But what about video games? (blog.glyph.im)

A recent discussion highlights the energy and resource implications of using local large language models (LLMs) versus playing video games. While some argue that running a local model for coding is comparable to gaming, the analysis points out critical differences: video games are interactive and consume resources mainly during play, whereas LLMs are intensive batch processes that can run continuously for extended periods, consuming significantly more energy. For example, developing a simple application using an LLM took three weeks of continuous processing, underscoring the disparity in energy consumption. Moreover, the training of LLMs requires vast amounts of power, far exceeding what is typically needed for video game development. Even local models depend on previous training done by large-scale providers, perpetuating the energy demands of the AI ecosystem. While local LLMs could offer some advantages, such as reduced privacy concerns, they are not a direct substitute for video gaming in terms of energy efficiency or resource usage. This conversation is essential for the AI/ML community as it emphasizes a deeper understanding of the ecological impact of AI technologies and the need for responsible AI usage.