Miles Cranmer - The Next Great Scientific Theory is Hiding Inside a Neural Network (April 3, 2024)
ฝัง
- เผยแพร่เมื่อ 4 เม.ย. 2024
- Machine learning methods such as neural networks are quickly finding uses in everything from text generation to construction cranes. Excitingly, those same tools also promise a new paradigm for scientific discovery.
In this Presidential Lecture, Miles Cranmer will outline an innovative approach that leverages neural networks in the scientific process. Rather than directly modeling data, the approach interprets neural networks trained using the data. Through training, the neural networks can capture the physics underlying the system being studied. By extracting what the neural networks have learned, scientists can improve their theories. He will also discuss the Polymathic AI initiative, a collaboration between researchers at the Flatiron Institute and scientists around the world. Polymathic AI is designed to spur scientific discovery using similar technology to that powering ChatGPT. Using Polymathic AI, scientists will be able to model a broad range of physical systems across different scales. More details: www.simonsfoundation.org/even... - วิทยาศาสตร์และเทคโนโลยี
Another banger from youtube algorithm
Yes but not for everyone, only those with the capability to appreciate this for what it is
@@JetJockey87 Edgy teenager says what?
facts
It seems like very powerful idea, when AI observes the system, then learns to predict behaviour and then the rules of this predictions are used to delivery math statement. Wish the authors the best luck
00:00-Introduction
01:00-Part I
03:06-Tradititional approach to science
04:16-Era of AI (new approach)
05:46-Data to Neural Net
13:44-Neural Net to Theory
15:45-Symbolic Regression
21:45-Rediscoverying Newton's Law of gravity
23:40-Part II
25:23-Rise of foundation model paradigm
27:28-Why does this help?
31:06-Polymathic AI
37:52-Simplicity
42:09-Takeaways
42:42-Questions
Amazing talk, and great Research!
Love the definition of simplicity, I found that to be pretty insightful.
So here we are, you guys seems to be chosen by algorithm for us to meet here. Welcome, for some reason.
It is precisely what I'm working on for some time now, very well explained in this presentation, nice work! (the idea of pySR is outrageously elegant, I absolutely love it!)
John Koza had Genetic Programming which is basically the same thing in the 90s. He made documentaries, talking about reusing learnt functions and everything, very interesting. Didn't really take off though, it just suffers from being slow like most evolutionary methods (unless you parallelise massively like OpenAI Evolution strategies) and can't learn more complex tasks that deep learning can. In another timeline it could've got more attention and maybe become better than neural nets
@@gumbo64maybe its application will be better suited for some other situations or environments or scales in the future if NNs hit some type of thing they cannot overcome quickly enough.
@@Fx_- We're currently hardcapped on current AI models with hardware but I am building a full stack system that takes advantage of currently existing hardware with implementing a daughter board to speed up the analogous computational requirements for large scale implementation. You'd be surprised at how little you need extremely large supercomputers when you scale more efficiently. Well and also leverage quantum computers for their relation with randomness.
There are multiple different awesome ideas in this presentations.
For example, an idea of having a neural net discovering new physics, or simply of being the better scientist than a human scientist. Such neural nets are on the verge of discovery or maybe in use right now.
But I think the symbolic distillation in the multidimensional space is the most intriguing to me and a subject that was worked on as long as the neural networks were here. Using a genetic algorithm but also maybe another (maybe bigger?) neural network is needed for such a symbolic distillation.
In a way, yes, the distillation is needed to speed up the inference process, but I can also imagine that the future AI (past the singularity) will not be using symbolic distillation. Simply, it will just create a better single model of reality in its network and such model will be enough to understand the reality around and to make (future) prediction of the behavior of the reality around.
We call it abstraction🎉🎉🎉🎉
And with all this advancement we don"t have fresh good water and we don"t have long term stable electricity and not enough minerals for development
@@shazzz_landthats because of the higher ups/elites not AI or technology.
@@denzelcanvasYT People don't fear AI, they fear capitalism
What an amazing fck of presentation. I mean, of course the subject and research is absolutely mind-blowing, but the presentation in itself is soooo crystal clear, I will surely aim for this kind of distilled communication, thank you!!
