turing Machines (UTMs) to generate training data used to expose networks to a broad range of patterns. We provide theoretical analysis of the UTM data generation processes and meta-training protocols. We conduct comprehensive experiments with neural architectures (e.g. LSTMs, Transformers) and algorithmic data generators of varying complexity and universality. Our results suggest that UTM data is a valuable resource for meta-learning, and that it can be used to train neural networks capable of learning universal prediction strategies." [[2312.09323v3] Perspectives on the State and Future of Deep Learning - 2023](https://arxiv.org/abs/2312.09323v3) adversarial testing as a tool for interpretability: length based overfitting of elementary functions in transformers zrovna sbírám papery ukazující kde je přesně limitace generalizace transformerů, Mamby, a podobných ideálně tak, aby z toho vznikla aplikace nasměrovávání transformerů apod. směrem k lepší generalizaci na circuit úrovni než to dělat blindly, nebo na design lepších architektur (ideálně víc explainable) Bylo by cool kdybychom do pár let měli nějakej pořádnej matematickej model generalizace Kterej bychom mohli aplikovat na model a trénovací data a řeklo by nám to jak kvalitně to generalizuje I kdyby byla totálně ugly, hlavně ať je prediktivní Rich Sutton: "finding good abstractions in state and time is one of the biggest things that's missing - finding features that generalize well, which is an unsolved problem" [I Talked with Rich Sutton - YouTube](https://youtu.be/4feeUJnrrYg?si=aStAIFwHJm0_JLNM&t=1465) zajímalo by mě či by tohle šlo použít na prevenci toho overfittingu, preventnout přes analyzování tý konkrétní overfitting tendency u long sequence length (či to zobecnit) na který jste přišli a podle toho při terénovaní nudgovat learning algoritmus (nebo trénovací data) aby ten overfitting preventnul přes např increased step size (podle geometrie cost landscapu) najít metody jak to donutit 😄 hmm nebo jiný parametry v celým procesu zobecnit je a experimentovat jaký všechny možnosti dávají jaký všechny výsledky dle té metriky hmm, třeba model generalizace by mohl být nějaký weaker no free lunch theorem - some architectures and their details are more general (with more net free lunch) than others in different domains XD metaoptimize everything s co nejmenší kombinatorickou explozí 1/ Announcing the new Lindy: the first platform letting you build a team of AI employees that work together to perform any task — 100x better, faster and cheaper than humans would. https://twitter.com/Altimor/status/1721250514946732190 Prompting methods [[2311.12785] Prompting Frameworks for Large Language Models: A Survey](https://arxiv.org/abs/2311.12785) Goal oriented prompting [[2401.14043] Towards Goal-oriented Prompt Engineering for Large Language Models: A Survey](https://arxiv.org/abs/2401.14043) [[2401.16657] Recovering Mental Representations from Large Language Models with Markov Chain Monte Carlo](https://arxiv.org/abs/2401.16657) [[2303.17276] Humans in Humans Out: On GPT Converging Toward Common Sense in both Success and Failure](https://arxiv.org/abs/2303.17276) [e/acc Leader Beff Jezos vs Doomer Connor Leahy - YouTube](https://youtu.be/0zxi0xSBOaQ?si=h9snDNbFtfikhSvy) https://twitter.com/MLStreetTalk/status/1753507278660026508?t=PeTBBHOiQ2SZMm8oo4cg3g&s=19 E/Acc cost function is maximizing the total free energy dissipation of our civilizational cybernetic complex dynamical system in time, the bigger the system gets, the more free energy it dissipates, which I think is what a big amount of protechnology longtermist effective altruists want too, but with utilitarian spice (minimize suffering) added to it more often, which can be in synergy with eachother. The disagreement for a lot of people seems to be on the level of - evaluating how various technologies or other things pose existencial