outright admitting they are less valuable than trump social
I just don't understand how Reddit makes money as a publicly traded company unless it makes itself unusable by selling ad space until it's another Facebook where you gotta scroll through 18 promoted posts to find one actually made by users.
Any AI programmed with Reddit's posts will murder itself, possibly us too.
I think that AI actually will change the world, but none of us will like the changes.
I mean will be like everything. People higher up will view it as a way to get rid of employees and it "just works!". Meanwhile much further down the chain there's a load of people being paid pennies on the dollar to sift through shit to feed into the AI ingest.
It's actually great for certain things, like being a better Google at times for searching up specific queries, but it sucks complete and utter ass at coding, painting and writing and I doubt it will ever properly work. Since all it is doing is drawing from a dataset to generate something "new" by slamming "old" shit together and adding a couple fingers and ears.
I signed up for the IPO and offer to buy shares at half the projected share value![]()
It depends what you want. Copilot auto completes my code successfully 80% of the time, but misinterprets my intent and generates valid-but-incorrect code 20% of the time, which means my job is more proofreading than writing code when I am using it. It doesn't make me that much faster, and fails to do the hard part of the job, which is deciding precisely what functionality to implement. Great for docstrings, though.
For writing/art, it is getting better and I do wonder how it will end up. The way these models work, they will never actually generate something innovative because they are limited by their training sets. However, there is a lot of unexplored territory within their training sets. You could explore that territory by having it make more Seinfeld episodes (boring) or by having it mash up unrelated concepts. Some mashups are already done, like Macbeth as a story about lions (The Lion King), but some could be novel. There is a lot of in domain content for these models that is unexplored.
I could see using AI to make storyboards and would be shocked if people aren't already doing it.
Also, it is worth being real about how derivative the vast majority of entertainment is. You think AI can't write an early 2000s style romcom or Adam Sandler movie? Please
archi begins and ends with AI
It is not good at coding at all, can it generate code? Yeah sure. Is it actually good code? no
Same for other 2, it's "passable" but again it just copies and generates off of things already known which is of course a form of art, but it can't create anything new and exciting. There are lots of very odd or weird circumstances that go into writing and art that can't be replicated. Can i tell an AI "write me a script, but also you're budget is x and you're working another job at the same time to make rent and your father died half way through the process" because that can all play into how something gets put out by an actual human.
It writing scripts and storyboards and stuff, means that it just gets fed up to someone else who needs to review and edit it who's going to get paid much less then being the "writer" but has to do all of the same legwork in the end.
I do agree on that, but that makes me even more apprehensive because we're already so gun shy about any new IP because "it's not tested! And we need to spend 20 billion dollars on marketing obviously or it won't work so we'll lose money!"Also, it is worth being real about how derivative the vast majority of entertainment is. You think AI can't write an early 2000s style romcom or Adam Sandler movie? Please
There is a lot of AI art that is uncomfortably good, especially fantasy/sci-fi/characters. It came a long way very, very quickly.
I think that may be partially because there is a *HUGE* amount of essentially free and already categorized art in those genres too.
I think if you take a cursory glance it seems good but it falls apart if you look at it for any length of time. So many elements just flat don't make sense. Shit coming out of nowhere or disappearing into nothingness etc. Fantasy art is where it excels because there isn't any realism for it to bump up against.
These are literally the first video generating AI models, though. It is reasonable to expect their quality to increase dramatically in the coming year or two because the data exists. Imagine reading a video caption and then watching the video. When the AI has reached the point where it is as good as its data source, then IMO the easy gainz are over.
I don't know where the next performance boost for high end generative text AI models is going to come from. It is already basically as good as the training data. I have seen some propose adding specialized sub models (like having a separate model for math questions) and then making a framework that lets you seamlessly combine them. Could work!
Making the models more efficient (cheaper to train) is going to decrease the performance edge that premium models have over open source ones and it does look like that is coming, so I expect free models with comparable performance to GPT-4 in the coming year.