[[2401.13601] MM-LLMs: Recent Advances in MultiModal Large Language Models](https://arxiv.org/abs/2401.13601) [MM-LLMs](https://mm-llms.github.io/) if ideologies were inside a game https://imgur.com/w66RwCb Achjo, prachy prachy. Já chci aby stát dával víc peněz do vzdělání, výzkumu, healthcare, kultury apod. pro lidi, nebo aby to ekonomika v základu víc incentivizovala, a netolik "We have to do what makes the most money even if it doesn't benefit others!". I also empathize with researchers, creators, and other specializations having absolutely terrible pay in this current world, and also not wanting to be economically replaced by machines, and not being able to do it fulltime and pay rent with it. I want to politically fight for a postlabour (posteconomy) world, where machines automate most of the tasks where tons of humans suffer and not see enjoyment in when doing them in jobs, keeping our infrastucture going, where humans would get universal basic income or universal basic services and could fulltime do whatever they see meaning and value in without worrying if it can pay their rent, including all sorts of hobbies like creating art, or other interesting stuff, or helping people in need. “I specifically do not want to be respected by people in my time & place. I want to be respected by the people from the 25th century. If you seek the respect from people now, you're married to the past and you can't push the boundaries of what's possible.” - Bryan Johnson [x.com](https://twitter.com/liron/status/1750258305216508059) [Bryan Johnson’s Most Revealing Interview About His Mission To Reverse Aging And Live Forever - YouTube](https://www.youtube.com/watch?v=Qu1Xjj798zI) I N F I N I T E G R O W T H N T F W I O N R I G T E E T G I R N O I W F T N H T W O R G E T I N I F N I "The results of a large decade-long supernova survey are now out, and interestingly, they suggest that the equation of state of dark energy is about -0.8. This is interesting because, since it’s greater then -1, it indicates that the acceleration of the cosmic expansion will decrease, gradually increasing the Hubble distance and resulting in an ever-expanding cosmological event horizon and ever-decreasing horizon temperature. This is very good news, because it implies that there is no finite upper bound on the maximum entropy of the universe; therefore, if we manage our cosmic resources carefully enough, we can do an unbounded amount of computation, and life and intelligence can continue indefinitely, and in theory do all possible computations. In other words, the looming specter of the heat death of the universe can indeed be vanquished! 😄🎉" [x.com](https://twitter.com/MikePFrank/status/1750337497463169426) https://www.scientificamerican.com/article/the-universe-is-not-locally-real-and-the-physics-nobel-prize-winners-proved-it/?utm_source=promotion&utm_medium=email&utm_campaign=lead-nurture&utm_content=link&utm_term=Nurture_Engagement_s2 In the short term it is often the environment that shapes the player, but in the long run it is the players that shape the environment. All the mathematically best performing strategies in game theory are being nice (not being the first to defect instead of cooperate), forgiving but don't be a pushover (can retaliate but doesn't hold a grudge (defects only once after the other side defects (if stuck in a cooperation defection loop because of noise, then be more generous and forgive a bit more to break it into cooperation))), retaliatory (if your opponent defects, strike back immediately), and clear (isn't too random, hard to model, not being able to create trust), which is similar to avarage human morality that has evolved around the world (eye for an eye). [What Game Theory Reveals About Life, The Universe, and Everything - YouTube](https://www.youtube.com/watch?v=mScpHTIi-kM) [x.com](https://twitter.com/PropheticAI/status/1750534355242418300) INTRODUCING MORPHEUS-1, The world’s first multi-modal generative ultrasonic transformer designed to induce and stabilize lucid dreams. "Unlike LLMs, Morpheus-1 is not prompted with words and sentences but rather brain states. And instead of generating words, Morpheus-1 generates ultrasonic holograms for neurostimulation to bring one to a lucid