or smaller risks or unsustaibility and therefore increasing significianly the chances of radically minimizing this number that's being optimized, - and with this difference E/Acc is also more fine with taking bigger risks to get to higher utility equilibrium faster, - and also with the assumption that top down governance in practice mostly leads to lower utility as power seeking agents explot it for tyrrany happens more often and is more probable than governance towards growth, - and assuming almost bottomup decentralized unregulated technocapital machine itself is better at this growth and freedom than topdown control - and if we discuss UBI or Universal Basic Services, E/Acc would more likely opt in for a decentralized version Free energy (thermodynamics, free energy principle) Dirac equation Can we abstract decision making into generál frameworks with true false flow charts? https://twitter.com/Andercot/status/1754281477712552044?t=2mwMjJIMmI-UGxqiGN3rTQ&s=19 https://medium.com/swlh/a-deep-conceptual-guide-to-mutual-information-a5021031fad0 VITALIK BUTERIN EDWARD WITTEN PETER THEIL ANDREJ KARPATHY GEORGE HOTZ https://www.researchgate.net/publication/377723187_Large_Language_Model_based_Multi-Agents_A_Survey_of_Progress_and_Challenges I think therefore I am? More like beyond conceptual comprehension ieffable void is shapeshifting in the shape of all of reality [Geoffrey Hinton | Will digital intelligence replace biological intelligence? - YouTube](https://youtu.be/iHCeAotHZa4?si=xAsrsCz1Vg8TPgSZ) [AI X-Risk: Connor Leahy v. Beff Jezos Debate Recap + Terminal Race Condition Redux and Update - YouTube](https://youtu.be/pCtdI1eCO4E?si=GKKknX2oa75Rexfv) [Moral Graphs: Interview with OpenAI Grant Winners! Meaning Alignment Institute: Aligning Humanity! - YouTube](https://youtu.be/bC2pQ78o754?si=rnHWMx79LtmGYHQo) Transformers are Better than State Space Models at Copying https://twitter.com/_akhaliq/status/1754334655405326482 MIT books https://twitter.com/burny_tech/status/1754624591262023755?t=4hjxa-srPyfloRSgn-z8QA&s=19 [Paper page - StepCoder: Improve Code Generation with Reinforcement Learning from Compiler Feedback](https://huggingface.co/papers/2402.01391) Magdi multiagent https://twitter.com/IntuitMachine/status/1754628616443347247?t=clKx8v8b7QnGpnhtxw5lnQ&s=19 Atlas robot manual tasks https://twitter.com/tsarnick/status/1754619657376686460?t=RWwRBqfesRmwYshX74kOIQ&s=19 The world is more changeable than most people think AI ID bypass https://twitter.com/josephfcox/status/1754514949995384996 The world is more changeable than most people think. Our generation constantly talks about giving up because everything is fucked anyway and there's no hope for future and everything will be terrible and we will loose as life. This is selffullfilling prophecy. If you believe in this fatalistic conclusion, it's more likely to happen. If you don't believe in this conclusion and actively fight against it happening, it's less likely to happen. Giving up is the worst solution we can make. Fight against it instead and bring bright future to the world. Axiomatic AI doomer? Ontological skill issue The laws of physics are the only limit to how we can make life more free, growing, happy, sustainable 1) konkrétní část bohatých se nemusela pro jejich wealth tak namáhat relativně k ostatním 2) a konkrétní část k tomu došla nemravníma metodama (ničení země, ničení wellbeingu lidí, korupce, kradení, ničení naší budoucnosti, atd...) A nemyslí si to jen chudí, myslí si to i ti co vydělávaj větší prachy a mají v sobě špetku morálky, to je fakt blbej argument, znám hodně takových lidí. Já teďka od novýho roku vydělávám 750 Kč na hodinu (což by bylo 130 000 Kč měsíčně na plnej úvazek) a chci aktivně proti všem těmhle rakovinným buňkám v našem systému tlačit. Lidi mají tendenci si pomáhat, pokud nejsou čistě selfish. Ale vidím že ty nejsi úplně selfish, protože