The folding analogy looks a lot like convolution. Also, the piecewise continuous construction of functions is used extensively in waveform composition in circuit analysis applications, though the notation is different, using multiplication by the unit step function u(t).
Oragami manifold🎉🎉🎉🎉🎉🎉🎉of course🎉🎉🎉🎉🎉🎉🎉🎉
Folding goes into compression and data theory and is the basis for the holographic universe theory.
Thought the same thing. Can do the Evaluation as a convolution of the two activation functions. Nevertheless, i guess the representation is somewhat more intuitive this way, as the middle part can be extracted as well if needed.
Thought the same! (This vid appeared in my recs after watching the 3B1B convolutions video!)
On what he's actually describing with the folding (11:10), I think it's actually pretty easy to miss, since he assumes you kind of anticipate or half-understand what he's about to say, so he goes over it pretty quickly
So for anyone who coming to this completely naive or who might have missed it the first time, like I did...
The chart (d) essentially traces out chart (c) while (b) is increasing, then traces it in reverse while (b) is decreasing, and then traces it forwards again as (b) increases again
Some people might get slightly mad at me for pointing out the obvious
Well, it IS simple, and it's easy enough to intuit why it would happen once you see it, BUT it is only obvious once you see it, and it's easy to miss in real time (at least I think!)
Amazing lecture!!
I came here to read all the insane comments, and I’m not disappointed.
We love our crackpots don’t we folks
;) The typical crackpots are here to submit their opinion and here I can't even get past half of it for how insanely hard this topic this.
Great minds are .,..,...
It's so cool when people are simply arrogant, and offer nothing to counter those ideas with which they take issue! Keep it up!
Well then, allow me. EUUUUUAAAHHHHH EUAHHHHH AAAAA SKYNET GRAY GOO!!! Omg I DON'T UNDERSTAND MATH HOW CAN YOU DO IT BY YOURSELF? Ancient aliens!!!! David Ike, D.u.m.b.s, Robert Bigelow taco bell space station!!! REEEE SCREEEEEEE.
You're welcome. Also I looove math and science and astronomy. Happy learning!
been in the rabbit hole lately so glad this popped up you rock miles!
This is a brilliant idea. I hope this goes places
This is SO cool! My first thought was just having incredible speed once the neural net is simplified down. For systems that are heavily used, this is so important
Incredible lecture
This is a very nice idea. I hope it will work! It will be very interesting to see new analytical expressions coming out of complicated phenomena.
Solving problems is the essence of the Hegelian dialectic.
Problem, reaction, solution -- The Hegelian dialectic!
Neural networks create solutions to input vectors or problems, your mind is therefore a reaction to the external world of problems!
Thesis (action) is dual to anti-thesis (reaction) creates the converging or syntropic thesis, synthesis -- the time independent Hegelian dialectic.
Concepts are dual to percepts -- the mind duality of Immanuel Kant.
Vectors (contravariant) are dual to co-vectors (covariant) -- Riemann geometry is dual.
Converting measurements or perceptions (vectors) into ideas or conceptions is a syntropic process -- teleological.
Your mind is building a "reaction space" from the input or "problem (vector) space" to create a "solution space" and this process is called problem solving or thinking (concepts) -- Hegel.
Targets, goals, or objectives are inherently teleological and problem solving is a syntropic process -- duality!
"Always two there are" -- Yoda.
Syntropy is dual to increasing entropy -- the 4th law of thermodynamics!
I was wondering or missing the concept of Meta-Learning with transformers, especially because most of these physics simulations shown are quite low-dimensional. Put a ton of physics equations into a unifying language format, treat each problem as a gradient step of a transformer, and predict on new problems. In this way, your transformer has learned on other physics problems, and infers maybe the equation/solution to your problem right away. The difference to pre-training is that these tasks or problems are shown each at a time unlike the entire distribution without specification. There has been work to this on causal graphs, and low-dimensional image data of mnist, where the token size is the limitational factor of this approach, I believe.
All i always wanted to hear is in this video ! thanks !
Very cool visual at 28:12 - where would harmonic analysis fit?
It makes intuitive sense that a cat video is better initialization than noise. It's a real measurement of the physical world
I think it is mostly the fact that, as he said, cats don't teleport or disappear, so you have some sense of structure and continuity that aligns with the PDEs you want to solve.