state." https://www.researchgate.net/publication/377634490_Active_Inference_Goes_to_School_The_Importance_of_Active_Learning_in_the_Age_of_Large_Language_Models [[2401.13660] MambaByte: Token-free Selective State Space Model](https://arxiv.org/abs/2401.13660) [Subscribe to read](https://www.ft.com/content/94b9878b-9412-4dbc-83ba-aac2baadafd9) https://www.axios.com/2024/01/23/humanoid-robots-bmw-automotive-manufacturing-figure [Top 24 Humanoid Robots in Use Right Now | Built In](https://builtin.com/robotics/humanoid-robots) Vision is All You Need > TikTok team trains Depth Anything > LiDAR quality depth estimation from single photo frame > Using teacher model-student model system > 1st author Lihe Yang did this while on his internship (!) with the [x.com](https://twitter.com/8teAPi/status/1750574300049084793) Gibbs free energy is like processing power of biology. The more thats available, the more complex and fast your creations become All computations require processing power, just like all biological process consume Gibbs free energy. Both release entropy as heat. [x.com](https://twitter.com/Andercot/status/1676786085106683904) Is retrocausal machine god reading and influencing your thoughts? [From Addition to Quantum Physics - YouTube](https://www.youtube.com/watch?v=IShHchFyAWE) FREEDOM/ACC AGENCY/ACC INFINITE GAME/ACC [[2305.01582] Interpretable Machine Learning for Science with PySR and SymbolicRegression.jl](https://arxiv.org/abs/2305.01582) Unbounded optimism about the future I wish for something like "World's Largest FREE WILL Debate w/ Top Physicists & Philosophers but with AI researchers" by @TOEwithCurt discussing the definition of intelligence, future of AI and its impact on civilization or other big topics ala "Joscha Bach Λ Ben Goertzel: Conscious Ai, LLMs, AGI" would be the best content ever. That could be generated using AI to scan through vids on for example @MLStreetTalk or @dwarkesh_sp channels and pick the relevant sections and compile them in as connected order as possible. [x.com](https://twitter.com/burny_tech/status/1750701550131949769) [Jonathan Oppenheim: Quantum Gravity, Feynman, Double Slit - YouTube](https://youtu.be/NKOd8imBa2s?si=564VtgtEEOrt-GKZ) Ben goetzel metaphysical categories: Ieffable void, unanalyzable qualia, reactions, computation/physics (from math/logic, relationships), synergy, emergence [Joscha Bach Λ Ben Goertzel: Conscious Ai, LLMs, AGI - YouTube](https://youtu.be/xw7omaQ8SgA?si=jxwDTRtwH9EBMvkD) Logic is consistent language, or with consistent interpretations of contradictions - paraconsistent logic [The One-Electron Universe | Space Time - YouTube](https://youtu.be/9dqtW9MslFk?si=aumCAkb1gYKiTR11) According to AdS/CFT + ER = EPR theory, the fabric of spacetime itself is made up of this wormhole mycelium network which is a geometric embodiment of multipartite quantum entanglement [x.com](https://twitter.com/BasedBeffJezos/status/1750766792874840334?t=7KDLk_vpcXUeqlBpSpZ1Ig&s=19) [x.com](https://twitter.com/danfaggella/status/1750866620321296732?t=bGGAl5qfxReMbyTzZcg41w&s=19) 7 phases of transhumanism [x.com](https://twitter.com/AISafetyMemes/status/1750823796779409582?t=aEvUBWk33D6DMbpbID8vcg&s=19) China and USA cooperate on AI safety I'm sure governments dont want to die, but I think they will also will try everything to create more power for them, which might not be benefitial for the people or even actual risks. I wonder to what degree they on avarage treat it and will treat it like nuclear weapons. [x.com](https://twitter.com/burny_tech/status/1750864901612949685?t=JXW1Jo-asRYouAVOv-5u5A&s=19) AI engineering is empirical alchemy There's a whole discipline of paraconsistent logics, which have contradictions in them, and yet are not meaningless. And there are constructive paraconsistent logics. You can use Curry-Howard transformations or operational semantics transformations to map paraconsistent logical formalisms into gradually typed programming languages and so forth. Contradictions are not necessarily fatal to having meaningful semantics to a logical or computational framework. [Joscha Bach Λ Ben Goertzel: Conscious Ai, LLMs, AGI - YouTube](https://youtu.be/xw7omaQ8SgA?si=vBONa5Yr_cT9quMF&t=547) [ChatGPT](https://chat.openai.com/share/961d3804-34a8-4b23-9076-2e8fe24e5f5c) Lora