It is often said that “good manners cost nothing,” but in the modern world, this isn’t strictly true. A simple ‘please’ and ‘thank you’ can actually be pretty expensive in the age of artificial intelligence (AI). Earlier this year, an X user (@tomieinlove) mused: “I wonder how much OpenAI has lost in electricity costs from people saying ‘please’ and ‘thank you’ to their models.” OpenAi’s CEO, Sam Altman, quickly replied, “Tens of millions of dollars well spent.”
While Altman’s reply may have been tongue in cheek, it is undeniable that gratuitous language consumes additional AI processing power, which, in turn, costs money. Training advanced language models to understand and respond politely, contextually, and helpfully doesn’t come cheap. Billions of parameters need to be fine-tuned to handle tone, empathy, and nuance, which consumes vast amounts of computing power and energy. To provide some context, according to ChatGPT, training GPT-4 (the fourth-generation large language model in OpenAI’s GPT series) cost over US$100 million in computing resources alone, and fine-tuning for specific capabilities like politeness costs millions more. Running the model every day requires huge ongoing cloud infrastructure costs, which are also likely to be hundreds of millions per year. Back in 2023, when OpenAI was running its earlier GPT-3 model, Dylan Patel, Chief Analyst at semiconductor research firm SemiAnalysis, estimated that ChatGPT could be costing OpenAI over US$700 000/day to run.1
In terms of energy use, training a large language model can consume as much energy as hundreds of US households use in a year. It has been reported that training GPT-3 used approximately 1287 MWh of electricity and emitted around 550 t of CO2. Tech giants are also consuming vast amounts of water to keep their AI data centres cool. Researchers believe that Chat GPT gulps up approximately 16 oz of water every time you ask it a series of between 5 to 50 questions.2
Yet despite these costs – financial, environmental, and infrastructural – AI offers extraordinary, transformative benefits for the downstream oil and gas sector. As AVEVA explains in its article starting on p. 19 of this issue, the integration of digital technologies – particularly advanced analytics and AI – can transform the vast amounts of data gathered in the sector into deeper insights and real business value. It can optimise operations, enhance safety and compliance, drive innovation, and support improved decision-making. From predictive maintenance that minimises unplanned shutdowns, to real-time process optimisation that improves throughput and reduces emissions, AI enhances operational control and responsiveness across highly complex industrial systems. It can support advanced digital twin models, automate routine tasks, support regulatory compliance, strengthen cybersecurity, and much more.
Just one example of AI’s practical applications in the downstream sector is provided by Clariant and Navigance on p. 15 of this issue. Felicitas Cokoja, Catherine Basilides Schwarz, and Lisa Krumpholz explore how next-generation AI tools are being used to optimise catalyst performance, thereby enhancing operational efficiency and safety of chemical production plants.
So while a polite AI might cost millions to train and run, the return on investment for downstream oil and gas is clear: smarter, safer, and more sustainable operations. In this context, a little computational courtesy is, perhaps, a price worth paying.
- https://africa.businessinsider.com/news/chatgpt-could-cost-over-dollar700000-per-day-to-operate-microsoft-is-reportedly/b64jzzq
- https://apnews.com/article/chatgpt-gpt4-iowa-ai-water-consumption-microsoft-f551fde98083d17a7e8d904f8be822c4?utm_source=newsletter&utm_medium=email&utm_campaign=newsletter_axioslogin&stream=top
