On jan 1 2027, i will send the best (non-explicitely-specialized for this task) model at the time a movie (file) and the pompt ‘balenciaga it’. If it turns the movie into a balenciaga meme version of the movie at a quality comparable or better to the balenciaga generations today (in mid 2026), this market resolves yes. (Slightly modifying the prompt to clarify you want it to balenciaga the entire movie is ok, but clarifying what balenciagaing means is not allowed) can pick a movie that doesnt have copyright if need be
1,000People are also trading
Fable 5 fallbacks to Opus 4.8 on this task due to their 'dangerous capabilities' classifier triggering strongly. There's also mentions in the System Card to the effect that they're silently sandbagging on balenciagaing it to stop other labs from catching up.
Updating down on odds of you can just balenciaga it by EoY with a deployed model. It's still possible imo for internal models if we're far into recursive self-improvement by then, though that seems unlikely. Either way they probably wouldn't release it to the wider public.
Also, I watched Harry Potter by Balenciaga 2 (2026) again, and I'm updating on this requiring superhuman writing / creativity / planning capabilities. I'm not sure who made that video. My guess is on a nation state of some kind, maybe Russia or China.
@Bayesian are there even any models that can currently produce 1.25 hours of video regardless of balanciaga inclusion?
@Bayesian I think you'll find cost problematic when testing it. For best models 1h of AI video costs $1000-$2000.