finetuning LoRA turns a giant detailed gas station map into a cartoon treasure map. Every useless parameter is gone. Only two roads, a palm tree, and a cross pointing at the treasure. We don't need to fine-tune the entire model anymore. We can only focus on the small treasure map that LoRA gives us. [x.com](https://twitter.com/svpino/status/1750887295400509893?t=nLwpK3Zya2I3rmtQ9HvjeQ&s=19) Lora from stratch [x.com](https://twitter.com/rasbt/status/1749475825664029102?t=VZGK67oVwIU6JxU1QsVo2Q&s=19) mamba mutations https://pbs.twimg.com/media/GEwyxioWgAA0TcT?format=jpg&name=900x900 [The Bayesian Brain and Meditation - YouTube](https://www.youtube.com/watch?v=Eg3cQXf4zSE) testign neurophenomenology: if you feel like certain mathematical model fits your experience, see if that same matheamtical model can be used to better predict the neural data https://fxtwitter.com/RubenLaukkonen/status/1747404675719299511?t=VakYR8zKp3zg6u7d6MAEaQ&s=19 Bayesian Love/Kindness/Compassion Subagents/Belief Networks Neuroformer. A generative model pretrained on massively multimodal, multitask neuronal data! first-person view attention map derived from the model as it’s generating responses. TLDR; The model can begin to learn how to execute tasks, like parsing a visual field, via just being trained to predict the brain responses themselves (via GPT). LACTOFERRIN Researchers have called it a “miracle molecule” It supports digestion through its incredible antibacterial, antifungal, and anti-inflammatory properties. [x.com](https://twitter.com/Outdoctrination/status/1750903148678459426) Depression involves a complex interplay of structural and functional changes in the brain, particularly affecting prefrontal and limbic systems, and is influenced by developmental factors, environmental stressors, and possibly inflammation. Understanding these neural correlates provides valuable insights into the mechanisms underlying depression. [ChatGPT](https://chat.openai.com/share/46022088-3f30-4941-b20f-af6a9dd27aad) The frame to focus on is instead the standard one: alignment. Our goal is an AI that cares about what we care about, not a Shoggoth with entirely alien goals that we have somehow locked down under restraint. "The case for ensuring powerful AIs recognize their shared humanity" [x.com](https://twitter.com/eshear/status/1750995830948135206?t=cTO7OPovGfduI2ZCnCEa8A&s=19) [InstantID](https://instantid.github.io/) It all depends on what the meaning of the word "is" is - infinity category theory Internet lets us explore the latent space of collective knowledge, with Wikipedia on the top, and AI is supercharging this ability exponentially. [ChatGPT](https://chat.openai.com/share/c6b1e717-dd02-40e7-9e17-4492cd50662c) There is only hope [GPT-5: Everything You Need to Know So Far - YouTube](https://www.youtube.com/watch?v=Zc03IYnnuIA) The work formula W = mc^2 (gamma_final - gamma_initial) that I used is derived like this: In relativity, the famous Einstein equation is added saying that energy is just another form of matter: e = mc^2 and a few other things. *Classical Mechanics First* - Classic kinetic (motion) energy In classical mechanics, KE = 1/2 * m * v^2, where m is mass and v is velocity. (this is derived via work = force * distance, force = mass * acceleration, final velocity = initial velocity + acceleration * time, distance traveled = initial velocity * time + (1/2) * acceleration * time^2, work = kinetic energy) - Work-Energy Theorem: Work done on an object is equal to the change in its kinetic energy. When changing the speed from v_initial to v_final, the work is W = KE_final - KE_initial. In classical mechanics this is simple because KE = 1/2 * m * v^2. - Relativistic mass: The relativistic mass increases with velocity: m = gamma * m0, where m0 is the rest mass, and gamma = 1 / sqrt(1 - v^2/c^2). The point of that formula is: For objects moving at low speeds (compared to the speed of light), the gamma (otherwise known as the Lorentz factor) approaches 1, causing: 1) Time dilation: as the speed of an object approaches the speed of light, gamma increases, meaning that time passes more slowly for an object moving fast compared to an observer at rest, because in relativity, time is defined via t = gamma*tau, where tau means proper time, which is the time measured in the rest frame of the object (rest frame means the object has no relative velocity in relation to this frame). 