@@lbgstzockt8493 You're saying the same thing. "Structure and continuity" come from this measurement of the real world (it's a video of a real cat, experiencing real physics).
@@lbgstzockt8493 Sounds like you've never had a cat. Structure and continuity is not a guarantee. XD
I think this is the ultimate proof that cats are fluids, so it helped the fluid simulation.
@@fkknsikk lol
This was amazing-- confirms my suspicions.
Nice! I interpret this as, “these are the standard models - we can use them to kind of explain why AI is growing so exponentially in languages we can’t even understand, but really - we have no idea what’s going on and this is why to complex for our linear models.”
Awesome explanation
This is some ingenious work
Jesus christ, okay TH-cam I will watch this video now stop putting it in my recommendations every damn time
You can press 'Not Interested' and it should stop suggesting it.
Being able to derive gravity laws from raw data is a cool example. How sensitive is this process to bad data? For example, non-unique samples, imprecise measurements, missing data (poor choice of sample space), irrelevant data, biased data, etc). I would expect any attempt to derive new theories from raw data to have this sort of problem in spades.
That is a really good question.
This is actually really important
I would say this is not as important as the book... called "where's my cheese". Have you seen it?
The 'Avada Kedavra' potential of that pointy stick is immense. Brilliant presentation.
Read another book
There's a paper on Feature Imitating Networks that's gotten a few good applications in medical classification, and subtask induction is a similar line of thought. FINs are usually used to produce low dimensional outputs, but I was thinking about using them for generative surrogate modeling. FINs can help answer the question of how to use neural networks to discover new physics.
An idealized approach would turn every step of a coded simulator into something differentiable.
It occurs to me that the approach of this talk, and interpretability research generally, is essentially the inverse problem of trying to get neural networks to mimic arbitrary potentially nondifferentiable data workflows.
This is a great talk, laughed a lot at "literally".
Surely genetic algorithms struggle heavily with local minima. Does PySR avoid this with whatever method it uses?
I love the idea of using a foundation models approach for PDEs of different families to deal with small sample problems.
Never heard of either SR or program synthesis until this talk but both seem related to my interests, glad I watched this!
Adversarial examples for science is fucking insane and I love that guy's question.
Could you pre-train some layers (i.e. turn the standard activation functions for a few layers into pySR estimated functions) as a way to increase/change the dimensionality of the input data? Possibly could decrease the number of layers needed or time taken to train the network.
If not run early training with parallel genetically pruned custom activation layers to approach the space from different paths while trying to find the minimum loss.
This is the first exciting concept I’ve heard in the current AI revolution
Cool idea! Essentially, we can deduce symbolic, testable scientific theories from deep learning models using things like PySR. Making foundation models (which are trained on a wide variety of phenomena, not necessarily related to the area of application) for specific scientific application gives ANNs an advantage. Simplicity (explainability, legibility) comes from familiarity with a problem area, so we should be training models on lots of diverse examples to help them “get used” to solving these types of problems, even if the examples may seem irrelevant (cat videos & differential equations 🐈)
Interesting application of explainable AI 🎉 Congratulations on your research
interesting! was just learning about neural networks, so this is a pretty cool application :)
Solving problems is the essence of the Hegelian dialectic.
Problem, reaction, solution -- The Hegelian dialectic!
Neural networks create solutions to input vectors or problems, your mind is therefore a reaction to the external world of problems!
Thesis (action) is dual to anti-thesis (reaction) creates the converging or syntropic thesis, synthesis -- the time independent Hegelian dialectic.
Concepts are dual to percepts -- the mind duality of Immanuel Kant.
Vectors (contravariant) are dual to co-vectors (covariant) -- Riemann geometry is dual.
Converting measurements or perceptions (vectors) into ideas or conceptions is a syntropic process -- teleological.
Your mind is building a "reaction space" from the input or "problem (vector) space" to create a "solution space" and this process is called problem solving or thinking (concepts) -- Hegel.
Targets, goals, or objectives are inherently teleological and problem solving is a syntropic process -- duality!
"Always two there are" -- Yoda.
Syntropy is dual to increasing entropy -- the 4th law of thermodynamics!
