Shower thought: Someone with no knowledge of programming but uses an LLM is a D&D sorcerer, while someone that has the understanding of CS would be the wizard class. In D&D rules, sorcerers can cast "more magic" per day than wizards as they are unencumbered with any thoughts about the technical details. Wizards can cast a vast range of complex customized spells that the sorcerer is unable to perform, and thus are able to maneuver far more complex situations.
And their patron is capricious and doesn't seem to actually understand anything that the warlock wants. Less monkey paw, more "I wasn't listening, here, have X because I think I heard you say Y"
I like it! For LLM creators that self-host their modal, they wouldn't have to worry about their patron leaving them. However, the patron only grows in power based on the wishes of other agents other than the warlock. Wizards, in theory, have unlimited power of their hardware environment [cough except DRM].
I) DEFINITION.
Magick is the Science and Art of causing Change to occur in conformity with Will.
(Illustration: It is my Will to inform the World of certain facts within my knowledge. I therefore take “magickal weapons”, pen, ink, and paper; I write “incantations”—these sentences—in the “magickal language” ie, that which is understood by the people I wish to instruct; I call forth “spirits”, such as printers, publishers, booksellers and so forth and constrain them to convey my message to those people. The composition and distribution of this book is thus an act of Magick by which I cause Changes to take place in conformity with my Will.)
In one sense Magick may be defined as the name given to Science by the vulgar.
- Aleister Crowley, "Magick in Theory and Practice."
The "magickal weapons" of the 21st century are technology, engineering, and computing; a newly emerging "magickal language" is the LLM prompt; we are building new machine "spirits" in the form of ChatGPT, LLaMa, and friends.
> it's worth considering Magick from the psychological perspective
That's right. Israel Regardie ("The Middle Pillar," "The One Year Manual") has more to say on viewing magick and Crowley's work through a psychological, Jungian lens, instead of a metaphysical, or even anti-scientific one.
Many things which are often quickly dismissed tend to have rich and interesting symbolic interpretations.
One example is a kind of "ghost". For example, if someone passes away and you miss them deeply it's possible for you to be "haunted" by their ghost / memory. And how do you get rid of the ghost? By using salt, i.e. crying. (Obviously this example is meant to be illustrative and basic.)
Jung has some really interesting interpretations of symbols but he can sometimes go a bit overboard. I recently read Aion and while the first few chapters were intellectually interesting, he absolutely lost me when he went on an 8(?) chapter detour rambling about opaque fish symbolism. Admittedly, maybe I'm just a bad reader and was unable to fully grasp the full intricacies of his points.
This might sound like a quibble, but details matter in this sort of thing. Both sorcery and theology are sciences as traditionally understood. Perhaps you meant "anti-pure-materialist?"
After all, there are notable and influential scientists who very much believed in the supernatural. Now maybe some people get a dopamine rush from thinking "Lol Newton wuz dumb," but on the other hand maybe the world's greatest minds weren't actually stupid.
Jacques Lacan's "Real, Symbolic, and Imaginary" categories come to mind.
Not everything we experience has a material basis (eg: emotions, inner dialogue, expectations, etc) but can be "anchored" in our minds through symbols. Those symbols can become a kind of vocabulary with which we can communicate with the subconscious (dreams, intuitions, what we notice/don't notice, ritual, etc) to change what and how we perceive.
Crowley's and Israel Regardie's approach to this field is worth looking into, even though there may be less esoteric ways to acquire the knowledge they shared.
Personally I think Peirce's semiotic was digging in the right place, but sadly even his genius intellect wasn't able to bring those conceptions to unity in his allotted time.
Exactly - cultivating one's ability to self-reflect, and increasing the critical distance between oneself and one's thoughts, to the extent that one can experiment with changing how one thinks in order to observe what one thinks as a result - that is the project.
Working in the realm of the abstract, writing software, is a kind of playground that allows for this kind of experimentation.
Paradigm shifts in system architecture and in how we conceive of daily life are in many ways alike.
Recognizing that we have (even if limited) some capacity to adjust to our liking how we think, so we can think what we want (and feel how we want to feel, do as we want to do, etc) is magical and rewarding. To understand that it is only nihilistic fatalism and helplessness of blaming circumstance or others for what - in many cases - is our own shortcoming (with regard the above notes) is actually empowering and inspiring on so many levels.
I think the source of LLM magic is not the compute, but instead written human knowledge. The LLM special sauce--what makes an LLM distinct from a compiler or a web server or a linter--is training on human generated input data. In fantasy novels a common trope is that humans generate magic (like a tree generates oxygen) which diffuses into the world to be used fuel for spells. Similarly, written knowledge is created by all of us in the course of our day to day lives, and is omnipresent in modern society. It fits!
> In fantasy novels a common trope is that humans generate magic (like a tree generates oxygen) which diffuses into the world to be used fuel for spells
Is it? I have read a lot of fantasy and have never come across this. Magic is instead always sourced from some supernatural/higher power that humans struggle to understand. If you have some examples I'd love to hear them, so I can branch out.
Depends on the fantasy world. Mana or MP for example is internal.
In Chinese Wuxia mythology there's a concept of Chi. In Naruto (which isn't Wuxia but borrows a lot) it appears as chakra.
In D&D it's the weave which is this higher power.
But this digresses from the point. I think Magic is a bad analogy. When I looked into ML I was surprised to find how trivial it is.
