global economy is pyramid scheme https://www.facebook.com/andrew.ic.948/posts/pfbid0BAy8GeZE1bqLswAcQGtDuUd8o1isBzshhp5ju1ZyG49wnJxsXMzzMzZPJfD4rEksl Rozkládá zatím malý language modely na srozumitelný části pomocí trénování sparse autoencoderů 😄 před tím to skoro nešlo protože featury se kompresují do různých tvarů v topologii tý neuronový sítě na zmenšení využití dimenzí s hodně překryvama mezi neuronama (superpozice) a do teď jsme to neuměli rozmotat. Zatím to umíme u malých modelů, teď to snad půjden naškálovat. [Towards Monosemanticity: Decomposing Language Models With Dictionary Learning](https://transformer-circuits.pub/2023/monosemantic-features/index.html) [Anthropic Solved Interpretability? - YouTube](https://www.youtube.com/watch?v=HAxd8DoZaW4) https://twitter.com/ch402/status/1709998674087227859 Violence as risk The current biggest crises, as of 2023, are quite multifaceted and spread across various domains including geopolitics, climate, and technology. Here’s a breakdown based on multiple sources: 1. **Geopolitical Crises:** 2. Palestina s dát - **Russian Aggression:** Russia, described as a rogue state, is posing severe threats to global security, especially to Europe and the U.S. Its actions in Ukraine and asymmetric warfare against the West, alongside escalating cyberattacks and disinformation campaigns are contributing to global instability【32†(Time)】. - **China's Authoritarianism:** Under Xi Jinping's strong command, China's centralized decision-making and assertive foreign policy are creating global challenges, from public health to economic stability, and are provoking resistance from the West and Asian neighbors【32†(Time)】. - **Conflict in Ethiopia:** The civil war in Ethiopia, particularly in the Tigray region, has led to a humanitarian crisis with grave human rights abuses and food shortages【8†(source)】. - **Afghanistan Crisis:** The situation in Afghanistan remains dire with humanitarian concerns following the Taliban takeover, and the international community is urged to provide aid【6†(source)】. 2. **Climate Crises:** - **Climate Change:** The climate crisis is a paramount issue, with rising temperatures leading to more extreme weather events, wildfires, and sea-level rise【7†(source)】. - **Central Sahel Crisis:** This region is facing the impacts of climate change exacerbated by long-term economic underdevelopment and political marginalization, leading to armed conflict and community vulnerability in Mali, Burkina Faso, and Niger【5†(source)】. 3. **Technological and Cyber Risks:** - **Weapons of Mass Disruption:** The advent of generative AI and other technological advancements are posing risks of mass disruption, enabling the creation of deepfakes and manipulation at an unprecedented scale【32†(Time)】. - **Cyber Attacks:** Increased cyber threats and attacks are affecting governments, businesses, and individuals globally, highlighting the need for better cybersecurity measures【9†(source)】. 4. **Health Crises:** - **COVID-19:** The pandemic continues to affect global health, economies, and daily lives, although the exact extent of its impact in 2023 is not detailed in the sources. 5. **Humanitarian Crises:** - **Yemen Crisis:** Yemen continues to face a severe humanitarian crisis with widespread hunger, disease, and displacement【10†(source)】. - **Global Food and Water Scarcity:** Food and water scarcity are also significant concerns, further exacerbated by climate change and conflicts【11†(source)】. 6. **Economic Crises:** - **Global Economic Uncertainty:** Economic challenges like inflation, debt, and the potential of a global recession are looming, exacerbated by the aftermath of the pandemic and geopolitical tensions【12†(source)】【13†(source)】. 7. **Other Crises:** - **Migrant Crises:** Various regions are facing migrant crises, including the U.S. southern border crisis and the crisis in Belarus and other parts of Europe【14†(source)】. - **Education Crisis:** The education sector globally has been severely affected by the pandemic, hindering progress towards achieving the Sustainable Development Goals【15†(source)】. These crises interconnect and often exacerbate each other, underscoring the complex challenges the world is confronting in 2023. Wiki list of crises visualizing utopia studies on mental health Effective Altruism and utilitarism are a map, not a territory, creating on avarage results with less suffering [Hindsight bias - Wikipedia](https://en.wikipedia.org/wiki/Hindsight_bias) https://www.lesswrong.com/tag/planning-fallacy