Training LLMs on code doesn't teach them to reason a bit better, it teaches them to reason a LOT better. It makes sense if you think about it: what do you learn when you (a human being) learn to write software? You learn a new way of thinking.
That notion of pre-trained NN's discovering new mathematical operations and generalizations is so fascinating! It's so difficult to imagine there would be huge conceptual holes in our version of mathematics, but there's no reason why they couldn't exist! They're probably already there in our foundation models, just waiting to be discovered!
At 17:53, he has a plot on the right side, but he seems to attain only an expression in the variables x and y. There is no equation, so how is he even able to make a plot against those 2 variables? If you try plotting some of the given expressions by equating them to a constant (e.g. 2(x+sin(y+1.3))=3 ), you don't get anything that looks like his plot.
If there is a 3rd variable (e.g. z, or something like f(x, y)), then the plot should be a 3D plot. Instead, the plot is 2D.
it's a mistake, they're implicitly equated to 0
This is the reason why I like TH-cam
Well not sure this will go anywhere except maybe modify some of our archaic equations for nonlinear terms. The problem is probably related to NP hardness and using more expansive nonlinearity methods to crack certain problems that are more specified. We will always not know what we don't know. Using more general nonlinear models was bound to greatly improve our simulations. The real question for NN is this the MOST ACCURATE or most INSIGHTFUL and BEST of nonlinear methods to do so? Somehow I doubt this, but it's certainly a nice proof of principle and place to venture off further. To put all our faith in it might be a mistake though. We might be looking at long predicted by mathematicians limits to reductionism, and our first method to not overfit billions of parameters will give us an illusion that this is the only way, and we could be looking at a modern version of epicycles. If we want to really go further we need to use such models to not just get better at copying reality, but finding general rules that allow it's consistent creation and persistence through time. Perhaps one way to do this would be to consider physical type symmetries on weights.
RE: what you said at the end there - You're thinking of PINNs, check out Steve Brunton and Nathan Kutz
Hmm do you think resonance and harmonics might fit in here. I imagine that patterns of connections within NN/neural networks that are self-stabilizing in some way would tend to persist throughout iterations (a kind of memory). Physics gives us resonance and harmonics that describe periodic behavior in everything from atoms to predator-prey relationships to solar systems. The fourier transform essentially gives us a logic chain to describe any signal, but as some combination of periodic frequencies instead of linear lengths. It is a concept that arises again and again. Both quantum and relativistic perspectives of spacetime are highly influenced by periodic or near-periodic behavior. Maybe this is fundamental to NN as well and the cat videos taught the AI how to recognize low-dimensional periodic relationships in data. Which could explain why it helped as a preset for totally unrelated data. I'm not exactly sure if that was at all similar to what you were suggesting but it seemed related in my mind.
Half-baked thought sources:
www.quantamagazine.org/how-the-physics-of-resonance-shapes-reality-20220126/
www.sciencedirect.com/science/article/abs/pii/S0893608012002584 (machine learning with adaptive resonance)
I didn't get what is the reason to use symbolic regression. Analytical relationships/models are not the same as symbolicly representables. "Derivability" is required.
35:21 Good pretrained in some epochs by using Polymathics results does not mean training from scratch has a worse error. It is just a matter of time the good model will have the same quality.
Right, the point is energy efficiency and optimized speed/quality for multiple applications. The pretraining is done once for the foundation model, which safes efforts for the various latter applications.
I see similarity with physics informed neural network especially with Sparse identification of nonlinear dynamics (SINDy)
Briliant ideas
Fantastic. At 55 minutes though, it is suggested that we don't have a simple concept like + built into us. Perhaps not in a blank neural net, but we for example are not born with a blank slate. It is clear that any toddler understands in some way, the concept of 'more' and 'less' even though they lack empirical understanding. With sufficiently robust generalized data sets based on physical principles, information theory as language and perhaps even the nature of emotions, given enough GPUs to sustain large inter-operational neural nets, would this not give rise to something more than the sum of it's parts?
So am i the only one that going to point out that SORA from OAI is basically a generalization for a 3d engine that might let us preform experiments!
Simplicity is the absence of relative complexity.
So if you sought to get what an LLM knows out into some equations we could understand, what could they be like?
36:36 becoming more basically intelligent because of understanding spacio temporal connectivity. The flashing faces in peripheral vision illusion it shows us The monsters we create when we lack that.
