AI isn't getting smarter, it's getting more power hungry - and expensive
Image generated by Gemini AI
MIT researchers have quantified the computing power required for 809 large language models, revealing significant resource demands that could strain data centers. The study highlights a trend towards increasing costs associated with AI model training, emphasizing the need for more energy-efficient computing solutions to sustain growth in AI applications.
AI's Growing Demand for Computing Power Drives Up Costs
Recent analysis from MIT highlights a troubling trend in the development of large language models (LLMs): a significant increase in the computational power required for training these systems is leading to higher operational costs. The research indicates that the computing power necessary for 809 of the largest language models has surged dramatically, raising concerns about the sustainability of AI advancements.
The average computational demand for training LLMs has increased by more than 10 times over the past two years, driven by the complexity of the models and the scale at which they operate. Training a single LLM required approximately 1.5 million GPU hours in 2020; by 2023, this figure has risen to around 15 million GPU hours.
Additionally, MIT researchers noted that the energy consumption associated with these models is climbing at an alarming rate, potentially contributing significantly to carbon emissions if sustainable practices are not implemented. This is concerning as the AI industry seeks to balance innovation with environmental responsibility.
These findings highlight the financial burden of powering AI systems, which could reshape business models and influence the accessibility of AI technologies across different sectors.
Related Topics:
📰 Original Source: https://www.zdnet.com/article/ai-isnt-getting-smarter-its-getting-more-expensive-mit-report-finds/
All rights and credit belong to the original publisher.