říkáš že chceš pomoct svým potomkům. Ale pak máš například billionares za různýma korporacema nebo v politice, co vydělávají nekonečný prachy na něčem co už teď způsobuje utrpení, ničení zdraví, stability apod. ve společnosti a našem systému přes ničení klima, znečištění prostředí, ničení mentálního zdraví lidem přes sociální sítě, kradení, destabilizaci zdravý ekonomiky, politiky, podpory lidem apod. a lobují šíleny prachy ekonomicky, kulturně, technologicky apod. aby se nepřecházelo na cleaner energie, neznečištováváný netoxický materiály, lepší sociální siítě, zdravější politický, ekonomický, kulturní systém apod. protože by to aktivní ničení našeho systému by jim osobně už tolik nevydělávalo na úkor všech. Tak funguje i rakovina v těle, který pak zabije celej systém. Já nevím jak ty ale osobně kdybych vydělal miliardy jenom pro sebe metodama co ostatním přímo nepomáhají a postavil si palác, mezitím co všichni kolem mě trpí (ne každý má tu možnost, štěstí, motivaci apod. tak vydělávat, což má taky bambilion důvodů) kterým bych mohl pomoct za miliontinu toho co mám, ale já se radši topil ve zlatě ve svým paláci izolovanej od světa, tak bych se cítil fakt blbě a odloučeně od našeho druhu, i kdybych je vydělal eticky nebo zdědil nebo k nim došel štěstím (a ještě horší by bylo kdyby to bylo vydělaný neeticky přes ničení země, ničení wellbeingu lidí, korupce, kradení, ničení naší budoucnosti, atd...) A šílený peníze fakt nejsou ultimátní cesta ke štěstí, spíš tolik kolik potřebuješ na žití, což několik studií potvrzuje, a např Johny cage vydlěal miliony a pak na konci života zpíval o tom jak je to ve výsledku jenom "empire of dirt". Fully random noise and completely patterned order are effectively identical. It’s in the transition between the two where all the interesting bits happen. https://twitter.com/eshear/status/1754679319442723002?t=AmrlFDwS5utgDH6qK-da8g&s=19 AGI is incredible technology and i'm incredibly optimistic about the futu- BUY LAND AND COMPUTE NOW HOLY HSIT https://twitter.com/yacineMTB/status/1754444712352661559?t=BUdrAinEyWjAePenYf4EBA&s=19 DeepSeekMath The 7B open LLM performs nearly on par with GPT-4 on MATH benchmark [[2402.03300] DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://arxiv.org/abs/2402.03300) [Moderna’s mRNA cancer vaccine works even better than thought](https://www.freethink.com/health/cancer-vaccine) [[2401.02731] Parameter-Efficient Sparsity Crafting from Dense to Mixture-of-Experts for Instruction Tuning on General Tasks](https://arxiv.org/abs/2401.02731) [US firm plans to build 10,000 qubit quantum computer by 2026](https://interestingengineering.com/innovation/10000-qubit-quantum-computer-2026) [First UK patients receive experimental messenger RNA cancer therapy | Medical research | The Guardian](https://www.theguardian.com/science/2024/feb/04/first-uk-patients-experimental-messenger-mrna-cancer-therapy-treatment) [Nuclear fusion reaction releases almost twice the energy put in | New Scientist](https://www.newscientist.com/article/2414681-nuclear-fusion-reaction-releases-almost-twice-the-energy-put-in/) [AI chatbots tend to choose violence and nuclear strikes in wargames | New Scientist](https://www.newscientist.com/article/2415488-ai-chatbots-tend-to-choose-violence-and-nuclear-strikes-in-wargames/) [A One-and-Done Injection to Slow Aging? New Study in Mice Opens the Possibility](https://singularityhub.com/2024/02/05/a-one-and-done-injection-to-slow-aging-new-study-in-mice-opens-the-possibility/) [Prophylactic and long-lasting efficacy of senolytic CAR T cells against age-related metabolic dysfunction | Nature Aging](https://www.nature.com/articles/s43587-023-00560-5) [2024 is the Year of the AI AGENT - YouTube](https://www.youtube.com/watch?v=hmt5MnStKUI) 2024 is the Year of the AI AGENT: Summary - Tencent's "Web Voyager" is an end-to-end web agent that can perform tasks like