other biases Nedávno v AI safety komunitě vybouchl tenhle paper, co z language modelů už nedělá jenom takový black boxy nerozluštitelných matic. Rozkládá zatím malý language modely na srozumitelný části pomocí trénování sparse autoencoderů 😄 před tím to skoro nešlo protože featury se kompresují do různých tvarů v topologii tý neuronový sítě na zmenšení využití dimenzí s hodně překryvama mezi neuronama (říkají tomu polysemanticity nebo superpozice) a do teď jsme to neuměli rozmotat do monosemantických featur. Zatím to umíme u malých modelů, teď to snad půjden naškálovat. [Representation Engineering: A Top-Down Approach to AI Transparency](https://www.ai-transparency.org/) [Towards Monosemanticity: Decomposing Language Models With Dictionary Learning](https://transformer-circuits.pub/2023/monosemantic-features/index.html) [Anthropic Solved Interpretability? - YouTube](https://www.youtube.com/watch?v=HAxd8DoZaW4) https://twitter.com/ch402/status/1709998674087227859 Líbí se mi jedna alignment agenda co veškerou lidskou morálku, od různých kulturních vzorců v chování a přemýšlení <https://en.wikipedia.org/wiki/Moral_foundations_theory> <https://en.wikipedia.org/wiki/Cultural_universal> , po různý etický teorie <https://dsef.org/wp-content/uploads/2012/07/EthicalTheories.pdf>, dekonstruuje na koodinační/kooperační protokoly, takže technicky pro alignment chceme, ať v základu pořád mají tohle jako background context, pokud se nám podaří tyto vzorce (featury) zoperalizovat, lokalizovat a zobecnit. Jeden projekt se ještě pokouší najít různý featury (takže technicky i ty potřebný pro morálku) v lidským mozku (plus je milion studií na neurální koreláty) a zkopírovat je přes podobnost do transformer architektury, kopírovat stav mozku lidí do transformerů. [Activation vector steering with BCI | Manifund](https://manifund.org/projects/activation-vector-steering-with-bci) Staví na předpokladu, že ty featury jdou v lidským mozku taky lokalizovat, a že platí nějaká forma univerzálnosti učích systémů, kde můžeš přes dostatčnou podobnost featury přesouvat z lidskýho mozku do transformerů a potenciálně naopak. :D if the universe is infinite there could be infinite free energy, but we would have to not dissipate heat as entropy aka useless energy for work, where heat death is all energy converted to useless energy, reversible computing is only possible to approximate asymptotically Fundamental Energy Limits and Reversible Computing Revisited [Fundamental Energy Limits and Reversible Computing Revisited. (Conference) | OSTI.GOV](https://www.osti.gov/servlets/purl/1458032) In future we will look at our ancestors from our transhumanist psychedelic neodharmist superhedonist hypermeaningful hyperinteresting infinitelyloving generally superintelligent allknowing supercooperation, one with the universe, global hyperorganism ultrautopia cluster where we are all utterly connected in the void beyond space and time and laugh about how we permanently transcended evolution's tendency to implement and reinforce "needed existence of suffering and lack of sense of freedom and making distinctions in experience between us in the open individualist field of conscious physics" local minimas while swimming in infinite freedom, lightness, bliss, novelty, knowledge, meaning, love forever as coordination failures and other destabilizing factors and existencial risks have been already solved. But can we beat the second law of thermodynamics (tendency towards entropy) as well? https://twitter.com/burny_tech/status/1711078463212257699 https://pbs.twimg.com/media/F774sWQXQAAUb6z?format=jpg&name=medium bryan johnson ultimate inner alignment of subagents and other shards against one goal of not dying https://twitter.com/bryan_johnson/status/1710810269558136965 Ai alignment Summary [US presidents rate AI alignment agendas - YouTube](https://www.youtube.com/watch?v=02kbWY5mahQ) Agendas: [A comparison of causal scrubbing, causal abstractions, and related methods — AI Alignment Forum](https://www.alignmentforum.org/posts/uLMWMeBG3ruoBRhMW/a-comparison-of-causal-scrubbing-causal-abstractions-and) [GitHub - montemac/activation_additions: Algebraic value editing in pretrained language models](https://github.com/montemac/activation_additions) https://www.lesswrong.com/tag/shard-theory This is how GPT4 could be finetuned for love, being better coding assistant, researcher, or how to make general image generators more specialized! [Steering GPT-2-XL by adding an activation vector — AI Alignment