Written by: John Fourness
In a matter of years artificial intelligence has gone from a futuristic concept you only saw in sci-fi movies to an everyday tool. AI tools like ChatGPT and Claude are now being used in classrooms, homes, and offices, and we are seeing these tools get progressively integrated into the workflow of companies. In this article, we will discuss what artificial intelligence is, how large language models work behind the scenes, the source of their information, and why their growing use creates not only opportunities, but also risks. In the US alone, it was reported in 2025 that 40% of employees report using AI at work, which is a 20% increase from 2023 (Appel et al., 2025). In an observational study conducted by Anthropic using Claude data, it was found that AI can speed up some tasks by 80% (Huang et al., 2026). Some of the more common tasks that AI is used for are drafting emails, writing and debugging code, creating schedules, and automating repetitive work. Datareportal estimates that over 1 billion people use AI tools each month (Kemp, 2026). This goes to show how quickly and easily these tools can enter everyday life.
However, the vast majority of users use AI without understanding how it works behind the scenes. A major reason for this is the fact that AI was not developed in a traditional way. Software is generally created by human programmers who give computers explicit instructions. This was not the case for AI. ChatGPT was built on a neural network and was trained on a dataset of billions of words. Therefore, no one on Earth fully understands what really goes on with LLMs. Researchers have recently been trying to get a better grasp of the situation; however, this process will take years (Lee, Trott, 2023). The basic idea of an LLM is pattern recognition and prediction. Instead of storing every answer to every possible question the models store and learn patterns they observed from massive amounts of data. When the model is being trained, it repeatedly looks at text and practices predicting what word or expression is most likely to come next. If the prediction is incorrect, the model adjusts its internal numerical settings, which are known as weights. Over time the weights become more and more accurate and are able to help the model recognize patterns in grammar, topics, tone, and common sequences of thoughts or ideas. What is currently known about AI is that it uses long lists of numbers, called word vectors, to represent various words in the human language. Each of these word vectors represents a coordinate, and words with similar meanings are then placed closer together. When a user enters a prompt, the LLM will break the prompt down into smaller pieces known as tokens. Those tokens will then be converted into numerical vectors and compared to the surrounding context in order to predict a likely response token by token. LLMs use this to generate responses by predicting the next likely word incrementally. However, since these vectors are built from the way humans use words, they often end up reflecting many biases that are present in the human language (Lee, Trott, 2023). Through this process, LLM responses are able to sound natural due to its ability to predict common continuations in language.
This can also help us understand where AI gets its information from. According to Martech, LLMs index websites and then rank the information; however, LLMs will then present the information as an answer instead of a list of where they found the information. Many LLMs also rely on different indexing strategies. For example, Gemini indexes primarily on Reddit, YouTube, and Amazon, whereas ChatGPT indexes primarily on Google (Pastore, 2025). In order to index such large datasets, LLMs use vast amounts of resources. LLMs are highly dependent on massive data centers, which consume insane amounts of resources. According to the Lincoln Institute of Land Policy, even a mid-sized data center can consume as much water as a small town, and certain larger data centers require up to 5 million gallons of water every day. On top of this, in states across the US, lawmakers are encouraging the creation of data centers with tax breaks and other incentives (Gorey, 2025).
Another big reason why AI continues to become more and more widespread is because of its accessibility and what it allows individuals to do on their own time. Anthropic conducted an interview on how people view AI with over 80k people across 159 countries. The most common hope people expressed was to gain professional excellence through AI. This means that they wanted to use AI to handle routine work in order to focus on higher-level tasks. Simultaneously, many participants also expressed that they viewed AI as an entrepreneurial partner that could help them scale and build businesses (Huang et al., 2026). This helps us understand why AI has become so prominent in side projects. Being able to use AI to walk you through not only an idea but also its implementation lowers the barriers between a concept and an actual design.
We can therefore see that AI opens up a lot of opportunities for many users. However, with AI becoming more and more widespread, there are certain risks. Aside from the major infrastructure and resource costs of data centers, there are many concerns about accuracy and labor. As we have previously mentioned, AI trains itself off of information on the internet and therefore retains certain biases or possibly incorrect information. This is why we can sometimes see LLMs giving different responses to the same prompt, since all the LLMs are doing is predicting; when several predictions seem plausible, responses can vary. In terms of labor, according to Anthropic’s March 2026 research, computer programmers, customer service representatives, data entry workers, and medical record specialists are some of the professions most exposed to AI (Cerrulo, 2026). AI, therefore, is not only reshaping tasks and efficiency but also career paths.
In conclusion, artificial intelligence can best be understood as a powerful system that is made to learn patterns from huge amounts of data. Through these patterns, it is capable of making predictions, language, code, images, and many other things. This creates real possibilities and can increase efficiency in many fields. Simultaneously, AI comes with major energy costs, the risk of misinformation, and a newfound pressure on aspiring workers. As AI continues to be integrated into the professional world as well as everyday life, the question is not what AI can do, but rather what society will choose to do with it, how it will be regulated, and how to respond to the opportunities and the consequences it may create.
References
Appel, R., McCrory, P., Tamkin, A., McCain, M., Neylon, T., & Stern, M. (2025, September 15). Anthropic economic index report: Uneven geographic and enterprise AI adoption. Anthropic. https://www.anthropic.com/research/anthropic-economic-index-september-2025-report
Cerullo , M. (2026, March 9). Anthropic is tracking which jobs are most exposed to AI. these 10 professions top the list. CBS News. https://www.cbsnews.com/news/anthropic-ai-jobs-most-exposed-risk
Gorey, J. (2025, October 17). Data drain: The land and water impacts of the Ai Boom. Lincoln Institute of Land Policy. https://www.lincolninst.edu/publications/land-lines-magazine/articles/land-water-impacts-data-centers/
Huang, S., Carter, S., Eaton, J., Pollack, S., Callender, D., Makagiansar, N., Gonzalez, M., Carr, S., Hong, J., Handa, K., McCain, M., Millar, T., Julapalli, M., Yun, G., Alt, A., Larrson, C., Leibrock, J., Gallivan, M., Sumers, T., … Ganguli, D. (2026, March 18). What 81,000 people want from ai. What 81,000 people want from AI \ Anthropic. https://www.anthropic.com/features/81k-interviews
Kemp, S. (2025, October 15). More than 1 billion people use AI – DataReportal – Global Digital Insights. DataReportal. https://datareportal.com/reports/digital-2026-one-billion-people-using-ai
Lee, T. B., & Trott, S. (2023, July 27). Large language models, explained with a minimum of math and jargon. https://www.understandingai.org/p/large-language-models-explained-with
Pastore, M. (2025, June 27). Where do the popular llms find their information?. MarTech. https://martech.org/where-do-the-popular-llms-find-their-information/