https://www.argil.ai/

What happened on Dec. 23rd, 2023?

@Mira

Mira sold her shares all at once

LLMs are dumb as rock.

https://www.goldmansachs.com/intelligence/pages/gs-research/gen-ai-too-much-spend-too-little-benefit/report.pdf?ref=wheresyoured.at

https://www.tiktok.com/@rizz.records.yt/video/7389286025635433758

AI "brain rot" videos seeing a lot of success on TikTok. This one has 1.3 million hearts and counting.

max joe justice trailer
https://www.reddit.com/r/aivideo/comments/1dxxi49/algenerated_movie_trailer_final_justice_the_first/ and we're only in 2024 Q2

this could easily do a die hard cheesy blow em up action film in 2028

I'm listening to this interview with Luma Chief Scientist Jiaming Song: https://a16z.com/podcast/beyond-language-inside-a-hundred-trillion-token-video-model/. He claims Luma will achieve real-time video generation in less than a year.

I made a market on it: https://manifold.markets/jim/will-luma-ai-release-realtime-video

my credence is 64% yes

Mine is 10% yes

Four years from Jan 1 to resolution. 1/8 of that is gone. Who is willing to claim we have advanced 12.5% of the way to a full high-quality movie to a prompt in that time?

I might remind you that Sora has still not been released.

runway just released gen-3 like yesterday

and do you think that is 12.5% of the way?

not really answerable,. Also, progress will accelerate

We will see.

seeing what sora/runway-gen-3 is producing, + latest lipsync ex https://humanaigc.github.io/emote-portrait-alive/ + elevenlabs voice + udio/suno audio

you could have a ML pipeline stitching it all together

short 4-8 second shots to look like StarTrek II : Wrath of Kahn,
https://www.imdb.com/title/tt0084726/

I guess, is the question asking , can it do a 'big budget' movie of 2028, or 'big budget' of 2024,2010,1980s etc..

I'm a little confused by the question., I totally think SORA and SOTA models and SOTA voice models have gone probably about 30-40% of the way here. Obviously that number is a bit arbitrary, because the final 2% is the hardest.

Each 1% should be equally easy, or you haven't divided up the percentages correctly.

Which is why claiming the models are 30-40% there is obvious nonsense.

@DavidBolin I think there's a lot of use in defining fractional progress in ways other than "ease" or "amount of resources put in to get there." E.g., the notion that the last 0.1-5% is the hardest has been very useful to explain delays in other AI technologies like self-driving cars. There, % is something like fatalities or collisions per mile driven or another safety metric, which seems easier to define than film production but not by too much compared to other creative pursuits (e.g., revenue something would generate if released, quality rating from human evaluators, time or wages of human labor that it displaces).

Regardless, we are not "30-40% there" by any sensible metric, even one with easier rather than harder parts.

The best argument for yes I have seen so far.

unfortunately we're plateauing with gpt-4o

Forty total GPTs is now called plateauing??

Is it easier to make a movie than to take an order for a cheeseburger?
https://edition.cnn.com/2024/06/17/tech/mcdonalds-ai-drive-thru-program/index.html