The potential for artificial intelligence is huge – but its equally great energy consumption needs to limit themselves to asking shorter questions a way of achieving, said a UNESCO survey revealed Tuesday.
A combination of shorter queries and using more specific models could reduce AI energy consumption by up to 90% without sacrificing performance, UNESCO said in a report published to mark AI for a good global summit in Geneva.
Openai CEO Sam Altman recently revealed that each request sent to its popular Generative AI -App Chatgpt uses an average of 0.34 wh of electricity, which is between 10 and 70 times a Google search.
With Chatgpt receiving about a billion requests a day, it is 310 GWh annually, corresponding to the annual electricity consumption of three million people in Ethiopia, for example.
Furthermore, UNESCO calculated that the AI energy needs are doubled every 100 days as generative AI tools are embedded in everyday life.
“The exponential growth in calculation power needed to run these models puts increasing impact on global energy systems, water resources and critical minerals, raising concerns about environmental sustainability, fair access and competition for limited resources,” warned the UNESCO report.
However, it was able to achieve an almost 90% reduction in electricity consumption by reducing the length of its query or prompt as well as using a smaller AI without a decrease in performance.
Many AI models like Chatgpt are general models designed to respond to a wide range of topics, which means they have to silence through a huge amount of information to formulate and evaluate answers.
The use of smaller, specialized AI models offers large reductions in electricity needed to produce an answer.
So did the cutting prints from 300 to 150 words.
When you are already aware of the energy issue, tech giants now offer all miniature versions with fewer parameters for their respective large language models.
E.g. Seller Google Gemma, Microsoft has Phi-3 and Openai has the GPT-4o Mini. French AI companies have done the same; E.g. Has Mistral AI introduced his model, minister.