Polymathic AI 🤖 is a wonderful idea 💡
Fantastic work, I thought we would take AI in this direction and here we have that reality.
As an artist using image generation models, it's become obvious that foundational models trained on very wide content perform much better in general.
It's similar to an artist drawing nudes and studying skeletons in order to draw fantasy characters better.
It's also been shown that newer foundational models that have their dataset neutered do not perform as well, even though they might be higher resolution, or generate more detail.
This is why I think it could be argued that training is transformative and falls under fair use. Unfortunately the marketing has been centered around making images that looked like other people's work (copying) which is a mistake. This has attracted people to file lawsuits against AI companies.
This could be mitigated if AI companies worked closer with actual artists in order to better understand the creative process and how that relates to presenting the technology as a tool for artists, similar to how this presentation is illustrating how to use these tools for scientists.
Great presentation. Its marvelous to see a take on AI from a broad, scientific/mathematical perspective without too much focus on technicalities. Really exited to see how this might improve or add to our understanding of the/(this?:) ) universe.
It's*
@@JorgetePanete Thank you for pointing this out. It shows that LLms are already surpassing humans (like myself) in many respects - Chat GPT makes no spelling mistakes.
Slime mold is my favorite way of imagining it
My ass smells like fish and I haven't eaten fish in a good while.
Wow!
Regarding simplicity: I think that you are missing something important about the addition operation that makes it "simple". We are also familiar with division (the arithmetic operation) and it is also useful, but we would not say that division is "simple" in the same way the addition is simple (or we would say that addition is simpler than division, even though both are "familiar" and "useful").
That is because addition is infinitely more "useful" than division. Literally any group of things, whether physical or not, coming together in some sense is addition. There are a lot of things next to each other in the universe lol. It is because it is so fundamental that it seems so "simple", because it is and they are just two different ways of saying the same thing.
@@samuelwaller4924 I was thinking of simplicity in an algorithmic sense: addition can be performed by a simple and fast parallel circuit, while division must be performed in a stepwise, linear way, where each step depends on the result of the previous one. Multiplication is similarly simpler than division, whereas subtraction exactly as simple as addition. My point is that these arithmetic operations are not "simple" or "complex" just because of our subjective experience with them, but because different operations actually have different innate properties, and it is a glaring flaw of analysis to think otherwise.
No feasible for UHDLSS Feature Selection.
it's seen it before so it pattern matches and i think it will be useful to scale up and pattern match and have other people pattern match and we can train it generally and scale up and train it generally and fluid simulation and we found matching outperform train it on more data and it does better
the title of this video overdelivers
model mining? brain digging? fascinating, i guess we gonna need some tools to uncover these gems from the nural nets or do we need to build the nets/models in a more comprehensive way?
I am re-reading once again the book By David Foster Wallace History of Infinity. There he describes the book by Bacon Novum Organum. In book one there is an apt statement that I would like to paste
8. Even the effects already discovered are due to chance and experiment, rather than to the sciences. For our present sciences are nothing more than peculiar arrangements of matters already discovered, and not methods for discovery, or plans for new operations.
Your Analytic Distillation sounds like an algorithm for Low-Rank Adaptation (LORA). Considering also semantic relationships in latent space (e.g. the vector pointing from Woman to Man added to Queen returns King), I speculate there may be something like a basis/spanning set approximation we could come up with for any arbitrary concept. Like, what if we consider lots of things we consider "good" and "evil" and try to analytically model that? Would it give us insights into morality?
Yeah but to be more precise, it would reflect our 'view' on morality. Language is a model used to simply convey the often repeated patterns of our super complicated psychological mess. I think of it as a some sort of a symbolic model. So, studying a model on Language, which itself is a model of our perception, would teach us about our perception of morality.
Since there is no such thing as intrinsically good or evil (it's defined by social&instinctive rule) it is greatly affected by the culture of the time. So doing an comparative analysis on same word vectors extracted from different times would probably show some interesting results. Such as training models on text data from World War 2, medieval time, hippi movement Era, etc. Then we would be able to quantitatively compare the moral culture of each eras.
