Politeness with ChatGPT: why you shouldn’t save on “please”
22 April 2025 11:58
The use of polite phrases in communication with ChatGPT costs OpenAI tens of millions of dollars in electricity bills. This was announced by the company’s CEO Sam Altman, commenting on a request from a user of the social network X, "Komersant Ukrainian" reports citing techradar.
“I wonder how much money OpenAI has lost on electricity bills because of people who write ‘please’ and ‘thank you’ to their models.”
– this message from @tomiinlove started a discussion in which Altman appeared. The company’s CEO answered this question, immediately noting that all the money spent was worth it.
“Tens of millions of dollars well spent – you never know what can happen,”
– Altman replied, hinting at the importance of maintaining ethical communication with artificial intelligence.
According to a survey conducted by Future PLC in February of this year, about 70% of users are polite when communicating with AI. Interestingly, 12% of respondents admitted that they do so because of fears of a possible “robot uprising” in the future.
Industry experts point out that every message sent to ChatGPT consumes energy of powerful servers that ensure the system’s operation. However, most users do not even think about the environmental consequences of even simple requests to AI.
Should we be polite to AI?
Experts assume that the significant energy costs mentioned by Altman are related to individual messages that contain only the words “please” or “thank you,” rather than polite wording within the main request.
A study conducted by TechRadar journalist Becca Caddy showed that being polite when communicating with AI can even improve the quality of responses.
“Polite, well-structured queries often lead to better answers and in some cases can even reduce bias. This is not just an advantage – it is a critical factor in AI reliability,”
– she noted in her article.
With the development of artificial intelligence technologies, experts predict that politeness may become a built-in feature of AI systems. It is possible that models can be configured to provide better responses to users who communicate respectfully.
So, although polite communication with AI may be less energy efficient, it potentially improves the ChatGPT experience.
Читайте нас у Telegram: головні новини коротко
Artificial intelligence: background
Artificial intelligence (AI) is a branch of computer science that deals with the creation of systems that can perform tasks that normally require human intelligence. This includes functions such as image, speech, and text recognition, decision making, language translation, and information generalization. Modern AI systems are based mainly on machine learning methods, especially deep learning using neural networks that are trained on large data sets.
In recent years, artificial intelligence has become an integral part of many areas of life, from personal assistants in smartphones to medical data analysis systems and autopilots in transportation. AI can be divided into “narrow” (or specialized) AI, which is created to solve specific tasks, and “general” AI, a theoretical concept of a system that could perform any human-level intellectual tasks. Currently, all existing AI systems belong to the category of narrow artificial intelligence.
The development of artificial intelligence poses both enormous opportunities and serious challenges for humanity. On the one hand, AI can help solve complex problems in medicine, science, ecology, and other industries. On the other hand, there are ethical, security, and socioeconomic issues related to labor automation and the possible use of AI for harmful purposes. Therefore, the development of artificial intelligence is accompanied by efforts to create appropriate regulatory frameworks and ethical principles.
What is LLM (Large Language Model)?
LLM (Large Language Model) is a type of artificial intelligence that refers to large language models trained on huge amounts of textual data to understand, generate, and process human speech. These models use transformer architecture and billions or even trillions of parameters to analyze context and generate relevant answers. Modern LLMs, such as GPT (by OpenAI), Llama (by Meta), Claude (by Anthropic), and others, can write texts, answer questions, summarize information, translate between languages, and perform many other tasks related to natural language processing.
The LLM training process includes a pre-training stage, during which the model processes huge amounts of texts from the Internet, books, articles, and other sources, learning the statistical patterns of language and accumulating knowledge about the world. After that, many models undergo a fine-tuning phase using reinforcement learning with human feedback (RLHF) to make them more useful, accurate, safe, and aligned with human values and needs.
With the advancement of technology, modern LLMs have evolved from simple text-based models to multimodal systems that can work not only with text but also with images, audio, video, and other types of data. This expands their capabilities and allows them to be used for content creation, programming, data analysis, business process automation, education, entertainment, and many other industries. Despite their impressive capabilities, LLMs have limitations, including the possibility of hallucination (giving out false information), bias, dependence on the quality of training data, and ethical challenges associated with their use.
Читайте нас у Telegram: головні новини коротко
