I haven’t seen any creatively useful AI functions since DeepDream or content aware fill/object removal.
I am boggled by what they think would actually be useful for, because it seems like AI developers legitimately don’t understand anything about the filmmaking process or even photo composition. Users just keep slapping the Roll Dice button until it spits out something that looks like a marvel flick or hentai, so it’s not like they’ve had any useful feedback whatsoever.
Yeah, there are so many amazing uses for the tech, but typically everything useful is just a single task product or function which can run locally and doesn’t need much or any compute power after being built in the first place.
They just keep pushing the stupidest and least efficient use cases and it’s consistently bad at it.
I got a 90 day trial of Gemini 3.1 pro and it wasn’t even good at being a chatbot! Like, what the fuck is the point of doing all this crap if even the most powerful models still suck at holding a conversation as of May 2026.
And other things did them better and faster, like llms never needed to exist. Everything good that can be done a llm was already either being done better by something else or some sort of llm or ml already existed for it like alphafold in protein folding
I haven’t seen any creatively useful AI functions since DeepDream or content aware fill/object removal.
I am boggled by what they think would actually be useful for, because it seems like AI developers legitimately don’t understand anything about the filmmaking process or even photo composition. Users just keep slapping the Roll Dice button until it spits out something that looks like a marvel flick or hentai, so it’s not like they’ve had any useful feedback whatsoever.
deleted by creator
Yeah, there are so many amazing uses for the tech, but typically everything useful is just a single task product or function which can run locally and doesn’t need much or any compute power after being built in the first place.
They just keep pushing the stupidest and least efficient use cases and it’s consistently bad at it.
I got a 90 day trial of Gemini 3.1 pro and it wasn’t even good at being a chatbot! Like, what the fuck is the point of doing all this crap if even the most powerful models still suck at holding a conversation as of May 2026.
deleted by creator
And other things did them better and faster, like llms never needed to exist. Everything good that can be done a llm was already either being done better by something else or some sort of llm or ml already existed for it like alphafold in protein folding
That is actually a thing that exists now, there’s a million of them if you search for it. No idea how effective they are, though.