CTO l Building Responsible AI solutions | Speaker I ML research I Networking technologies I Digital solutions | Sustainability | Board Member | Advisor | Mental Health Advocate
Here is a #lightbulb moment: I’ve seen some videos by “experts” claiming that a #chatGPT query consumes 10X more energy than a #Google search. Or as much as a light bulb being lit for 10 min. With OpenAI’s announcement of GPTsearch yesterday to compete with Perplexity and Google’s #Gemini, I think it’s worth examining whether this is true. A typical incandescent light bulb: ~60 watts (0.06 kW) Energy (kWh) = Power (kW) × Time (hours) For 10 minutes (which is 1/6 of an hour): 0.06 kW x 1/6 hrs = 0.01 kWh A typical LED light bulb: ~10 watts (0.01 kW) LED: 0.01 kW X 1/6 hrs = 0.00167 kWh ➡️ A typical chatGPT query consumes 0.0004kw, which is less than a light bulb burning for 10 min. But when you compare it to Google search, the claim of a chatGPT query taking 10 times more energy is more of an outlier. I suppose it is plausible when you consider extreme cases of higher estimates of GpT queries and lower estimates of Google search. As the graph below shows, the more typical and realistic cases show that a ChatGPT query consumes about 4x more energy than a Google search. That is not to say that #AI models are not a huge energy hog given the nature and quantum of queries, that is already becoming an environmental nightmare. #GenAI models consume more energy due to the intensive computations required by their large neural network models during inference driving higher cooling requirements in the data centers. In contrast, #Google #Search is more energy-efficient, leveraging decades of optimization, pre-built indexing, and highly efficient algorithms to quickly retrieve and rank search results. #searchwars #sustainability #energy
𝗔𝗜 𝗧𝗿𝗮𝗻𝘀𝗳𝗼𝗿𝗺𝗮𝘁𝗶𝗼𝗻 𝗘𝘅𝗽𝗲𝗿𝘁 * 𝗕𝗲𝘀𝗽𝗼𝗸𝗲 𝗔𝗜 𝗗𝗲𝘃𝗲𝗹𝗼𝗽𝗺𝗲𝗻𝘁 𝗣𝗮𝗿𝘁𝗻𝗲𝗿 𝗳𝗼𝗿 𝗙𝗼𝘂𝗻𝗱𝗲𝗿𝘀 & 𝗦𝗠𝗕 * 𝗦𝗲𝗿𝗶𝗮𝗹 𝗘𝗻𝘁𝗿𝗲𝗽𝗿𝗲𝗻𝗲𝘂𝗿 * 𝗜𝗻𝘃𝗲𝘀𝘁𝗼𝗿 * 𝗦𝗽𝗲𝗮𝗸𝗲𝗿
5moVery informative, thanks for clarifying.