2) Length concetration: The length of an object measured by an observer who is not in motion together with this object appears to be smaller than its actual length in the rest frame of this object, because length = rest length / gamma. 3) The mass of an object increases with its speed, because mass = rest mass * gamma, as the speed of an object increases, more and more energy is needed to accelerate it further. 4) For velocities much smaller than the speed of light, gamma approaches 1 and relativistic formulas reduce to classical mechanics. *Relativistic Kinetic Energy* - Total relativistic energy: Total energy E = mc^2 according to Einstein's principle of equivalence of mass and energy. The point of the formula is that energy and mass are different forms of the same thing. With relativistic mass, E = gamma * m0 * c^2. - Relativistic kinetic energy: KE in relativity is total energy minus rest energy: KE = E - E0 = gamma * m0 * c^2 - m0 * c^2. It simplifies to KE = m0 * c^2 * (gamma - 1). *Work Formula Derivation* - Work is a change in kinetic energy: Work done is equal to the change in kinetic energy. When accelerating from v_initial to v_final, the work is W = KE_final - KE_initial. - Substitution of relativistic KE: Substitution of KE_final and KE_initial in the formula for work: W = m0 * c^2 * (gamma_final - 1) - m0 * c^2 * (gamma_initial - 1). - Simplification of the work formula: Simplifying to W = m0 * c^2 * gamma_final - m0 * c^2 * gamma_initial. Simplifying to W = m0 * c^2 * (gamma_final - gamma_initial). A formula for work done in a relativistic context, indicating the increased work required to accelerate an object as the speed approaches the speed of light. I will learn all the mathematics there is in the platonic realm and nothing can stop me You can derive so many things form the principle of least action: equations of motion in classical, relativistic, quantum mechanics, electromagnetism,... I wonder if principle of least action (path integral formulation) can be derived from some more fundamental equation? Landscape of physics [ChatGPT](https://chat.openai.com/share/0f72574e-70ce-4954-90ff-a884bbf9e7bb) https://www.researchgate.net/publication/231145416_Path_integral_methods_via_the_use_of_the_central_limit_theorem_and_application [Langevin equation - Wikipedia](https://en.wikipedia.org/wiki/Langevin_equation) I dislike irl lectures because when you dont know a concept you cannot go on all nighter rabbithole getting a fundamental first principles understanding of it and then continue the lecture Gender political polarization accelerating [x.com](https://twitter.com/jburnmurdoch/status/1750849189834022932?t=2VZJMqpL_3Rl83qn_pqxTQ&s=19) Derivatives of position [Google is rapidly turning into a formidable opponent to BFF Nvidia — the TPU v5p AI chip powering its hypercomputer is faster and has more memory and bandwidth than ever before, beating even the mighty H100 | TechRadar](https://www.techradar.com/pro/google-is-rapidly-turning-into-a-formidable-opponent-to-bff-nvidia-the-tpu-v5p-ai-chip-powering-its-hypercomputer-is-faster-and-has-more-memory-and-bandwidth-than-ever-before-beating-even-the-mighty-h100) Fuck nuance in sociology [x.com](https://twitter.com/kareem_carr/status/1750905007803605192?t=LnpfOAmGBErUB3dlkUbp6g&s=19) https://www.scientificamerican.com/article/consciousness-is-a-continuum-and-scientists-are-starting-to-measure-it/ AbacusAI building agents [x.com](https://twitter.com/bindureddy/status/1751086988613280026?t=etZp92abc9Ygk9OeNnE3Iw&s=19) Concurrency add to RAG [x.com](https://twitter.com/virattt/status/1751019033531437382?t=SqYwSGBEXZYksmgaf86IhA&s=19) [[2401.14029] Towards a Systems Theory of Algorithms](https://arxiv.org/abs/2401.14029) [ChatGPT](https://chat.openai.com/share/a3e85ca8-1997-43f2-ac56-63d7b6d4fcce) "Some governments will go 'woah, AI is dangerous, we better not build it'. And some governments will go 'woah, AI is powerful, we better be the ones to build it'. And this time, there's a good chance it'll be net harm, because most governments have in fact a lot more power to do