#30: That Obscure Object of Desire
This week: AI-film making and the AI Film Academy Awards // Things to read, see and do in the ecosystem this week // High-quality data-sets for AI training, plus Bunuel, Chris Marker and more...
So we’ve hit the big 3-0 this week.
This week’s seen a big bump in new readers thanks to a recommendation from
. A huge welcome - I hope you enjoy being here with us as we delve into the emerging edges of art, ideas and technology.It took a bit longer to 30 issues than I thought - the whole of April off on leave, and then a half-term break earlier this month when I was writing my latest piece for The Art Newspaper.
But 30 issues in, there still feels like a world of things we haven’t got to yet.
Next week will be a recap where we index the ideas and themes of the last 10 issues, then we’ll dive back in to the ocean of Creative R&D.
Happy reading and thanks for being with us.
ART! // Un Chien AI-ndalou
Film-making of all creative disciplines probably has the most to gain and the most to lose from the emergence of generative AI.
The movie-industry, now so based around the large-scale teams producing visual FX, is scared about what generative AI means. The risks to the jobs of those VFX artists - the kind of high-paid, high-productivity jobs governments care about and champion - seem very near and present.
We saw that earlier this year when the very good horror film “Late Night With the Devil” got caught using gen-AI for three still images which appear as interstitials during the film.
But the promise is rich - of a medium so dependent on resources becoming in some way closer to writing a novel or poetry, an individual skill that might conjure up tomorrow’s Kafka rather than its Kubrick.
We are still a long way from that.
But that promise was very much on show with the announcement of the first AI Film Academy Awards at Lisbon’s NFC conference last week.
Co-founded by newsletter reader Leo Crane (thanks Leo!), the awards celebrated new talent and provided a brilliant moment to look at where we’re at in the emergence of whatever new form will come out from this new technology.
You can see the pool of winners and nominees over on SuperRare where they’re for sale as NFTs.
Three stood out for me.
I LOVED AIInferno by MFluxmusic -highly enjoyable, high-octane mutating madness. There’s a risk of a certain technological and philosophical seriousness that weighs on a lot of this kind of work - an immediate novelists turn to existential profundity and meditativeness which this film kicks to the kerb and just goes beserk. Very good.
Also very good was Birthday Story by Rainisto.
This went all in on the existential heaviness, but I thought it embraced very boldly the unstable nature of representation in AI-generated imagery - the way no two frames properly quite lead to each other, and the way that instability can become an aesthetic feature rather than a bug.
There’s an echo of Francis Bacon in this that is resonant and meaningful. It’s apocalyptic ending also had a nice wildness to it.
Last up, I also really liked Bacillus Anthracis by Hallidonto, which won the Community Award. The voiceover is bonkers (check the subtitles below), but the imagery has a lovely organic weirdness to it.
I enjoyed watching these films, but felt very much like we are still at the beginning of something rather than any kind of maturity.
And that the artists needed to forget what they know about “film” and reach for some new grammar that’s barely begun to be articulated yet.
There were two films I found myself thinking about as a result.
The first is Bunuel and Dali’s Un Chien Andalou - the great Surrealist short, still shocking a century after it was made.
It’s weirdness is so much in the juxtaposition of its images. Its plainness allows its harshest moments - the famous cutting of the eye - to jump out harder.
The second is Chris Marker’s La Jetee, a short-film which strips film-making down to a sequence of still images, the base medium from which we fake the illusion of motion. 62 years and counting and it still feels as fresh and odd as it did the day it came out.
It’s from this kind of alternative thinking that I think the new medium of AI-enabled visual storytelling will really emerge.
But I’m glad these early attempts to cluster talent together and give them a voice exist.
There’s so much to do here.
Keep watching.
IDEAS! // What to watch, read and see in Creative R&D
I wrote back in issue 10 about how important I think Berlin’s LAS ART Foundation are - I think they’re amongst the most vital art makers in the world. So when they’ve got a new show coming … if you can, you need to go. Opening 12th July at Halle am Berghain, the new show, THE SOUL STATION presents a major new work by Danielle Brathwaite-Shirley and a survey of their work to date. MUST GO.
This journal piece makes disturbing reading on the physical impacts on the brain of excessive internet usage, particularly amongst young people. What are we doing to ourselves? I had to go and watch Videodrome straight after reading this - David Cronenberg’s parable about the horror of TV makes more sense than ever.
I am glad that a specific award for drone photography exists - get entering now!
I LOVE the look of this immersive experience coming to a forest near Guildford next Sunday. Sensing the Forest is a sound installation that will transform environmental data from the forest into an auditory experience. By converting often hidden processes into sound, the installation aims to makes complex ecological interactions tangible. It’s only on for one Thursday afternoon - but looks amazing. Get there if you can.
Last-up this week. More madness from the litigious world of digital-meets-copyright. Digital artist Krista Kim Mars has won a lawsuit declaring her the sole artist behind Mars House, a design for a digital house she sold as an NFT at the height of the NFT hype cycle. The freelance renderer she used to help make it had sued for co-authorship - the courts said no. Agree better contracts people!
The Mars House - coming soon to a courtroom near you.
TECHNOLOGY! // High-quality data for the future of AI
Andrei Karpathy was head of AI at Tesla and one of the founding team at Open AI, so apart from knowing about megalomaniac bosses, he also knows his AI cheese.
So his tweet this week about FineWeb-Edu, a huge high-quality data-set for training AIs and its impacts is worth paying attention to.
There’s a very long and detailed blog explaining how this all works over on Hugging Face - but the principle is simple.
Big, high-quality data makes AI’s produce better outcomes than the very messy data scraped from the web most model so far have been trained on.
That might seem obvious, but it takes a hell of a lot of work to build that volume of good data. 1.3 trillion tokens (broadly this means 1.3 trillion words of content) is a lot of stuff.
The background this is happening against is the spree of investment the big AI companies are making into licensing training data from major publishers. The table below shows just how much of what we’re seeing as outputs from ChatGPT will now be based on paid content.
This is just news, editorial and digital media - there’s so much more good data that needs to be properly prepared for use by AI, and properly licensed.
This really is my challenge out there to data-owners in the cultural and creative sectors - there’s a lot of you reading this, or policy people round the edges: get your data ready and start making deals.
If we want what we care about - the understanding of history held in digital collections and digital arts - to influence the future, this HAS TO BE a primary activity for the years ahead.
I worried in my piece on AI in The Art Newspaper last week about the distancing effect the summarising interfaces of AI chatbots might have on access to cultural content. That’s a major issue.
But a bigger issue is that our data might not be there in its best form in the first place.
Museum directors, fund managers, politicians - please start thinking about this: what do we need to do to build our own equivalent to FineWeb-Edu, how do we do it, who’s doing the deal … and why haven’t we started already?
This is mission-critical - and as far as I know it’s something that’s not even begun.
Get moving!
I apologise profusely for the “Un Chien AI-ndalou” section title. Dad joke of the worst kind (if your dad is a film geek anyways).
It did make me laugh though, and the title of the newsletter therefore became a reference to Bunuel’s last and greatest film, where the young woman Fernando Rey keeps changing from one actress to another. No reason given - a cruel and funny metaphor for how unstable love is, and a pre-echo of the visual instability of AI imagery?
To next week!