ChatGPT's new feature for generating Ghibli-style images has seen a warm reception from social media users. However, this surge in popularity recently caused the company's servers to crash. In light of this, CEO Sam Altman issued a plea for users to exercise restraint. While he praised the enthusiasm for creating Ghibli-style images through ChatGPT, he also expressed concern, stating that generative AI, while making tasks easier, presents significant environmental risks.
In his message, Sam encouraged users to limit their image creation, pointing out that the strain on their graphics processing units (GPUs) was becoming overwhelming. As a result of the increasing demand for Ghibli-style images, the company has imposed a daily cap on image generation for free users, allowing only three images per day.
You might wonder how much electricity is consumed to generate an image with this AI tool. To put it into perspective, creating a Ghibli-style image uses about the same amount of electricity as running a 60W light bulb for approximately 3.25 minutes. In fact, generating an image consumes ten times more electricity than producing text.
According to reports, generating a Ghibli-style animated image consumes electricity at a rate of 2.5W to 3.5W per hour. This amount of power can charge your smartphone by about 50 percent. Large data centers equipped with GPUs have been established to handle the image processing demands. When at full capacity, these GPUs can draw up to 700W of electricity.
In addition to electricity, vast amounts of water are also used to keep the GPUs cool in data centres. Each AI-generated image requires around one trillion floating-point operations (FLOPs). To put this in context, a high-end AI accelerator needs to perform about 100 billion FLOPs to generate text using AI.
ALSO READ: New research indicates this AI model outperforms humans in Turing Test evaluation