Specific density squared ; Volume Quebed ; Vice Versa und AUgmentation Cycle
Flux TEMP composite material Augmentation Cycle und NEURO CELLULAR GEN_REGEN CYCLE @ NEUROPLASTICITY U V STABILIZER 7:50
I'm going to state the obvious. That is smart. Yes it draws questions about AI explainability regarding deep learning NNs but what this chap is saying is quite brilliant. For me, as long as the conventional approach is combined with the model he is propounding, there should be some excellent science out of that. Then there can be even more science when we start to understand the reasons and mechanisms by which the deep learning neural networks some humans build are doing and are capable of what they are so. Let's not miss the point of what he is saying, at least what I interpret that he is saying... The NN is finding some order through patterns, it really is those patterns that are probably most related to something interesting, ie of scientific interest, then we can sift through the rest of the noise to see if something was missed, let's say we do that if questions are presented that don't have an answer. So all in all, it is a very powerful way of cutting through the fluff. If we then want to scientifically describe the fluff itself, it is now more distinct. I think what this guy is saying is brilliant. Incidentally, I think we ultimately find out that deep learning neural networks come to sensible decisions because the have the fidelity to tap into the innate intelligence structure of reality itself, but that is a next topic, although entirely pertinent.
someone so smart, only listenable at 4x
You might want to think about simplicity in terms of Kolmogorov complexity e.g your NN should try to emit the least complex, in the Kolmogorov sense, syntax tree.
Also, I think "+" is simple because it is closed over the field of integers. I think that if your operation takes you from one domain to another its more complicated. In that way you might consider using Category Theory. You could think about penalizing models that "move' further away into other mathematical spaces from a 'base" space.
Kolmogorov complexity can be thought of the ideal “lower bound” for a compressor/predictor in unsupervised learning.
But it’s also uncomputable which would make it hard to implement in practice 😅
@@user-hy6cp6xp9f true, I think I was trying to get at a weighting of symbols used. I’m not sure if that could be learned or would have to be assumed.
I think 1+1 is simple because is in some ways assumed ( forgetting Russell) whereas something difficult like say the Kullback-Liebler Divergence is defined in terms of simpler primitives
Edit: big picture would be you need some sort of error term to trade off against accuracy otherwise your tree grows without bound either in depth or complexity of the operators Consider it something like dropout or pruning.
@@DensityMatrix1 Yeah that's interesting! I feel like any theory with a sufficiently complex symbolic representation could be factored into smaller bits that could themselves be learned as features.
It's a big search problem, so I guess it's about allowing the algorithm to search deeply + generate complicated symbolic representations, but having it bias towards shorter ones (since they're more likely to be true).
Honestly a big problem I have no idea how to solve.
Solomonoff induction isn't tractable for beings with finite compute and AFAIK there's no standout best approximation to it. Myopic piecemeal modeling is probably better in many cases than trying for a theory of everything.
I would love to see some breakthrough in Dark Matter regime. There is so much data regarding Dark Matter yet no theory to back it up.
30 mins to say that you can fit simpler models to a neural network data-generating process, and another 30 to say that more training data (even if relegated to what we call “pretraining”) improves performance.
ps: things are simple because they are ubiquitous and they are ubiquitous because it’s how the world works (law of conservation of mass and energy, i.e. addition), not because it’s “useful”
With enough inputs you can make any curve or field match the current data. So it this even science ? I am very skeptical.
It will provide very little real insight, when you have inscrutable AI model able to predi t something. It might as well be the oracle of Delphi.
Add some thermodynamic constraints?
The issue is discovering the higher-ordering principle which subsumes a continuum of self singularities and discontinuities. Linear math works well in-between the singularities, but cannot extrapolate through them, in a sense they are like mathematical worm-holes. Attempts to linearize across the discontinuities will fail. A whole harmonically-related series will only be properly understood from the perspective of a higher-ordering principle, similar to the idea of projection from a higher magnitude to a lower dimensional space, or from the idea of negative curvature. The point is the epistemological assumption of a static model is problematic, the real world has static islands which are bounded within areas of great change, and so the basic function changes completely there, that is to say, the dynamics of change themselves change. So to bridge that gap you can’t just ignore it, or flatten it, you have to seek how to remap it in such a manner that it is no longer infinite, but cyclical, as Gauss did with the complex number domain.
