š„From Ghibli-style art craze to melting GPUs š°All About GPT-4o, GPU, and Ghibli |
|
|
A meme portraying how everyone is generating Ghibli-style images using GPT-4o. Source: Facebook
OpenAI's newly upgraded image generation feature in ChatGPT-4o has sparked a global trend: creating images in the iconic Studio Ghibli style. Known for masterpieces like Spirited Away and Howlās Moving Castle, Studio Ghibli is beloved for its warm colors and whimsical aesthetics.
The trend has grown so intense that OpenAI CEO Sam Altman tweeted, āOur GPUs are melting.ā The very next day, he followed up with another tweet begging users to "chill"āhis team was losing sleep. Most recently, OpenAI announced that over one million new users joined within a single hour.
Today, weāll dig into the wave behind this trend:
-
Why are GPUs āmelting,ā and what can be done about it?
-
Are there copyright concerns with Ghibli-style image generation?
-
Does Hayao Miyazaki (head of Studio Ghibli) really hate AI?
|
|
|
Why are OpenAI's GPUs Melting? |
|
|
Sam Altman 's tweet.
š¤ Why is GPU usage spiking?
Among AI models, image generation requires significantly more computation than text generation. While generating text mainly involves calculating word probabilitiesāa relatively lightweight taskāimage generation involves gradually denoising within a latent space. This process requires heavy parallel computation, which leads to substantial GPU usage.
š¤ So, how can we address the GPU shortage?
ā
Model Optimization Techniques like LoRA (Low-Rank Adaptation) and quantization reduce model size and computation without compromising performance. These methods improve efficiency and help ease the burden on GPU resources.
ā
AI-Specific Chips Hardware specialized for AIāsuch as Googleās TPUs (Tensor Processing Units) or Appleās NPUs (Neural Processing Units)ācan reduce reliance on traditional GPUs and deliver faster, more efficient processing.
ā
Cloud & Distributed Computing By dynamically allocating GPU resources in the cloud and using distributed computing across multiple servers, we can distribute the load and prevent overloading any single server or GPUāensuring more stable performance.
ā
New Model Architectures Moving beyond traditional Transformer-based models, next-generation architectures can handle tasks more efficiently. Sparse modeling techniques and lightweight neural networks offer better performance with fewer resources.
|
|
|
GPT: Ghibli Production Tools? š¤ |
|
|
Not long after expressing concern that GPUs were "melting," Altman followed up with another tweet, pleading for people to "chill" on generating images āsaying the team needs sleep. In response, one user cleverly coined a new acronym for GPU, GPT, and AGI, based on the ChatGPT + Ghibli-style image frenzy.
|
|
|
ā¶ļø GPU: Ghibli Processing Unit ā¶ļø GPT: Ghibli Production Tools ā¶ļø AGI: All Ghibli Images
The global obsession with Ghibli-style image generation has taken over timelines everywhereāeven Sam Altmanās Twitter profile now sports a Ghibli-inspired portrait.
But while much of the world is swept up in the nostalgia and novelty, thereās a growing conversation happening in the backgroundāa more serious one. Artists and creators are raising important questions about AI, authorship, and copyright. Letās take a closer look at whatās being said. |
|
|
Itās not easy to take one clear side. These debates often fall into a familiar back-and-forth:
āEveryone learns by studying past worksāhow is that illegal?ā vs. āThis isnāt the same. This is copying and mass-producing replicas.ā
Right now, the spotlight is on illustrators and visual artists. But whoās to say it wonāt be your story next? Itās important to marvel at progressāand yes, even enjoy it. But itās equally important to pause and listen to the voices of those who may be losing something in the process. |
|
|
Is Hayao Miyazaki "Disgusted" by AI? |
|
|
Hayao Miyazaki is widely quoted as saying AI is ādisgustingā and āan insult to life itself.ā These strong words stem from a 2016 NHK documentary, where Miyazaki reacts to developers demonstrating an AI image generation model. The scene captures his blunt, heartfelt responseāone that has since echoed across the broader conversation on AI and creativity. |
|
|
Hayao Miyazaki watching an AI-generated video. Source: NHK
|
|
|
To understand the full meaning, we need to know the context.
The developers presented a scene generated by AI that depicted zombie-like movementsāa creature with missing limbs crawling across the floor. It was a graphic and provocative video, shown as an example of what AI could create.
Upon viewing it, Hayao Miyazaki recalled a close friend with a disabilityāsomeone who struggles with muscle controlāand said he could never find that video amusing. He emphasized that only those who are utterly ignorant of human suffering could make something like this. He called it ādisgustingā and said it was āan insult to life itself.ā
When comparing various translations of the scene, there seem to be two common interpretations of what Miyazaki says next:
-
I would never want to use this kind of 'technology' in my work.
-
I donāt want my work to be associated with this kind of 'thing' at all.
Curious about the nuance, I looked up the original Japanese subtitles and translated them myself. |
|
|
The phrase often translated as "this kind of technology" was actually just ākore (ćć)" in the original Japanese, meaning simply āthis.ā
That subtle difference suggests that Miyazakiās criticism may not have been directed at AI technology itself, but rather at what was madeāgraphic, provocative content created without empathy for human suffering. While he does express concern about the direction of technology later in the clip, it doesnāt come across as outright hatred or disgust toward AI as a whole.
š You can watch the video yourself and decide what he really meant. |
|
|
We still donāt know exactly what Hayao Miyazaki thinks about AI-generated Ghibli-style imagesāand even if we did, he doesnāt speak for every artist. Whether AI-generated art constitutes copyright infringement remains unresolved. And thereās no easy fix for the āmeltingā GPUs struggling to meet soaring demand.
Weāre living in a world where the boundaries are blurry and the answers rarely come easy. In so many areas, there are more questions than settled rules. Thatās why itās more important than ever to explore different perspectives before forming opinions. In an era where AI offers quick takes and ready-made views, we need to put in the work to shape our own.
|
|
|
Solution for Trustworthy AI
Datumo is an all-in-one data platform for your smarter AI
|
|
|
Datumo Inc.
š contact@datumo.com
|
|
|
|
|