> ... i think google will take the lead here as they just have monopoly on dataset ...
I think there's plenty of data for people to train substantial models, either drawing directly from the public domain or scraping content. The cost of compute and disk space will continue to drop exponentially fueling the ability of small businesses and amateurs.
Compute is still following Moore's law, more or less, when you allow for GPUs. Hard drives are relatively on the same path, with maybe a price halving every 3 years or so.
> ... while on open source mark zuckerberg will be continuously leveling the playing field so the general population can keep up in terms of availability of access and end-result productivity.
Meta's offering is not open source. I agree that all the big players will make substantial moves in this space but I'm much more interested in the niche models that are actually libre/free/open source. I suspect FOSS will eventually eat all the big players lunch but I don't have a good read on it (nor do I know anything about China's OSS models).
> it's gonna be great to see lots of micro communities offering better content than netflix/disney/hbo ...
It sounds like a "Black Mirror" episode but I can't wait (and I don't think it'll be as bad as "Black Mirror").
> i wonder how much of the industry will transition into pure prompting too ...
My read on this is that we're in the "experimental art house" phase of this type of generative AI. So many people create weird things, experimenting with the technology but ultimately having low expectations in terms of what gets produced.
Sometime soon (1-2+ years?) my guess is we'll see finer control with someone, either an actor, director, etc. that can provide dialogue, facial expression and body language with minimal setup, a camera or voice recording, say, that can make or refine avatar/agent performance. I would guess this would be available to specialized visual graphics shops but will eventually bleed out so that amateurs have access. I think we might be seeing this available already.
Eventually, we'll have "make what I want", but I would imagine that's 5-10+ years away.
Regarding the video generation stuff, I can foresee a Canva style UI that allows you to define the position of your camera, movement, lighting etc. coupled with a prompt. Or a catalog of exiting movie scenes that you can select as a starting point to define the setting, cinematography, etc.
I think there's plenty of data for people to train substantial models, either drawing directly from the public domain or scraping content. The cost of compute and disk space will continue to drop exponentially fueling the ability of small businesses and amateurs.
Compute is still following Moore's law, more or less, when you allow for GPUs. Hard drives are relatively on the same path, with maybe a price halving every 3 years or so.
> ... while on open source mark zuckerberg will be continuously leveling the playing field so the general population can keep up in terms of availability of access and end-result productivity.
Meta's offering is not open source. I agree that all the big players will make substantial moves in this space but I'm much more interested in the niche models that are actually libre/free/open source. I suspect FOSS will eventually eat all the big players lunch but I don't have a good read on it (nor do I know anything about China's OSS models).
> it's gonna be great to see lots of micro communities offering better content than netflix/disney/hbo ...
It sounds like a "Black Mirror" episode but I can't wait (and I don't think it'll be as bad as "Black Mirror").
> i wonder how much of the industry will transition into pure prompting too ...
My read on this is that we're in the "experimental art house" phase of this type of generative AI. So many people create weird things, experimenting with the technology but ultimately having low expectations in terms of what gets produced.
Sometime soon (1-2+ years?) my guess is we'll see finer control with someone, either an actor, director, etc. that can provide dialogue, facial expression and body language with minimal setup, a camera or voice recording, say, that can make or refine avatar/agent performance. I would guess this would be available to specialized visual graphics shops but will eventually bleed out so that amateurs have access. I think we might be seeing this available already.
Eventually, we'll have "make what I want", but I would imagine that's 5-10+ years away.