Hey everyone! I recently heard Sam Altman mention that polite phrases like 'please' and 'thank you' can lead to significant computing costs for OpenAI. This made me wonder if there are ways to make AI operations more sustainable without just focusing on using greener energy sources.
For instance, instead of a lengthy explanation for something like the weight of a blue whale, shouldn't the answer just be a straightforward '1,500,000,000 pounds'? I'm curious if there's potential for a service that could efficiently shorten prompts and direct queries to more efficient models without compromising on politeness.
Is anyone aware of ongoing developments in this area? Or if any services already exist that take this approach? Thanks a lot for your insights! 🙂
3 Answers
As a user, you actually have the power to guide how verbose or concise the model’s answers are. Just frame your questions to specify whether you want a brief answer or a detailed one!
Local AI options can be greener, since they don’t rely on massive data centers and the energy costs associated with them. If you are concerned about energy efficiency, exploring local models might be a good route.
While the length of a response doesn’t directly dictate energy use, what actually matters is the complexity of the associations made during processing. GPUs handle different loads depending on the task. You can’t effectively gauge energy use without access to the data from the major providers. Plus, if prompts are processed in batches, your individual prompt won’t significantly affect the overall GPU usage. It’s all about the deeper processing involved.
Related Questions
xAI Grok Token Calculator
DeepSeek Token Calculator
Google Gemini Token Calculator
Meta LLaMA Token Calculator
OpenAI Token Calculator