The Cost of Cutting-Edge Generative AI according to Qualcomm | Midjourney ai | Bing Image Creator Free | Come Usare Midjourney | Turtles AI

The Cost of Cutting-Edge Generative AI according to Qualcomm
DukeRem14 agosto 2023
  A new report from Qualcomm analyzes generative AI trends - costs dropping 90% in 2 years as hardware improves, but more parameters need exponentially more data and computing. The report examines the substantial compute, parameters and costs needed to develop advanced generative AI models like GPT-3. Specialized chips like TPUs can help reduce expenses. The rapid advancement of generative AI models has led to increased interest and adoption from businesses and consumers. However, these powerful systems require substantial computational resources to develop and deploy. A new report examines the costs and hardware needs associated with state-of-the-art generative AI. According to the report, developing a generative AI model on the scale of GPT-3 requires thousands of petaflops of compute and billions of parameters. The expense to train such a system can exceed $10 million. Ongoing costs are also considerable, with each query to GPT-3 estimated to cost $0.002. The report notes that specialized AI accelerators like TPUs can help reduce training times and costs. For example, training Meena, Google's conversational chatbot, took only a few weeks using TPUs compared to thousands of years on GPUs. Dedicated AI chips allow for more efficient parallel processing and optimized performance. As generative AI continues to advance, even more powerful models will emerge. The report forecasts models reaching trillions of parameters over the next decade, enabling new possibilities in content creation, reasoning, and personalization. But realizing the potential of generative AI will require innovations in specialized hardware, scaling methodologies, and cost optimization. Highlights: - Generative AI models like GPT-3 require thousands of petaflops of compute and billions of parameters - Training costs for large models can exceed $10 million - Queries to GPT-3 estimated to cost $0.002 each - Specialized chips like TPUs reduce training times and costs - Models forecast to reach trillions of parameters in coming years This report provides valuable insight into the resources needed to push generative AI forward. As these systems become more powerful and accessible, we must also consider the implications for security, ethics and fairness. What policies and safeguards should be put in place as generative AI is integrated into business and society? I invite readers to share their perspectives.