Posted by cheeaun 1 day ago
And yet, I am still quite good at mathematics.
Don't worry, there's still room for paintings in human culture, even though photos outnumber them billion to one.
So I decided to get my own custom slop. Opened Gemini and it said "video generation is free for a few days, try it out!". I uploaded a copy of Neuromancer and asked it to generate a scene.
MF had Cage inserting a power plug in his ear.
Ah and the time zone curiosity question? It went about it for a a minute or so and said "can't help you with that one, anything else?"
I think we've seen the best these LLMs can do and the race to the bottom has started. It's all cost cutting now, and I am very curious to see what will be of OpenAI and Anthropic. Probably will end up absorbed by MS, Amazon and Nvidia.
edit: https://share.gemini.google/hvH2A3Yva5Uj
First try.
I doubt their attempt failed because they wrote the question poorly. It was probably just the service being flaky.
So high quality I can certainly acknowledge because indeed the output from these models can be indistinguishable from human generated content. However, I am curious to know what you mean by creative here, in two senses:
1) Do you have an example of AI generated content being particularly novel or different, and not just a (admittedly highly consistent) rehash of what it was trained on? 2) What do you think it means to describe the output of a model that doesn't possess any agency as "creative"?
Instead, my goto examples of high quality creative works that are clearly grounded on AI tools is https://andreaciulu.com/projects/ and https://youtu.be/fyZhC2TXgcs
“model that doesn't possess any agency” implies a model acting on its own. Like a paint brush flying around the room painting the walls with no people around. Or a computer spontaneously turning on and publishing images to an online gallery with no on operating it. The creative process of using AI tools is one of exploration, pipelining and editing.
Commentators strawman image generators as magic paintbrush machine guns. But, the process of using them intentionally is more like nature/street photography. You can’t walk into the forest and declare “I will photograph an owl fighting a fox over a mouse today.” But, you can develop a sense for where to go to find good shots. Now imagine instead of walking in a forest, you can pop anywhere in the multiverse. You don’t know exactly what to expect each time. And, the space is infinite in many dimensions. But, you can still learn to navigate.
And, that’s just the process of direct prompting. One step past that gets you configuring pipelines of highly varied generative editing tools like https://www.runcomfy.com/comfyui-workflows/comfyui-product-r...
From there you get into integrations with 2D painting/3D modeling tools and video editors. Generating references which are used to generate style sheets, then characters, sets, blockings, dialog, motion, base renders, upscales, edits.
“Make me a pretty picture. Masterpiece, best quality” is easy to make fun of. And, it’s common because so easy many people are playing around with it. But, it’s tiresome to find all of the discussion assumes that’s all there can ever be.
He is rapidly coming to define himself through opposition, rather than promotion. Which is a dangerous and fruitless game.
Throughout human history there has never been a still useful technology that we as a species chose to stop using, and AI very much is. Maybe we'll come to regulate it, but the genie will never go completely back into the bottle and folks are going to need to get used to the new world.
I think that someone who's radically opposed to any use of AI is simply a poor developer.
You said it yourself - you have the knowledge and ability to know what to ignore and keep what makes sense, but that's because you spent years actually writing and understanding code.
The magic wand of AI is absolutely wreaking havoc with juniors engineers' ability to understand complex code bases and software engineering fundamentals. How will the engineers of tomorrow know what to keep and what to reject if all they've ever known is "you are a principal engineer. read this Jira ticket and build a solution. ultrathink".
I think that junior engineers can learn from AI, but the corporate setting isn't the best place for self-learning from AI. There's a lot to be learned from the average advice from the better AIs, as long as they are trying to understand the feedback. Please note that for these engineers, AI is replacing the old copy-and-paste from Stack Overflow. At least with AI they can dig deeper and ask questions.
I used to think there was no downside, but now I think there is.
You're developing the skill of choosing things out of a list to use. Your skill of creating that list is withering because you've offloaded it.
Now, I won't deny there's power in an AI code review, and for many people and occasions that tradeoff is worth it. For me, as well. But I'm convinced from my own experience it is a tradeoff.
Absurd. Scalzi fans are not going to stop reading his works because LLMs exist. If anything, established authors are the ones with the least to fear.
This talking point that when people dislike something it’s because it affects their personal finances is getting really tired. Not everyone is a greedy selfish prick who only thinks of themselves, people are allowed to dislike things for other reasons. That’s normal, it’s not an anomaly.
> but the genie will never go completely back into the bottle and folks are going to need to get used to the new world.
And apparently we’ll never share another original idea, either. “Genie is out of the bottle”, “you’re going to get left behind”. Apparently the people who called LLMs parrots got it wrong, they instead turn people into them, mindlessly repeating whatever the labs want them to.
The successful ones already knew how to write books - which directly contradicts the standard line about "It's just people who couldn't do it without AI."
That's just objectively wrong.
There have been some successful spammers who flood Amazon with tens/hundreds/thousands of titles, but that's more of an exercise in technology and marketing than writing.
Might just be final nail in the coffin for our cultural stagnation
On a utilitarian sense, no doubt a photography is more useful than a painting. But in the artistic/creative sense (what the conversation is about), I’m not sure the case is that clear-cut, hence the question.
Being rude about it doesn't make you right about it.
It sounds like people are repeating themselves because they have to keep repeating the obvious in every one of these threads:
The tools are useful, people are going to use them. Cope.
It's a new, and potentially very lucrative, technology. Given time the cost will go down, not up. The way that will happen is increased efficiency.
Do I know HOW that will happen? No, but I know that it will. It always has, and there are huge motivations for making it happen.
https://www.dpg-physik.de/veroeffentlichungen/publikationen/...
I've seen a few dim-bulbs or sales-types on social media say things of that sort, but never a serious adult.
Everything about the data centre buildout is a bet on the singularity, that's the whole operating premise behind what all AI researchers are doing. The most grandiose charlatans of silicon valley talk about that stuff and everyone else has got in line whether or not they believe it. Trillions of dollars have spent on bringing about AGI. If it doesn't usher in AGI (and it won't) then it's all a catastrophically stupid thing to do.