Tech Won't Save Us Patreon Preview: The Harms of Generative AI w/ Alex Hanna
Feb 17, 2025
Delve into the intricate world of generative AI as experts unpack the computational demands that drive these models. Discover how crucial specialized hardware, like NVIDIA GPUs, supports the complex calculations involved. The discussion critiques the rising costs and energy challenges associated with generative AI, painting a stark picture of sustainability as usage escalates. This eye-opening conversation reveals the hidden infrastructure strains that accompany the fascination with AI technology.
AI Snips
Chapters
Transcript
Episode notes
Computational Intensity of Generative AI
- Generative AI models are computationally intensive due to their large size and the training process.
- The training involves backpropagation, matrix multiplication, and specialized hardware like GPUs.
Inference Costs and User Growth
- Initial focus was on training costs, but inference costs are significant with increased user bases.
- Google's optimistic predictions about plateauing costs overlooked the widespread adoption and continuous inference.