Creating just nine images through AI is expensive

0
1
 Creating just nine images through AI is expensive.  As well as fully recharge your mobile

The AI ​​revolution is not cheap. The large computing resources required by generative AI models are accompanied by an uncomfortable fact: these training processes and the use of said models themselves cause significant CO2 emissions. We already know that ChatGPT and Bard are polluted, but Midjourney and DALL-E 2 of the world are even more polluted.

The study. Research conducted by researchers at Carnegie Mellon University and the startup HuggingFace reveals that using generative AI models requires a lot of energy. In the study, https://arxiv.org/abs/2311.16863, they revealed the use of a tool called Code Carbon to compare the CO2 emissions of the 16 most popular models used by HuggingFace Hub, an open platform that allows these models to be tested. and easy to use.

Read Also:  Reasons why the Chevrolet Aveo is one of the favorite cars of Mexicans
Imaging creates more emissions than any other process currently covered by AI and generative AI models.

ChatGPT consumes, but DALL-E consumes more. The conclusions make it clear that text-generative AI models—chatbots similar to ChatGPT—consume a lot of energy, but things are more worrying in the case of models dedicated to image creation.

Comparison of consumption. In the tests carried out, the generation of 1,000 inferences means consumption ranges between 0.06 and 2.9 kWh, with the average consumption being 1.35 kWh. The researchers then highlighted how, on average, a smartphone uses 0.012 kWh of energy to charge. What does that mean?

Charging your phone will give you nine pictures. With these numbers, the study concluded that creating nine images with these AI models is the same as fully charging a phone. This data puts the problem of perception into perspective because the normal thing in these processes of using generative image AI is to constantly generate several images until we see what we are looking for.

Read Also:  JPMorgan raises US economic growth forecast and no longer forecasts recession for 2023

And there are bad models. Although the average is that, there are models that are more energy-hungry, and according to the study, the worst of all consumes 11.49 kWh per 1,000 inferences, which means that each image costs the same energy as recharging the mobile phone once.

And the emissions? Again, the underlying problem is also the CO2 emissions, which are remarkable: on average, each image produced generates between 0.1 and 0.5 grams of carbon dioxide. In the most polluting study models, the figure is scandalous, and each image causes emissions of 2 grams of CO2.

It is not complete, but it gives clues. Those responsible for the study themselves emphasize that the debates on energy consumption and emissions focus on the training phase, but the next phase of use should be taken into account. They hope that this data will serve—even acknowledging that it is incomplete—to implement measures, good practices, and regulations that will help alleviate the problem.

Read Also:  Home Depot offers 10 thousand pesos discount on Mabe brand washer and dryer combo