Forum](https://www.alignmentforum.org/posts/5spBue2z2tw4JuDCx/steering-gpt-2-xl-by-adding-an-activation-vector) [Activation Engineering Steering LLM's by adding an activation vector - YouTube](https://www.youtube.com/watch?v=J2Gx6FFEaRY) Editable behaviors in neural networks, reinforcement learning models, looking for cheese in a maze, for example "cheese vector": [Understanding and controlling a maze-solving policy network — AI Alignment Forum](https://www.alignmentforum.org/posts/cAC4AXiNC5ig6jQnc/understanding-and-controlling-a-maze-solving-policy-network) Discovering latent knowledge in the structure of language models [[2212.03827] Discovering Latent Knowledge in Language Models Without Supervision](https://arxiv.org/abs/2212.03827) https://www.lesswrong.com/tag/eliciting-latent-knowledge-elk?sortedBy=new [Cognitive Emulation: A Naive AI Safety Proposal ‚Äî AI Alignment Forum](https://www.alignmentforum.org/posts/ngEvKav9w57XrGQnb/cognitive-emulation-a-naive-ai-safety-proposal) [Infra-Bayesianism - AI Alignment Forum](https://www.alignmentforum.org/s/CmrW8fCmSLK7E25sa) Uncovering Latent Human Wellbeing in LLM Embeddings This demonstrates how language models are developing implicit representations of human utility even without direct preference finetuning. https://www.lesswrong.com/posts/BDTZBPunnvffCfKff/uncovering-latent-human-wellbeing-in-llm-embeddings [GitHub - lhyfst/knowledge-distillation-papers: knowledge distillation papers](https://github.com/lhyfst/knowledge-distillation-papers) https://www.lesswrong.com/posts/bxt7uCiHam4QXrQAA/cyborgism [Barriers to Mechanistic Interpretability for AGI Safety — AI Alignment Forum](https://www.alignmentforum.org/posts/KRDo2afKJtD7bzSM8/barriers-to-mechanistic-interpretability-for-agi-safety) https://www.lesswrong.com/tag/functional-decision-theory https://www.lesswrong.com/tag/logical-induction These language models form finite state automata like assemblies of features. The line between symbolic and nonsymbolic AIs is blurring! If we can interpret the giant language models like this, we can construct a giant map of human knowledge! [Towards Monosemanticity: Decomposing Language Models With Dictionary Learning](https://transformer-circuits.pub/2023/monosemantic-features#phenomenology-fsa) [[2307.03201] Scaling Laws Do Not Scale](https://arxiv.org/abs/2307.03201) as the size of datasets used to train large AI models grows, the number of distinct communities (including demographic groups) whose data is included in a given dataset is likely to grow, each of whom may have different values. As a result, there is an increased risk that communities represented in a dataset may have values or preferences not captured by (or in the worst case, at odds with) the metrics used to evaluate model performance for scaling laws. Is minimal loss algorithm get picked assumption in AIs or biological systems true? Probably not. [Digital physics - Wikipedia](https://en.m.wikipedia.org/wiki/Digital_physics) [Phys. Rev. Lett. 64, 2354 (1990) - Unpredictability and undecidability in dynamical systems](https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.64.2354) Unpredictability and undecidability in dynamical systems [Gödel's incompleteness theorems - Wikipedia](https://en.wikipedia.org/wiki/G%C3%B6del%27s_incompleteness_theorems) [AI-Plans](https://ai-plans.com/) [Two Minute Papers - YouTube](https://youtube.com/@TwoMinutePapers?si=lZdUnGqGc36qEB09) https://twitter.com/burny_tech/status/1711199181753274666 This is insane! Creating a game development, software development, social media account company, all run by ChatGPT agents. [GitHub - OpenBMB/ChatDev: Create Customized Software using Natural Language Idea (through LLM-powered Multi-Agent Collaboration)](https://github.com/OpenBMB/ChatDev) ChatGPT can also use internet now, so you can search for all new relevant information Division into subtasks and specialization has neverending potential. I love how you can basically automate to oblivion and start with "Hey ChatGPT, build me a big company full of researchers analyzing from the internet or running experiments, writing techniques, selling books, selling substances that will increase wellbeing, productivity, blissing people out!" "Hey ChatGPT, build the most efficient company maximizing hedonium the most effectively in the universe" "Hey ChatGPT, build the most effective charity company" "Hey ChatGPT, build a company systematizing minimization of biorisk scientifically, technologically, sociologically, politically