◡◶▿ SOFT03 | Blank screens
🖥️ Using ChatGPT to write a script "for the funders." Plus: Béla Tarr workshop videos emerge from Fukushima. Sarajevo. Lem. God. The Trashmen. | Imaginary Software of the Filmmaking Future Week 03
Hello! Good morning and welcome, film students.
Let’s crack on with our guide to re-conceptualising filmmaking - or doubling down on your core beliefs! - in the age of auto-dial art.
First, please think back to last week’s lesson. We covered how:
When radically retooling, the filmmaker must identify what she wants from the future(s).
Filmmaking is not a futurist art, so the filmmaker should guide it into the future she desires.
The development of future tools and techniques requires delicate attention to today’s habits.
Tomorrow’s filmmaking might be something we can’t yet imagine - or it might be right under our collective (or individual) nose!
Missed a week? Joined late? Don’t worry about reading these lessons out of order. Each functions independently. They are sent in a sensible sequence but hardly reliant on it.
Today, we will look at some places to start with the software. The screen is on. The cursor is blinking. Here we go. We’ll cover:
🧵 How your generative software practice might begin with subtle acts of repair.
🧩 What to say when your artificial intelligence (AI) sparring partner was trained dumbly on a vast body of wrong words.
🆒 ‘Textual interaction’ as a hazy playground for wordsmiths and provokers.
🗑️ “Papa ooma mow mow, papa ooma mow mow.” IFYKYK.
I’ll also point out an opportunity to access some rare Béla Tarr workshop footage. And mention an excellent international (or, for Bosnians, national) opportunity for up-and-coming filmmakers from Southeast Europe. And there’s some stuff about ancient, lava-caked scrolls! At the end of class, below.
Please do your film school (me) a favour by sharing this post (or one from our growing archive) with a colleague or network who’ll appreciate it.
Dots and threads
You can hear me deliver this lesson by scrolling up to the header and clicking the play ▸ button.
It starts with dots and threads.
A stitch here, a patch there. Adding some stars, removing some hairs.
Mending your movie. Brightening the picture “smartly.” Swapping the sky or changing the light scheme; negotiating with sassy shadows. Asking your robot to erase a carton of juice drink that belongs to quite another period. Or perhaps to replace it with a better one.
Not creating, just mending - as though mending is not creative! As though mending is not the greatest unconscious creative act of all!
But at what point does mending become modification? At what point does modifying become generating? And generating, conceiving?
Will you draw a line between these processes? Or just dive in, your needles clanking furiously?
His master’s voice
“In the beginning was the Word,” wrote John1, at some point in the first century AD. “Well, everybody knows that the bird is the word!… Papa ooma mow mow, papa ooma mow mow,” sang The Trashmen, two millennia later.2
Both assertions are among the vast reams of written training material on which today’s so-called Large Learning Models are built. Both bits of writing are threaded right into the AI’s mind!
So, when you use AI software to generate a script, a motion clip, or a cure for your cinematographer’s cataracts, the AI must refer to both of these texts. Refer to these texts (among countless others) to understand your proposition and formulate a response.
Gosh!
It appears that John meant that God founded ‘his creation’3 on a rational basis. The “word,” for John, is a synonym for logic and intention.
The words and form of The Trashmen’s text appear to disprove John’s suggestion.
And yet, both texts appeal to people on both a visceral and literary level. What does this mean for the filmmaker in the age of generative AI?
Here’s a third text: Stanisław Lem’s novel, His Master’s Voice (Głos Pana). Or, more particularly, the two volumes of apparently random numbers translated from a mysterious space signal, around which Lem’s plot is organized. (He doesn’t include the full volumes in his book.)
In Lem’s novel, a man notices that the numbers in the volumes are not, in fact, random. They contain repeating patterns.
The very fact of discovering such patterns in a space signal has huge implications. And just as “Papa ooma mow mow, papa ooma mow mow” might translate as a spell to make humans (and perhaps birds) dance and, inevitably, mate, the numbers reveal a recipe for creating a simple alien life form and potentially powerful weapon. A whole new set of implications.
But they were just numbers! Just words! It was the grammar that was grotesque. The syntax, the melody, the rhythm, that animated them.
Something for the filmmaker to consider. To consider as she processes her own Word or words through those vast and incomprehensible engines. Engines of language that promise her such power. Power that she activates by reaching blindly into that network and connecting one word to the next.
Humpte Dumpte
“[T]he elaboration of artificial literary structures would seem to be infinitely less complicated and less difficult than the creation of life,” wrote François Le Lionnais in his second Oulipo manifesto in 1973.
Well, yes. And his literary friends proved it. Inventing absurd frameworks and algorithms for writing and using them to shape warmly human texts.
But somehow, AI text generators struggle to follow these artificial structures. Struggle despite their apparent suitability for the job.
The text machines can present the filmmaker, quite proudly, with the second line of Humpty Dumpty if she feeds them nothing but the first. But they will just as proudly offer her a screenplay sprinkled with letter e’s even if she was quite clear that the letter e was the one structural element she didn’t want them to use. For example.
And so it is with any structure she feeds it. However ‘artificial’ or organic her concept. If the filmmaker wants the software to produce a screenplay or a shot or a prop, she must accept that it plays by rules that it considers, for all practical purposes, optional.
And so, the work the software produces is a better or worse impersonation of what it pretends to be.
His master’s belch
The filmmaker may long for the days when she would guide her screenwriter with a wink or a groan or with her choice of writer’s room snacks. And re-balance the pressure with an in-joke or regional gossip.
But when she chooses to generate a movie or movie material with AI, she has to start somewhere.
