为什么说生成式 AI 只能生产平庸之作?
卡梅隆的完整表述,来自 三联生活周刊 的采访视频
下面是完整文字(中英双语,优先推荐阅读英文,意思更准确):
And you also describe the generative AI only can generate kind of average
Yeah well think about it logically if you’re feeding into a kind of giant blender that consists of everything that’s ever been done
it’s everything but it’s not any one thing, right?
So it’ll can only create a version of that which has already been done
Now if you ask a gen model to do something in the style of Avatar
you’ll get a glowing rainforest, right?
Okay, ask it the day before Avatar was released
Can’t do it
because it doesn’t know what the reference point is
It can’t create something new from scratch
because there never was a glowing rainforest before Avatar
and who knows what’s in somebody else’s imagination that they’re gonna create
So the human creative process is still the most critical thing
and the human creative process of creating characters is done by actors
and it’s based on the unique lived experience of the writer
then being fed through the unique lived experience of the actor
no two actors live the same life
They feel things differently
They express things differently in unique ways
And that’s why generative AI can at best
only do kind of the average thing
based on things that have already been done, right?
so it’s not true creativity
(采访者)
您还曾提到过 AI 只能生成一个平庸之作
(卡梅隆)
是的,我们可以从逻��上想一想
如果你把所有训练数据
都丢进一个巨大的“搅拌机”里
这些数据包含了所有曾经做过的事情
它包罗万象,但它并不代表任何一个具体的东西
所以它只能生成那些已经存在过的版本
比如说你让一个生成式AI模型
做一个《阿凡达》风格的画面
它会生成一片发光的雨林
但如果你在《阿凡达》上映前一天问它
它是做不到的
因为它不知道参照物是什么
它无法从零开始创造新的东西
因为在《阿凡达》之前从未出现过发光的热带雨林
谁知道别人的想象力会创造出什么呢?
所以人类的创作过程仍然是最关键的
而人类创造角色的过程
是由演员演绎完成的
是来自编剧的人生经历
再经过演员独特的人生经验
被重新表达出来
没有两个演员拥有完全相同的人生
他们感受世界的方式不同
表达情绪的方式也不同
而且都是独一无二的
正因为如此,生成式 AI
最多只能做到一种平庸之作
它只能基于已经存在过的东西
那并不是真正的创造力
🚨 The Future of Video Game Graphics Finally Revealed!
In the near future, AI models will act as the live rendering layer between game logic and your screen.
This means games won't just look "good". They will be fluid. You could switch from hyper-photorealism to anime, or change the setting to different historical eras in real-time. This is the future people at Nvidia is working towards.
One game. Infinite styles.
Here you can see The Witcher 3 made photorealistic (and anime style). Red Dead Redemption 2 in Japan below ⬇️
Introducing Camera Angles
Rotate views, shift angles, change perspectives, and preserve every detail. You’ve never moved a camera like this before
Now available on Freepik Spaces
We just acquired @weavy_ai, soon to be known as Figma Weave.
With Weavy, you’ll get more models, more tools, more ways to create within the canvas. More to come