really impressive work considering the reported size of the model and training hours.

50+ hours on 256 H100s is considered impressively low training?

Really makes me wonder if any of this incredibly computationally expensive research is worth it, which seems only useful in potentially promising a future in which humans are given less opportunity to express themselves creatively - while delivering them an infinitely produceable amount of ai generated 'content' to passively consume

>Really makes me wonder if any of this incredibly computationally expensive research is worth it

I'm wondering the same thing. 256 H100s were hot for two days straight to be able to make short clips of cartoons that almost don't look like shit?

It just isn't compelling to me.

So, costs roughly 15k?

$15k for some clips of tom and jerry that almost look passable. What a deal.

>Feels like hand wavy nod to a human soul

I have no idea what you're talking about. I've tried to understand where you're coming from with this and the only logical conclusion I can make is that you spend a lot of time engaging in debate about creativity and art as it relates to new AI technology, and you are simply re-igniting previous debates instead of engaging with me.

>It’s a really lame circular routine and lived experience being around my peers these days; oh I hate my job but this new thing is an abomination and affront to my chosen job.

It sounds like you're arguing with your peers, and not me, because I don't hate my job and I don't think AI is going to replace it any time soon.

>Are we that creative copy-pasting and git pull deps someone else maintains? IT is librarian work these days. Little in the day to day is novel creativity.

This isn't what I do at my day job, and if that's what you do... I think I have a good idea of why you interact with the internet like this.

Sorry, you're right lol. I'm just accustomed to other major lab gazillions of hours of training.

This is by no means a comment about the quality of the project, but my god it's very uncanny in some frames. I feel like this would open up a lot of doors to creepypasta content. I'd love to play around with this

Looks pretty bad but considering this was impossible a couple years ago(as far as I know) it’s very impressive progress

Aside from memes I do not see the progress value.

The main progress value is that Test-Time Training appears to work very well in practice. I think that as labs begin to test it as scale in LLMs, it will become commonplace in next-generation models.

Sure, Im not saying they’re not useful tools but let’s not buy into the hype and pretend they’re some silver bullet. Im aware they’ll change how we do programming and other tasks but I don’t think they’ll completely displace human thinking. As for art, i’m not sure artists will cease to exist either. Unless we as a species cease to exist, but then what is all this progress for?

Are you saying you don't see the value in video generation? The potential for unlimited high quality and customizable content generation?

who said that personalised, infinite content generation is a good thing? I watch movies and listen to music because I want to be challenged in some way. I don’t want tailored content that prevents me from exploring new territory and keeps me trapped in a personalised echo chamber.

Why do you think tailored content will prevent you from exploring new territory? You are choosing the content.

I can't wait to watch the first entirely generated short film and create my own.

if something is done for yourself by yourself is not challenging by definition IMO

Reading the prompts reminds me of this interesting short story from Steven Millhauser called "Cat 'N Mouse"[1]

Would be really cool to just use this (or parts of it) as one of the prompts and see what results.

[1] - https://www.newyorker.com/magazine/2004/04/19/cat-n-mouse