For the most part, it seems to be that it starts with a word.4 This initial word alone carries the weight of the massive and casually selective corpus of literature on which the software is founded. Each word is networked with every word and sentence in the Frankenstein mind of the software. So, the filmmaker’s first word is a big responsibility.
Her second word is just as pregnant. And it further introduces the issue of structure. Oh god!
The software developers have sought to ease this process. They’ve sought to ease it by modelling their word interfaces after the human chat. Chat is very popular. Chat is informal.
And chat is seductive: all it takes is that first word. One word, and you’re on your way to a quick script. Or a visual concept for your hero’s socks. Or a shot. A discrete moving image pulled, blinking and confused, from that dark pocket of ‘artificial intelligence’!
But this informality encourages a lack of rigour. A lack of care in the choice of words and word-structures (“prompts”) humans offer the machines. Much prompting is, as François Le Lionnais might put it, “eructative,” which is to say: belched. The effect is instant impact without precision.
And so the filmmaker considers her first word. And as soon as the filmmaker:
adds a second word, or
runs her prompt mentally through a second thought
composing, structuring, editing, rather than tapping in the first impulse and the next and the next
she has entered a new and probably necessary realm of artistry.
In such a playground, words and images become like characters. The background or provenance of each one becomes more mysterious:
Where had each word been before?
Did the machine find that word, “word,” in the bible or on the dance floor?
What does it mean that it was chosen?
How will it play with the others?
And yet, as precise as her crafted prompts might be, the result will likely disappoint the filmmaker’s expectations or remain imprecise or plain wrong. The computer must root through a very deep and mysterious pocket to make the appropriate connections. And it can’t see outside the pocket.
If you choose to write a film or a manifesto or a blueprint for production with any level of artistry, you must adapt yourself - your tricks and expectations - to these limits.
But belching can also be useful. To release pressure. For leisure. Or to do those filmmaking tasks that have an inverse effect on your ability to make a movie to the best of your ability. For example, by using ChatGPT to write a script “for the funders,” and then writing your real movie with moss and plasticine.
Please share your thoughts, queries, and exercises from this week’s lesson in the comments.
Fukushima Cinema School crowdfund offers Béla Tarr workshop materials to supporters
Béla Tarr is in Japan! He is teaching in Fukushima for a couple of weeks, as we have foretold. His work will be supported and documented by my friend and colleague, the most excellent filmmaker, Kaori Oda.
Some of Kaori’s video documents of the workshop may leak to the general public. Leak in the form of rewards for supporting a crowdfunding campaign. A campaign to “create a place for continuous creation and learning to inspire and nurture filmmakers and other artists in Fukushima!” The money will go to Fukushima Cinema School to help Mr Tarr’s workshop participants complete their work. And to help the organisers do it all again.
I believe that Mr Tarr is teaching in English. So the video material will be, too. But, since the crowdfunding call is in Japanese, I present an auto-translation of it here. Trimmed for concision.
“Students from all over the world will gather in Fukushima, meet and feel Fukushima together with Béla, and their experiences will be expressed and recorded as film works,” writes Kaori Oda.
“I have been tasked with documenting the activities of the students as they encounter Fukushima and create their works, as well as how Béla himself guides the students in Fukushima.
“This time, as we are seeking support, I proposed sharing the recorded footage with those who support the film class. We are considering offering rushes (lightly organised footage) or rough cuts (lightly edited footage) as perks/returns with specified viewing periods and numbers of times.
“What will be captured in the footage is unknown, but we would be greatly honoured if you would consider supporting us.”
As a bonus, you may spot me, your principal, in a couple of photos on the crowdfunding page.
Talents Sarajevo opens for applications
Talents Sarajevo is a very cool spin-off, or eruption in its own right, of the ‘rite of passage’ (when you finally get selected) Berlinale Talent Campus.
And even better, it’s in Sarajevo, possibly the best city. And during the Sarajevo Film Festival, certainly one of the most fun film festivals and best places in the world to be in August, and certainly warmer than Berlin in February.
Up-and-coming filmmakers, actors, critics, etc., from 17 Southeast European countries can now apply to take part in the 2024 session. Please apply, if eligible! And share the opportunity - heck, share this whole email - with your cineaste friends in the region.
Down-and-leaving filmmakers need not apply. They can start their own film school instead.
Apply here.
A 2000-year-old blog post about how to enjoy life
Seems relevant: around about the time “John” was writing about “the word,” Mount Vesuvius erupted in Italy and buried “an ancient library of papyrus scrolls” in lava.
The scrolls were rediscovered a couple of centuries ago. And now, boffins have used AI to partly recover and translate the text - without unscrolling the scrolls!
What did they find? They found “a 2000-year-old blog post about how to enjoy life,” according to their sponsors. Damn! Nothing about productivity or the author’s progress with the keto diet, then?
Anyway, I hope somebody is printing these newsletters and rolling them up into scrolls for posterity. For future volcano explorers to sit and marvel at.
You can read more here.
Next week, we’ll look at the synthetic, the generative, and the automatic. Or, what happens when you trigger a filmmaking process that won’t stop?
Class dismissed!
~Graeme Cole.
(Principal)
📹 Unfound Peoples Videotechnic | Cloud-based filmmaking thought. ☁️
🐦 Twitter | ⏰ TikTok | 📸 Instagram | 😐 Facebook | 🎞️ Letterboxd | 🌐 Website
We have no way of knowing if he was right.
This claim, at least, you could disprove by surveying the general public. And yet, on some level, it rings true.
Again, there’s no way of proving whether any of this is God’s doing, except to use science, which frankly has its own issues to resolve.
The gradual emergence of non-verbal interfaces such as picture prompts, generative breathalysers, etc., aside.