my approach is hoarding all types of information processing systems, all definitons of intelligence, machine intellgience, reasoning, etc. and creating gigantic taxonomies maximal inclusion of all perspectives of all the intelligence nerds fighting that their perspective are the best ones and making collection of perspectives, and potentially bridging/synthetizing them ❤️ I think the most effective learning has many types of learning: imitation learning (watching lectures, reading textbooks, spaced repetition flashcards), reinforcement learning (doing exercises), exploratory divergent exploration (playing with structures) I try to not care if papers confirm or refute what I want to believe. I try to look at hard facts with pure empirical data in as least biased ways possible. But I still have optimism bias and other biases that are never possible to fully escape. LLMs generalize to some degree though That's more than just repeating information Otherwise novel discoveries in math wouldn't be possible with them, like in AlphaEvolve https://youtu.be/vC9nAosXrJw?si=pkgqUkTVpjtDbwvL We can do so much better when it comes to generalization and world model building, yes, but it's not just repeating, as we're already creating new knowledge with them And when you look inside using reverse engineering, you see the generalizing circuits, that are often broken and can be better, but they're there Is that novel math result not an original thought? It's for sure original I guess it depends on how you define "thought" It is different, because no other algoritm that we have that we tried could achieve this result How do you define thought? It has to be a representation on only biological substrate? I see the brain as operating under algorithms and building representations too, among many other things, which is studied by computational neuroscience Brain has different algorithms and differently structured representations on a different hardware, but it's also algorithms and representations on hardware with some similarities in those algorithms, representations and hardware It uses algorithms, that's what the field of computational neuroscience studies among other things Checkout for example https://en.wikipedia.org/wiki/Predictive_coding Algorithm is just a process or set of rules to be followed in calculations or other problem solving operations, which you can find in the brain You have a lot of dedicated regions for specialized information processing, like a region for facial recognition People do compare it in science though. There are differences for sure, but also lots of similarities. See Human-like object concept representations emerge naturally in multimodal large language models https://www.nature.com/articles/s42256-025-01049-z https://techxplore.com/news/2025-06-multimodal-llms-human-brain-representations.html "The resulting 66-dimensional embeddings were stable, predictive and exhibited semantic clustering similar to human mental representations. Remarkably, the dimensions underlying these embeddings were interpretable, suggesting that LLMs and multimodal LLMs develop human-like conceptual representations of objects." And https://transformer-circuits.pub/2025/attribution-graphs/biology.html It is different, hut not completely different, there are similarities that don't make it completely different "why computers can't really do what the brain does" Functionally they can already do various tasks that were previously just thought to be human things, because of those similarities Sure, there's still tons of unsolved problems And sure, whole brain emulation projects will still take a lot of time If you want 1:1 mapping I don't think we need exact replicas of the brain for intelligent machines, as we're seeing now I'm also at the same time fan of attempts at whole brain emulation and simulation that tries to do as much 1:1 mapping as possible and simulate it, on both biologically inspired hardware and digital computers But it's so far not solved enough and very impractical for AI, but maybe that will change in some years To me intelligence isn't limited to brains and biology generally https://youtu.be/6w5xr8BYV8M?si=hh-mXQiSOFT1i04s Novel systems with novel algorithms on silicon can also be intelligent to me, if they can to various degrees create models, generalize, generate novel knowledge, solve hard tasks by learning, adapt, etc. There are levels to the ability of being able to do all these things, so it's on a spectrum, and there are different types too. I guess your definition is radically human centric, maybe equating it with the equations that the brain uses. I guess we disagree there, as my definition is approximately this one, which isn't radically human centric, and can be implemented in other subtrates, like the computers we create, too, in various forms, as long as it does the things that I mentioned to define intelligence. And there's also Turing completeness that makes computers theoretically universal where they can execute any computable algorithm. But the efficiency depending on the Turing complete building blocks also matter in practice. And there's also: In computer science and quantum physics, the Church–Turing–Deutsch principle states that a universal computing device can theoretically simulate every physical process. https://en.wikipedia.org/wiki/Church–Turing–Deutsch_principle This is one of my favorite attempts at putting intelligence into an equation, Francois Chollet's definition! His motivation is primarily to get some better ways to measure it in machines. https://arxiv.org/abs/1911.01547 For him, intelligence is skill-acquisition efficiency, the efficiency with which you operationalize past information in order to deal with the future, which can be interpreted as a conversion ratio, and highlighting the concepts of scope, generalization difficulty, priors, and experience. The ability to generalize is important there. He uses algoritmic information theory. https://imgur.com/f7YuIXO And he's fan of neurosymbolic program synthesis approaches. His new presentation. https://www.youtube.com/watch?v=5QcCeSsNRks Interested in quantum gravity computers doing computation with indefinite causal structure under causaloid formalism? Quantum gravity computers: On the theory of computation with indefinite causal structure https://arxiv.org/abs/quant-ph/0701019 How would an AI working like this look like? probabilistic logical hypermetagraphs is interesting language Intelligence can be so much more grander I see creativity, out-of-distributionness, and intelligence on a spectrum, and sometimes as more granular or high-dimensional or with subtypes, and with diverse possible definitions i personaly see OOD on a spectrum nothing is 100% OOD but for example those who found quantum mechanics and general relativity IMO did some big degree of OOD leaps Myslím že krása je ještě z velký části symetrie, a komprese by mohla jít vidět jako identifikace konkrétních symetrií v datech a pomocí toho zmenšení redundance. Ale ty symetrie v kráse můžou být i víc obecný symetrie. Definitions of intelligence https://arxiv.org/abs/0706.3639 https://arxiv.org/abs/1911.01547 https://agisi.org/Defs_intelligence.html https://www.researchgate.net/publication/327275080_Getting_Clarity_by_Defining_Artificial_Intelligence-A_Survey https://en.wikipedia.org/wiki/Intelligence#Definitions What is intelligence? https://imgur.com/edP996M It's interesting that when people try to apply all these different definitions and measure them across all the different existing AI systems, how much diversity of classifications they can get from it. OOD je na spektru Mimo distribuci jsou novel scifi alien universes, ale furt jsou i v distribuci tím že používají (většinou skoro) stejnou fyziku nebo humanlike sociální pravidla. (Když poprvý vyšel Rick and Morty) Ale ještě víc mimo distribuci může být noise Čím míň podobností je mezi vzorama generated něčeho a in distribution věcí, tím víc of distribution to je How to create or automate the creation of entities that create outputs that are as out of distribution as possible, while still making sense to human pattern recognition machinery, and either giving us all sorts of interesting emotions like art, or being scientifically useful like scientific discoveries, or mathematically interesting like mathematical results, or philosophically interesting like philosophical theories? All abstractions and analogies are leaky 38:58 "plants could be doing inferencing, and have a small level of intelligence" https://youtu.be/PNYWi996Beg?si=l_ZQXi9fAlpqsX7Q&t=2338 The more we can expand the scope, diversity and scale of intelligence, the better we can understand what questions to ask about the universe, platonic realm and overall reality, to understand the nature of it all, with more predictive and explanatory power