and so on" "Hey ChatGPT, build safe AGI" "Hey ChatGPT, build the most efficient company maximizing cooperation in our society" "Hey ChatGPT, build company that creates lookup table for all important knowledge to prevent civilizational selfdestruction and aiming for utopia for all sentience" "Hey ChatGPT, write me a flappy bird game, but hyperrealistic in DMT hyperspace, morphing into a different game every 10 seconds" [Build an Entire AI Agent Workforce | ChatDev and Google Brain "Society of Mind" | AGI User Interface - YouTube](https://www.youtube.com/watch?v=5Zj_zstLLP4) [OpenAI‚Äôs ChatGPT Makes A Game For $1! - YouTube](https://www.youtube.com/watch?v=Zlgkzjndpak) I wonder how effective this could be! Techniques of making models more specialized are emerging everywhere. I wonder what will be possible with Google's Gemini (bigger GPT4 coming in the next few months probably), GPT5 or other models. Maybe even automating choosing the most optimal correct models. For some tasks you want general language model, somewhere you want... just plain old linear regression! :smile: Possibly also using Active Inference architecture for bayes optimal decision making? AI is getting crazy! "Hey multi agent ChatGPT, research and build even better multi agent ChatGPTs and engineer society into transhumanistic sustainable utopia!" List of automated multiagent systems to build arbitrary companies or tasks [GitHub - e2b-dev/awesome-ai-agents: A list of AI autonomous agents](https://github.com/e2b-dev/awesome-ai-agents) [GitHub - Paitesanshi/LLM-Agent-Survey](https://github.com/Paitesanshi/LLM-Agent-Survey) [AI Jason - YouTube](https://www.youtube.com/@AIJasonZ/videos) metagpt autogpt babyAGI agentGPT [GitHub - camel-ai/camel: 🐫 CAMEL: Finding the Scaling Law of Agents. A multi-agent framework. https://www.camel-ai.org](https://github.com/camel-ai/camel) [GitHub - OpenBMB/AgentVerse: 🤖 AgentVerse 🪐 is designed to facilitate the deployment of multiple LLM-based agents in various applications, which primarily provides two frameworks: task-solving and simulation](https://github.com/OpenBMB/AgentVerse) [Build AI agent workforce - Multi agent framework with MetaGPT & chatDev - YouTube](https://www.youtube.com/watch?v=pJwR5pv0_gs) [Generative Research Teams: Active Inference Compositions For Research and Meta-Science](https://zenodo.org/record/8164667) Generative Research Teams: Active Inference Compositions For Research and Meta-Science metagpt + chatdev or autogen multiagent workforce [Build AI agent workforce - Multi agent framework with MetaGPT & chatDev - YouTube](https://www.youtube.com/watch?v=pJwR5pv0_gs) [Autogen - Microsoft's best AI Agent framework that is controllable? - YouTube](https://youtu.be/Bq-0ClZttc8?si=bT7UGwcQVAan_qQ1) [AutoGen Tutorial üöÄ Create Custom AI Agents EASILY (Incredible) - YouTube](https://www.youtube.com/watch?v=vU2S6dVf79M) [Simulating an Epidemic with Autogen - YouTube](https://www.youtube.com/watch?v=Z9NadaZ-1Rc) [Autogen created Snake Game and Space Invaders wow - YouTube](https://www.youtube.com/watch?v=gnn1H4H81IY) [Open-source Code Interpreting for AI Apps](https://e2b.dev/) https://twitter.com/FrancescoSacco1/status/1711288381383028943 Physics meets artificial intelligence, self-organizing LLMs https://twitter.com/VividVoid_/status/1711235744910688420 list of lists of TPOT psychotechnologies and spiritual techniques [AGI Will Not Be A Chatbot - Autonomy, Acceleration, and Arguments Behind the Scenes - YouTube](https://www.youtube.com/watch?v=2bn3S4vOVN4) AGI Will Not Be A Chatbot body is love machine https://twitter.com/algekalipso/status/1711490232703766929 attention awareness amodal love has more wamth and safety empty objects have no reifying singularities (grass fire algorithm steven lehar) torus and other zero eurel characteristics shapes as formalization of cessation lots of chronic pain is muscle tension attention is contraction and implies informaton flow to a singularity cessations are parralel field lines in experience which kills all selfreference from intersecting field lines sauna excercise cold pounge map of substances used in psychiatry Studies on psychotherapheutic methods can say method a works with same efficiency with method b, but they tend to forget to analyze for which groups the various psychotherapies worked better, what is are the individual factors, because psychotherapies, substances, meditation traditions, philosophies or lifestyle methods work differently for different people, they're personalized, and