Didn’t feel altruistic when they told me I couldn’t put affiliate links in the sidebar for my subreddit.

I was gonna say… how is it not this? Weird world.

Here ya go. This is pretty much exactly whatcha describe.

Meanwhile, I just had Cluade turn a few obscure academic papers into a slide deck on the subject, along with presentation notes and interactive graphs, using like 5 prompts and 15 min.

It can be helpful to approach this from the other direction. The part of the brain that works like an LLM.

Interestingly, LG’s AI Wash pre-dates the public release of ChatGPT by almost two years. Truly pioneers.

I can’t really agree as a video producer. Luma, Krea, Runway, Ideogram, Udio, 11Labs, Perplexity, Claude, Firefly -> All worth more than they’re charging, most with daily free options. They save me a ton of time. Honestly, the one I’m considering dropping at the moment is ChatGPT.

“We” being anyone interested in scientific exploration, hence the papers linked in the post and comments.

Oh man, it saves me so much time - but it is like working with a bipolar intern. Sometimes it pulls off these amazing fills that would’ve taken me hours to do by hand. And sometimes it has a fit trying to fill in a cloudy sky, and I just do it the old fashioned way. I’m pretty systematic about testing tools in general, and gen fill resists all attempts at building a reliable workflow. Ya really gotta switch up tactics to react as ya go, which can be fun when it’s not irritating. Plus I find surreal hallucinations hilarious, so it makes up for some of its behavior issues with entertainment value.

On the flip side of that, the discoveries that accompanied attention transformers are changing the way we think about our own neurology:

A new study bridging neuroscience and machine learning offers insights into the potential role of astrocytes in the human brain.

Convergence of Artificial Intelligence and Neuroscience towards the Diagnosis of Neurological Disorders

And how we study it:

Artificial Intelligence shaping the future of neurology practice

I try not to be quick to assume that intuitions around similarities are correct or incorrect, but I trust they are worth exploring.

I can’t imagine willingly going back to before Adobe added Generative Fill to Photoshop. Gen AI will certainly remain more than an academic curiosity, at least until they can be replaced with something better.

Here’s the paper onno’s alluding to, for reference.

I think you mean, “As a large language model…”

Typically speaking, the appearance of similarities should be explored for functional similarities. Even in the world of prompting, a user gets demonstrably better results (in some contexts) by acting as-if they’re talking to a person. We can explore the similarities between interpersonal interaction and AI-interaction for overlap in effective modalities without claiming they’re equivalent.

Isn’t your comment more of a perspective on the public perception of AI, the missteps surrounding its implementation, and its current role - rather than an examination of the potential role (practically speaking) of generative AI in a more general AI model? As is the thrust of the post, generative AI will necessarily be part of a larger AI, in part to make up for its weaknesses, in part to utilize its strengths.

That said, generative AI isn’t nearly as endangered by generated training data as is commonly understood. Even if it were that bad, embodiment is rapidly changing the landscape. There are a ton of papers about how to use larger models to make smaller models more effective, using generative AI to improve generative AI along with efficiencies. Heck, novel efficiencies get developed almost as regularly as novel use cases. We’re always learning how to do more with less.

¯|(ツ)/¯ mostly looking to spark discussion.

Just looking at the mental space of the three fanboy types… Mac seems the chilliest to hang with.

e: lol, case in point

The voice library 11labs added includes some really reliable and expressive models. I’ve only trained a few voice clones, but I find them totally usable for swapping out short lines to avoid having to bring a subject back in to record. I’ll fabricate a sentence or two, but for longer form stuff, I only use AI for the rough cuts. Then I’ll practically record as a last step, once everything’s gone through revision cycles. The “generate a few and chop em together” method is fine for short clips, but becomes tedious for longer stuff.

Funnily enough, when I say roto, I really just mean tracing the subject to remove it from the background. Background removal’s so baked in to things now, I dunno if people even think of it as roto. But I mostly still prefer the Adobe solutions on this - roto brush in After Effects, for the AI/manual collaboration. As for roto in the A Scanner Darkly sense, I’ve played with a few of the video to video models, but mostly as a lark for fluff B-roll.

Full open source, nice! I respect the effort that went into that implementation. I pretty much exclusively use 11 Labs for TTS/RVC, turn up the style, turn down the stability, generate a few, and pick the best. I do find that longer generations tend to lose the thread, so it’s better to batch smaller script segments.

Unless I misunderstand ya, your controlnet setup is for what would be rigging and animation rather than roto. I do agree that while I enjoy the outputs of pretty much all the automated animators, they’re not ready for prime time yet. Although I’m about to dive into KREA’s new key framing feature and see if that’s any better for that use case.

It comes from when a full course dinner would always begin with soup and end with nuts.