shopping, travel research, and sending emails upon command. - The concept of a "Foundation Agent" is proposed, which aims to be a universal agent capable of generalizing across different realities, whether it's a video game, the physical world, or simulations. - The Rabbit R1, a handheld AI companion, has seen significant consumer interest, with sales reaching 60,000 units in a short period. - The development of these agents involves complex processes, including the use of neuro-symbolic algorithms and extensive data collection from human interactions with various software applications. - The video also explores challenges in AI agent development, such as the limitations of current models in navigating web environments effectively and the potential for AI models to improve in tasks like deception and cooperation in gaming scenarios. - Comparison of AI Models: The video highlights the performance gap between GPT-4 and other models, especially in tasks requiring memory, planning, world modeling, self-reflection, grounding, and spatial navigation. GPT-4 consistently outperforms its competitors, indicating its superior capability in handling complex AI tasks. - Web Voyager by Tencent: A significant focus is placed on Web Voyager, an AI agent that demonstrates a 55.7% task success rate, outperforming GPT-4 in web-based tasks. This agent can interact with websites like Amazon, All Recipes, GitHub, and others to fulfill specific user requests, showcasing its practical utility in navigating and extracting information from the web. - Emergence of Large Action Models (LAMs): The video discusses the development of large action models that excel in navigating various digital environments, including online shopping, spreadsheets, and more. These models represent a step towards more autonomous and capable AI agents. - Impact on Society and Economy: The potential impacts of AI agents on jobs, the economy, and online advertising are pondered. Questions are raised about the future role of AI in daily tasks and high-level decision-making, hinting at a transformative shift in how humans interact with digital environments. - Reflection on the Current Era: The video concludes with a reflective note, suggesting that we are at the cusp of a new era driven by AI advancements. It invokes a sense of nostalgia for the present moment, recognizing the significant changes on the horizon. https://twitter.com/ToughSf/status/1754550452996518099 A method to unilaterally disable all nuclear bombs on Earth, remotely and without countermeasure: https://arxiv.org/pdf/hep-ph/0305062.pdf It uses a 1000 TeV muon->neutrino beam to penetrate right through the Earth and decay near fissile material, forcing it to 'fizzle' and become useless. https://www.marktechpost.com/2024/02/03/a-memes-glimpse-into-the-pinnacle-of-artificial-intelligence-ai-progress-in-a-mamba-series-llm-enlightenment/?amp Mamba: [[2312.00752] Mamba: Linear-Time Sequence Modeling with Selective State Spaces](https://arxiv.org/abs/2312.00752) Mamba MOE: [[2401.04081] MoE-Mamba: Efficient Selective State Space Models with Mixture of Experts](https://arxiv.org/abs/2401.04081) Mambabyte: [[2401.13660] MambaByte: Token-free Selective State Space Model](https://arxiv.org/abs/2401.13660) Self-Rewarding Language Models: [[2401.10020] Self-Rewarding Language Models](https://arxiv.org/abs/2401.10020) Cascade Speculative Drafting: [[2312.11462] Cascade Speculative Drafting for Even Faster LLM Inference](https://arxiv.org/abs/2312.11462) LASER: [[2312.13558] The Truth is in There: Improving Reasoning in Language Models with Layer-Selective Rank Reduction](https://arxiv.org/abs/2312.13558) DRµGS: https://www.reddit.com/r/LocalLLaMA/comments/18toidc/stop_messing_with_sampling_parameters_and_just/ AQLM: [[2401.06118] Extreme Compression of Large Language Models via Additive Quantization](https://arxiv.org/abs/2401.06118)