Yeah, I read that like 5 times and have no idea what you’re trying to say.
Yes AI is definitely faster generating random ideas, and is also quicker fitting these random ideas to a data set. It’s a very powerful tool.
35:16 Yes, doing the model from scratch with traditional machine learning is worse compared to the pre-trained generative network, but only for the *same time frame*, if you give the traditional machine learning approach more *time*, then it can out-perform the pre-trained generative network, while the pre-trained network will just keep on spitting out the same type of results.
a proper comparison would require a 3 dimensional chart comparing model error vs #samples AND training time+network evaluation time.
The better approach is to use the pre-trained generative network to bootstrap samples for the genetic programming("Scratch-AViT-B") model thus getting the best of both.
The "folding analogy" is incorrect. That is not how composition works. It works only in this case because of the very specific nature of the "first layer"(in his example).
Indeed.
Can you tell me more about what is incorrect?
This is very trival knowledge if one has an open mind, but its great that it is now formally been empirically proven for those out there that need proofs.
I can't shake the feeling that someone is going to train an AI model on a range of differently scaled phenomena (quantum mechanics, atomic physics, fluid dynamics, macro gravity / chemical / physical dynamics) and accidentally find an aligned theory of everything, and they'll only end up finding it because they noticed some weird behavior in the network while looking for something else.
Truly, "the greatest discoveries are typically denoted not by 'Eureka' but by 'Hm, that's funny...' "
The problem is thinking about these things as if the universe is distinguishing between scales. Any true "theory of everything" will by definition be scale invariant and the structures we see at different scales will be a natural result of the fundamental phenomenon at that level.
We don't discuss that human beings very rarely exist entirely independently. If there is a human being in a place, there is an assumption that they had parents, were raised to maturity/independence, and that must have occurred in a finite time period. These are such basic assumptions that no one would believe someone who claims they came into being fully formed and were an independent creation by a God or randomness. We cannot know what the original person or primordial ooze came to be simply by looking at our current local environment.
Just like the guy who finds a severe vulnerability in linux ecosystems, accidentally by just benchmarking a database. And shits, that happened recently lol
Someone watched pi..
Fine-tune an LLM to interpret neural nets. Iterate and maybe symbolic regression (i.e. language) will help us supercharge LLM training. But hallucinations could be a major issue...
I already did that in February when I trained ChatGPT on quantum punctuation markers and de-markers.
Anthropic did this for GPT2
I tried this with my gynoid, but she she kicked me in the nuts.
Is this folding similar to convolution?
No, convolution layers, like 2d, take inputs and further extrapolate features by applying specific linear kernel methods (specific for 2d space or 3d), this seems to be doing something different where it is not a layer, but instead applies different layers together, by folding it over. Tbh don’t understand folding, but convolution layers are common in image problems so they are easier to understand
Serious questions here, isn't his "folding analogy" just superposition of waves? Or I am missing something?
broadly useful algorithms across different systems = mathematics
Did you see the Lifestyle Trader ad? Proof that money is not just a commodity but logarithmic.
the empirical fit part was a bit of a thinker, huh
I don't know about all the fancy stuff but as a programmer this makes me 30 to 50% more productive and my daughter, who is a manager, makes her about 10 to 15% more productive.
Great
Reminds me of a sociology paper with tons of seemingly complex math that, in the end, says something like, "school bullying is exacerbated when it goes unaddressed." So what was all the math for? Credibility.
one might reason out the implications of what he said here without him having to also provide the vision for how his work might be applied. or give it to a gpt and let it do it for you
My point being, he's no philosopher, but he's demonstrating something profound beyond his ability to express it
38:18 Why is “+” simple? Well, maybe because it is closely related to the concept of counting and “having” things. If I have two apples and I get (add) three more, I can just count the number of apples to verify that I now have five apples. That has got nothing to do with the concept of simplicity. Not sure if I even want to continue watching the whole thing…
Interesting! How could neural network be an empirical finding? It’s not sth tangible that we can see or touch. It’s hard to believe that AI is developing the way fluid dynamics did.
This has crossed my mind and this is exciting indeed. High dimensionality patterns are often hidden but the fact that they are high dimension makes for the discovery of robust natural laws. We are in need of territory. We no no longer have to rely on empirical, philosophical or mathematical models to create natural laws. Data in high dimensionality can reveal many laws. Exciting times!
