> You don't hear a lot of talk about emergent chemical phenomena.
What? All of chemistry is basically emergent. we do not for example have clean, reducible models of the details of why aufbau diagrams fill the way they do in the transition metals, it's very hand-wavey and extremely empirical. Why does iron freely access the +3 and +2 states in ferrocene? Like why are titanium [-2] and [+1] states even a thing at all. Why does zirconocene have just enough space to do an p-pi sigma bond in ziegler natta catalysis? What the fuck even with p-pi sigma bonds in the first place?
> chemistry is actually physics
You cannot throw a physicist at chemistry and expect insight or even understanding to come easily. The defining mentality and approach are just so different. So much so that probably the methodological approach should be a better definition of what constitutes chemistry vs physics.
It's always taught that oxidation states are essentially a book-keeping strategy, and so can lend to bizzare bottom lines in complex systems wherein assumptions of clean charge attribution are invalid or less valid. However, I think I get what you are trying to say: you cannot expect a molecular biologist enter an OT and perform a coronary bypass.
You know more about chemistry than me, but when I studied emergence it was considered a more transparent layer than, say, psychology. It was considered at least something worth questioning as recently as 2022. [1]
Anyway, no one outside of your specialty knows what you're talking about and so that is congruent with my "you don't hear much about" which I think you read as "does not exist." I don't know if there are emergent chemical phenomena, just that they are thrown around to mere mortals like me the way evolution or social behavior are. Or life itself.
This isn't about physicists and chemists, obviously or their praxis. It's about what is actually happening and if you can explain it solely using the tools of the underlying layer.
You have succeeded in out esotericking me, but I don't think anything I stated was incorrect.
How are you defining emergence? Because I've always thought of it as simple rules that compose creating completely counter-intuitive results, and man is there a lot of that in chemistry.
TBF that's because those rules are merely approximations of the actual rules and undergraduate chemistry essentially consists of learning a series of progressively less wrong theories over the course of several years. Emergence flows from real rules which in chemistry would seem to be QM. But we can't actually fully simulate (as opposed to grossly approximate) a QM system beyond two (or perhaps three?) atoms so ... yeah.
> But we can't actually fully simulate (as opposed to grossly approximate) a QM system beyond two (or perhaps three?) atoms so ... yeah.
Iirc only hydrogen has a satisfactory QM solution, so... More like two subatomic particles is the limit
> TBF that's because those rules are merely approximations
Not just. Often the rules you learn (until pushing arrows in ochem) are rules governing ground state which is not usually that interesting, because those are often quite stable molecules, and interesting things happen when you LEAVE the ground state.
> iirc only hydrogen has a satisfactory QM solution,
But we can in theory iteratively approximate to an effectively arbitrary level of precision. The issue being that in practice the algorithm is O(scary) leaving anything past a lone helium atom unrealistic.
I'd argue that the rules are needed though, because at some point we have to partition reality and say "x works this way, but y works more like this." So there's definitely less vs more accurate models, but the more accurate ones can have unnecessary mental overhead that makes it difficult to keep track of other details. I'm currently studying chemistry right now, and I've come to appreciate the layering of models happening (lewis dot diagrams for non-important details, pi bond delocalization for a planar molecule, band gap for light frequency, etc).
Also QM simulation is not completely hopeless! I'm actually doing computational chemistry research, and we're able to do reasonable approximations for around 70 atoms with a method called Density Functional Theory (DFT). And recent developments with machine learning have further given rise to Machine Learning Interatomic Potentials (MLIPs) (not to be confused with something like AlphaFold). Essentially it uses a GNN to predict the forces on all the atoms based on the current atomic arrangement, and then the simulation will advance through time by integrating (e.g. adding) the forces.
Agreed on all counts, but also I don't think you're actually refuting anything I said?
Certainly the various theoretical frameworks are useful. I was merely pointing out that they aren't the relevant ruleset when it comes to discussing emergent phenomena.
The approximations you listed while interesting and impressive nonetheless remain approximations. I haven't kept up with the state of the art but the computational complexity of full QM is utterly absurd. Normally that wouldn't be of any particular interest but it becomes relevant since when we talk about emergence we typically refer to simple and easily applied rulesets such as the various well known cellular automata.
> Agreed on all counts, but also I don't think you're actually refuting anything I said?
Fair enough. I guess when you said "progressively less wrong theories" I was thinking that the wrong theories are still quite useful in the right contexts, but I can see that you feel the same :)
> But the computational complexity of full QM is utterly absurd.
It's always a good sign when your naive algorithm scales 6^N /s. Not exactly simple in the sense of cellular automata like you mentioned, lol. But yeah, it's kind of insane. I'm hoping that we can train MLIP models for specific cases though, like could we fine-tune an MLIP to handle biochemical reactions at normal temperatures? It won't account for more exotic chemical reactions, but it would still open the door to model enzymes and receptors better. I'm hoping to go into research for QC methods because of all the cool opportunities it would open up (though man I have a long ways to go to even understand the math behind it).
Sometimes that's a pedagogy issue. The 18-electron rule is a good heuristic for the ground state, but literally all of inorganic and organometallic chemistry is 17- and 19- electron systems, because interesting stuff happens when you're not at the ground state. Took me halfway through inorganic to realize that.
Alternatively, as an explosion of complexity required to succinctly describe and/or predict consequences in a layer above a comparatively simple one. A classic example being Conway’s game of life.
My examples are basic things they teach you at undergraduate level (I'm not an inorganic chemist, my examples are inorganic/organo metallic examples, last I saw them was an undergraduate).
Probably not but if they granted buyers with no income 0% interest for 20 years then yeah? Sales would grow through the roof. iow it can be essentially fraudulent, depending on the quality of the loans.
I had the llms write an incredibly complex CV pipeline in zig through several iterations a and failed ideas and the only time it segfaulted was when I accidentally compiled it with avx intructions and deployed on a non-AVX machine
But what made "better not leak" challenging? I ask because it looked like you were positioning "incredibly complex CV pipeline in Zig" as a counterpoint to my claim that LLMs are bad at memory management. I still think they're bad at it, but I'm interested in evidence to the contrary, so I'd love to know why your application rises to the level of something I should expect an LLM to struggle with. I've seen (have in prod right now actually) relatively simple ideas not much more complicated than resetting an arena each request, which would paper over a lot of memory-management sins even with quite bad models. I'm curious if that's why the LLM did well or if there's something else going on.
I think zig would do to have "custom binary operators" which is not operator overloading, but basically give people what they want. It would suffice to be syntactic sugar elaborated by the parser. For example:
const @<+> = @import("op.zig").special_add;
const c = (a <+> b);
//syntactic sugar for @<+>(a, b)
Maybe it will add that. But just how important that is, though, I'm not so sure. Languages like Java, C, and Go don't currently offer custom infix operators, and they are more popular than many languages that do.
Note that there are some complexities, though, to infix operators regarding the order of operations (e.g. + vs * in ordinary mathematical notation) and associativity (e.g. is `(x <+> y) <+> z` equal to `x <+> (y <+> z)`?). They can be dealt with, of course, but not without adding complexity that may or may not be worth it.
Don't diss linked lists. Sometimes they are the best data structure. In my language's interpreter, I maintain "contextual layers" and the best data structure for this is a linked list (two different layers might be applied to the same previous layers)
reply