Pretty much all of intelligence can be simplified to this single endeavor of curve fitting. LLMs, GANs, deep learning... all of it are just different ways to find a best fit curve.
What I found is that this knowledge (which is trivial to obtain) reduced the magic. Intelligence to me was no longer magical. Understanding intelligence made it less magical and more mundane.
A better analogy would be those stories about characters walking into this epic void of lovecraftian incomprehensibility expecting only madness and more questions to arise from their journey that spans eons... only to walk back in 15 minutes with the realization that the answer is just 42.
> (It used to be that making a drainage system under a tower required consultations of magicians and terminology used was including dragons.)
It still does. I'm part of a hobby community that builds basically every part of houses from the ground up, but the one thing we never touch is plumbing. A lot of people aren't too keen on electrics which I guess is fair enough considering the dangers of getting it wrong, but the real dragon is water.
It sounds like a reference to “Merlin and the dragons” - I don’t know whether it’s canon or apocryphal, but young Merlin (then called Emrys) has his village invaded by a warlord, who intends builds a battle tower on a nearby mountain. Uther Pendragon eventually shows up to save the day.
I remember my mom reading it to me when I was a kid - and the story itself was framed as Merlin telling Arthur himself a bedtime story when he was a newly-crowned king.
From my rather vague memories of Geoffrey of Monmouth - Vortigern, the ruler of a part of Britain that he probably claimed was all of Britain had a good deal of problems with the Saxons, whom he had made some disadvantageous deals with.
His soothsayers said he find a young man fathered by a demon (incubus) on a virgin and soak the foundations in his blood.
So Vortigern found Merlin but Merlin was so cool he talked everyone into thinking he knew better than the soothsayers and he had them drain the foundations and there were two dragons fighting there at which point Merlin told some prophecies the gist of which I do not recall, but which caused him not to get killed and Vortigern to leave the tower and get killed shortly thereafter.
This story was basically used with embellishments making it logical and realistic in Mary Stewart's Merlin Books - I think the second one the name of which escapes me but plus 1, they are a great read if you like Merlin stuff.
Oh no that’s it, absolutely, that’s the story! I think The dragons’ colours represent Vortigern and Uther Pendragon, and when they fight, the one representing Vortigern loses, and Merlin informs him that it portends Vortigern’s impending loss against Uther in battle.
A goal of mine that I currently lack the time or energy to execute on is a simple open world game with most of the game world rules exposed in a Forth-like API/programming language.
Then take some runes/glyphs/circles/shapes and train an NN to translate them into programs using that Forth-like. I think it would make a game world with a significant amount of wonder in the magic. And a bit of unknowability as I'd probably add a bunch of extra features (astrological cycles, environment effects, etc) that slightly affect the spells.
Why Forth, of all available languages? Unless you're riffing on that really weird series of YA fantasy where the guy falls into a world full of demons and starts converting magic into programs because it makes debugging the universe easier. Think that involved Forth. Still not sure why.
Fun book, but no. And I wouldn't describe those as YA. YA doesn't typically rant about ANSI C conferences. It's more like Discworld in that it's quite silly but not necessarily targeted at youth.
My reason for Forth is quite simple: it has extremely little structure which makes it a much much easier output target to work with.
At it's simplest, a series of glyphs could directly transliterate into Forth words. More complicated schemes are more interesting, for this purpose, but it would defeat the purpose of its use to target, say, C.
Also, Forth is easily embeddable and interpretable with only the words I want exposed. Whipping C again, but targeting C would mean a whole mess of sandboxing and C bindings which sounds much harder than Forth words or some such.
I have many other ideas too. Multiple rings of circles represent threads or some kind of concurrency, mana is used like 'gas' in the EVM to naturally denote the cost of spells and prevent abuse, possibly compile from Forth to some WASM target or something for performance, etc.
Edit: put another way, Forth-likes blur the lines between a programming language and a mutable environment of high level bytecodes which makes it a natural intermediary language for some cases.
Try Reader Mode on Firefox. The article looks great, and renders how I choose — dark mode, font, etc. It’s a godsend when it works (almost always). It’s been an unfortunate necessity for my modern web experience.
This is a hilarious and fun take on ChatGTP. I can imagine that at some point we have deployed so many of these technologies that we might start living in a Terry Pratchett novel.
I recently had a thought that when a meaningful piece of text/image is written/uploaded on the internet, the energy or willpower if you will spent in producing it by a human reduces entropy a bit. Up until now that entropy reduction was unmeasured and unharvested, because really who would browse through 170000 forum pages and combine the bits and pieces to produce an answer on how to do a niche little thing, but by these LLMs being trained on this data and commoditizing it, the "magic" if you will is really harvesting and repackaging informational entropy reduction.
The real magic will be when we can get to the point where the objective function is “return the answer to the prompt that is optimally satisfying for the specific person asking”.
I think the real magic will be infusing these natural language systems with deterministic algorithms that have been developed in CS, etc. Imagine doing Dynamic Programming and Kalmal Filter in the domain of thoughts and ideas. A human being only thinks 3-4 recursion deep but these systems will be truly magical. They can very easily write books like "Reasons and Persons" by Derek Parfit
DeepMind has https://arxiv.org/abs/2309.03409 so as long as you can specify an objective scoring function you can get closer than manual prompt engineering.
You can get pretty close to that now with a) clever system prompt engineering to incorporate user behavioral and personality metadata and b) getting said user behavioral and personality metadata, which is the harder problem.