this can be explained through general languages, such as by various neurophenomenological models of the nervous system and consciousness. For example genetics, baseline fluidity, cultural upbrindging, memories, learned skills,... draw andres stuff thinking doing x will solve everything generates infinite motivation [Category Theory For Beginners: Abstract Algebra - YouTube](https://www.youtube.com/watch?v=3Du7rLvP_wI) Category Theory For Beginners: Abstract Algebra [Grothendieck topoi - 1Lab](https://1lab.dev/Topoi.Base.html) Nootropics https://twitter.com/burny_tech/status/1710137451702878628?t=kJ5KzZrxyPDYEd4VM2zqpg&s=19 [The 10 Best Nootropic Supplements to Boost Brain Power](https://www.healthline.com/nutrition/best-nootropic-brain-supplements) https://fxtwitter.com/nootropicguy/status/1637917960378408960 [Shop Qualia Mind](https://neurohacker.com/shop/qualia-mind) https://twitter.com/algekalipso/status/1710043056203059380?t=MuJAD-aY4qq6Y54K_p48Nw&s=19 [blueprint](https://protocol.bryanjohnson.co/) list of game engines inlucding VR for metaverse AI model optimized for science and learning general physical laws and their transfer between contexts [Polymathic](https://polymathic-ai.org/) https://fxtwitter.com/MilesCranmer/status/1711429121220465037 https://fxtwitter.com/mikemccabe210/status/1711481166375620995 Sam Altman AGI timelines [Sam Altman on AGI, Elon Musk, ChatGPT, Neuralink - YouTube](https://www.youtube.com/watch?v=5VCioa_eaPk) Lex Fridman podcast in metaverse [Mark Zuckerberg: First Interview in the Metaverse | Lex Fridman Podcast #398 - YouTube](https://www.youtube.com/watch?v=MVYrJJNdrEg) Mustafa Suleyman: OpenAI's Biggest Threat [Mustafa Suleyman: OpenAI's Biggest Threat - YouTube](https://www.youtube.com/watch?v=QM8BJ9Vqhs0) Future: protopian technological and societal adaptation How will future unfold with AI? Looking for third attractor Mustafa Suleymanai The Coming Wave animation [The Coming Wave animation - YouTube](https://www.youtube.com/watch?v=4ReMavRec7U) https://media.discordapp.net/attachments/991777510641651733/1161156050997760111/image.png z toho jak ho zatím chápu, tak se snaží najít středobod mezi ultimátní regulací/monitoringem (tlak pro větší bezpečnost) a ultimátní akcelerací/svobodou (tlak pro technologické inovace) co nepovede ani ke katastrově ani k surveilence státu (něco mezi PauseAI [Max Tegmark: The Case for Halting AI Development | Lex Fridman Podcast #371 - YouTube](https://www.youtube.com/watch?v=VcVfceTsD0A) [Pause For Thought: The AI Pause Debate - by Scott Alexander](https://www.astralcodexten.com/p/pause-for-thought-the-ai-pause-debate) a e/acc movementem [Joscha Bach—How to Stop Worrying and Love AI - YouTube](https://www.youtube.com/watch?v=YeXHQts3xYM) ) osobně se ale bojím, že automatizace tak jak pokračuje může vést k nekonetrolovatelným neudržitelný dynamice v našem systému typu https://www.lesswrong.com/posts/LpM3EAakwYdS6aRKf/what-multipolar-failure-looks-like-and-robust-agent-agnostic a když ještě vezmu v potaz další problémy čemu lidsvo čelí [What are the most pressing world problems?](https://80000hours.org/problem-profiles/) ale pořád věřím že jsou ty tychnologie zkorigovatelný do kontrolovatelný udržitelný dynamiky přes dostatečně rychlou společenskou, technologickou, ekonomickou, vládnoucí apod. adaptaci... Replict coding LLM How to do your own research [Do your own research. But do it right. - YouTube](https://www.youtube.com/watch?v=nD6hS8WV3ic) [List of academic databases and search engines - Wikipedia](https://en.wikipedia.org/wiki/List_of_academic_databases_and_search_engines) [Elicit: The AI Research Assistant](https://elicit.com/) https://www.lesswrong.com/tag/list-of-links https://www.lesswrong.com/posts/N9afB9b6vzCT2YxZX/ai-alignment-breakthroughs-this-week-10-08-23 https://fxtwitter.com/oscarmoxon/status/1708603929011863871 selfimproving agent with autogen custom AI tutor https://fxtwitter.com/Austen/status/1709234099884580955 Edward Frenkel: Langlands problem,, ToE, Infinity, Ai, String Theory, Death, The Self [Edward Frenkel: Infinity, Ai, String Theory, Death, The Self - YouTube](https://www.youtube.com/watch?v=n_oPMcvHbAc) https://www.researchgate.net/publication/364374410_The_role_of_exercise_in_the_treatment_of_depression_biological_underpinnings_and_clinical_outcomes [Spike-timing-dependent plasticity - Wikipedia](https://en.wikipedia.org/wiki/Spike-timing-dependent_plasticity) [Boltzmann distribution - Wikipedia](https://en.wikipedia.org/wiki/Boltzmann_distribution)