Could these models apply compression to themselves through techniques like quantization, pruning, and knowledge distillation becoming faster and faster and smaller until AGI emerges from a phone sized device which can invent warp drive?
Tbh all of reality can be encoded on one gigantic vector.
Probably not. There are hard limits on stuff like that.
There's a new thing in the market called laser pointer
Too little research into optimization and "understanding". We should be able to determine optimal compressed hierarchy. Hypothetically, all knowledge might be first compressed and divided into discrete tokens, say for example nouns, verbs, causation/temporality, and description.
Am I confused? It feels like he is explaining the calculus of variation and linear algebra.
The elemental functional priors he seems to be talking about are literally the concepts of functions and groups of related functions existing in hierarchial topics like trigonometry grouping sine, cosine, tangent together because they are mutually dependent and reduce the parameter space.
Students may ask why we learn both sine and cosine when we could just learn one and use a parameterized offset for the other. The synthesis in seeing how together they can convert a two positions into single time parameter given a fixed length and a pivot point. Similarly, an ellipse can be described by these same two equations with a single parameter, t for position along the curve and the axis lengths. These are all model building concepts from statistics though.
Am I missing something? It feels like he is explaining statistical model building. Yes, parsimony is great and admirable in a model. The push for larger and larger model is simply brute forcing and filling out the solution space with so many variables that it would be difficult for an answer to not exist if the idea previously existed in the world. However, they suck at low context situations where they need to make deductive leaps. If I'm talking about fear of a need to "abort", whether the conversation is happening at Kennedy Space Center or in a medical examination room completely change what we are talking about. If I don't tell ChatGPT the context, it may suggest language talking about "T-minus" for one contexts or "weeks" in another. At some level we are simply talking about different methods of representing temporal, spatial, social, economic, etc relationships and how abstracted from the ideas of initiating, terminating, increasing, decreasing, linear, exponential, repetition, regular, irregular, stochastic, or predictable. Whether one uses the term "sine" or "wave-like" or "repeating" is all just representation of the same linguistic concept
For "+," I do think it is simple because I hypothesize that the human brain does have built-in neurons specifically for counting small numbers (usually 5-9 varying between persons), so when you are an infant, you don't actually need to learn to count objects under this number (I suspect that in certain area of the brain, likely hippocampus, there are this amount of special neurons that are served as synaptic placeholders for the visual cortex in object identification. Then, it serves as the starting point to further learn the abstract concept of "+." That is also why "+" is the first mathematical operation that most humans (if not all) learned. If nothing is built-in, I wonder if someone can teach a human multiplication without them knowing addition. This experiment would be highly unethical, tho.
This is already well known. It's called 'subitizing'. I believe the research showed that subitizing is not implemented in separable neural substructures.
Thoughts from my deep ignorance
Regarding the idea that " +" might be assumed ( in replies) to be the first mathematical operation of human behavior. I wonder what would be different if looking at this from my perspective
"What if “ - “ is actually the first mathematical operation? What if the second operation, the “+ “ is the process of filling in the vacuum caused by the first “ - “ ? The first loss of coherence.. as an identifiable cellular membrane (ovum) being fully formed and then losing that coherence by the separation of the membrane experiences as a gap forced by penetration of new foreign material (sperm) that then becomes assimilated, exchanged. Not either or, + or - but shared - is part of + . And always was.
The proof is trivial! Just view the problem as an associative topological space whose elements are fundamental varieties.
Great idea. By the way he sounds like a science-oriented version of Mark Ruffalo
How do you avoid p-hacking your data?
Stop using p
@@whatisrokosbasilisk80 so stop using statistics
@@joelwillis2043 If you think that all of statistics boils down to p-value, you don't know statistics.
Feynman would've loved this age!
He already hates it when n is NOT equal to 3. He would despise them SO HARD, goverment would declare him domestic terrorist.
my man just reinvented the wheel with already existing meta and unsupervised learning. good luck ig
Kalle's Ninth Proof of Folding is here.
What kind of magical language they speak?
4:35
Era of AI
presenter uses stick to point on canvas
Dear TH-cam algorithm,
Please send me more like this
With love
-O