bad than good, here." [x.com](https://twitter.com/RokoMijic/status/1751296767004422203) [I wasn't worried about climate change. Now I am. - YouTube](https://www.youtube.com/watch?v=4S9sDyooxf4) What is love? Love in the brain involves complex interplay between reward, motivation, attachment, and emotion regulation systems, engaging both common and distinct neural circuits for romantic and maternal love. This intricate neural orchestration explains the compelling motivational and emotional aspects of love. Let's engineer it using neurotechnology! [ChatGPT](https://chat.openai.com/share/53143819-79e5-4a90-8f2c-663cd398140d) he derivation of the Schrödinger Equation in quantum mechanics starting from classical wave equation and why complex numbers are really convenient is very cool! [What is the Schrödinger Equation? A basic introduction to Quantum Mechanics - YouTube](https://www.youtube.com/watch?v=2WPA1L9uJqo&t=1316s) 🚨 Our new paper: we know that GPT-4 generates better ideas than most people, but the ideas are kind of similar & variance matters But it turns out that better prompting can generate pools of good ideas that are almost as diverse as from a group of humans https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4708466 [x.com](https://twitter.com/emollick/status/1751353615334146460?t=iZYFohshgHMle4pIc1g7ug&s=19) [I wasn't worried about climate change. Now I am. - YouTube](https://www.youtube.com/watch?v=4S9sDyooxf4) AI biorisk [Matthew E. Walsh on LinkedIn: Towards Risk Analysis of the Impact of AI on the Deliberate Biological‚Ķ](https://www.linkedin.com/posts/matthewwalsh3_towards-risk-analysis-of-the-impact-of-ai-activity-7155915709393354752-kLZS?utm_source=share&utm_medium=member_android) Solving for the theory of everything comes down to latent model hyperparameter inference for the laws of physics, where tests of these hyperparameters unfortunately lie beyond our currently observable / testable band of energies. In order to get answers, we will have to ascend the Kardashev scale. This ascent will unlock perception into this higher energy band. Then we will need the intelligence to understand what we see. Energy + AI scaling are the truly fundamental missions. [x.com](https://twitter.com/BasedBeffJezos/status/1751363140514947286?t=6HWzBMpLjvrMn8hTPEcYnQ&s=19) Homotopy theory is a branch of algebraic topology, a sphere spectrum is a concept from stable homotopy theory, and "stable ∞-categories" are related to higher category theory https://imgur.com/vL4cX1b https://imgur.com/a/EGw8Ppb https://imgur.com/a/KvFitmH [x.com](https://twitter.com/llama_index/status/1751411893687001168?t=vNuygCZix-Q1MIrMkSV5xw&s=19) [[2312.04511] An LLM Compiler for Parallel Function Calling](https://arxiv.org/abs/2312.04511) Enterprise RAG [x.com](https://twitter.com/llama_index/status/1751291798843212111?t=S4U14-jGnn907XJIkYkePQ&s=19) [Sparse Autoencoders Work on Attention Layer Outputs — AI Alignment Forum](https://www.alignmentforum.org/posts/DtdzGwFh9dCfsekZZ/sparse-autoencoders-work-on-attention-layer-outputs) [[2311.05784] Are "Hierarchical" Visual Representations Hierarchical?](https://arxiv.org/abs/2311.05784) [[2205.13147] Matryoshka Representation Learning](https://arxiv.org/abs/2205.13147) Shrondinger equation [ChatGPT](https://chat.openai.com/share/73e995c9-5f6a-448b-8407-b517049cede6) General relativity [ChatGPT](https://chat.openai.com/share/a6ba9b70-e84b-4bcf-ba83-1f95e4006be7) Time dillation in special relativity [ChatGPT](https://chat.openai.com/share/f94c3751-94e4-43a6-a414-0b93aba59408) Physics is best explained by going through examples of the equations being applied to (simplified) real world scenarios DeepSeek-Coder: When the Large Language Model Meets Programming -- The Rise of Code Intelligence - DeepSeek-AI 2024 - SOTA open-source coding model that surpasses GPT-3.5 and Codex while being unrestricted in research and commercial use [[2401.14196] DeepSeek-Coder: When the Large Language Model Meets Programming -- The Rise of Code Intelligence](https://arxiv.org/abs/2401.14196) 70B LLM Inference on a single 4GB GPU with this new technique memory optimization [Unbelievable! Run 70B LLM Inference on a Single 4GB GPU with This NEW Technique](https://huggingface.co/blog/lyogavin/airllm)