Summary of Generative
AI, LLMs, and Copilot
Summary
Artificial Intelligence is presented in the document as the science
of making machines think and act in ways that resemble human behavior. It is described
as an “eternal learner”: a system that observes millions of examples until it
can imitate patterns, make predictions, or generate new responses. Although AI
has existed for decades in tools such as GPS navigation and email spam filters,
it has become much more visible with the rise of generative systems that create
text, images, music, and other forms of content.
The document explains several types of AI through everyday examples.
Generative AI creates new content and is represented by tools such as ChatGPT,
Gemini, and Midjourney. Recognition AI identifies objects, faces, and places,
as in facial unlocking systems or augmented reality games. Recommendation AI
filters information according to a user’s preferences, while predictive AI
anticipates results based on past data, such as weather forecasts or bank fraud
detection. Together, these examples show that AI is not a single tool, but a
broad set of technologies used to observe, classify, predict, and create.
Current uses of AI are illustrated through cases that show how
people often help train these systems without noticing it. In Pokémon GO, for
example, players’ camera movements and location-based interactions help
artificial intelligence understand real-world spaces, objects, depth, and
surfaces. CAPTCHA systems are also described as a kind of classroom where
humans teach AI to recognize traffic lights, crosswalks, and road signs. Social
media algorithms are another example: they learn from small user signals, such
as watching a video until the end, skipping it quickly, or giving it a like.
These algorithms build mathematical profiles of user interests and prioritize
content that keeps people engaged, even if that content is not necessarily the
most truthful or educational.
Large Language Models, or LLMs, are introduced as the technology
behind tools such as ChatGPT and similar assistants. Their main task is to
predict the next word in a sequence. Although this sounds simple, it allows
them to produce coherent sentences, paragraphs, explanations, and complete
texts. The document explains that LLMs are first trained with huge amounts of
text through pretraining and then improved through feedback from humans or
other AI systems, a process called refinement. It also identifies the
transformer as the internal structure that allows these models to process
language efficiently. The term GPT is explained as Generative Pretrained
Transformer.
A central idea in the document is that an LLM should not be confused
with a search engine or a source of truth. It does not simply consult a
database of verified facts. Instead, it generates probable language by
predicting what words are likely to fit together. Because its priority is
fluency and coherence, not truth, it may produce answers that sound convincing
but are false. These errors are called hallucinations. They can include
invented dates, books, laws, authors, or citations that appear real. For that
reason, the document emphasizes that AI-generated content must always be
checked carefully.
The role of the teacher or professional user is presented as
essential. The AI is compared to a copilot that can generate drafts, ideas, and
initial text, while the human remains the captain who validates, corrects, and
decides what is useful. The document highlights the difference between a
novice, who may not detect mistakes, and a trained professional, who can
identify historical, pedagogical, or conceptual problems because of their
knowledge and experience. This reinforces the idea that information is not the
same as knowledge: AI can provide data and text, but human judgment supplies
understanding, ethics, and context.
The document also explains basic terms related to how AI works. A
data center is the physical infrastructure where AI systems run, described as a
building full of powerful computers that give the AI the capacity to respond. A
model is the trained digital brain, such as GPT-4 or Gemini. The interface is
the screen through which the user communicates with the system, and the text
box is where instructions are entered. A prompt is the question or instruction
given to the AI, and the quality of the prompt strongly affects the quality of
the answer. Tokens are the small pieces of text that AI systems process,
similar to building blocks rather than full words or letters.
Another important concept is the context window, which works like
the AI’s short-term memory during a conversation. It determines how much
previous text the system can remember and use when answering. The document also
describes iteration and refinement as the process of improving results by
asking the AI to adjust its response. For example, if a generated exam is too
difficult, the user can ask the AI to make it simpler. This back-and-forth
process helps transform a first draft into a more useful final product.
Finally, the document provides guidance for writing effective
prompts. Before asking the AI for help, users should define their objective,
explain the context, specify the sources or references needed, and make their
expectations clear. They should indicate the desired tone, audience, format,
and purpose of the response. Courtesy and clarity are also encouraged, because
respectful and specific instructions make it easier for the AI to understand
the request. The document closes with guiding questions that help users improve
their prompts: what the text should achieve, who the audience is, what tone is
appropriate, which data or references should be included, how the response
should be organized, and what exact result is expected.
In conclusion, the document presents generative AI and LLMs as
powerful tools for learning, writing, organizing information, and supporting
teaching, but it also warns that they require critical thinking. AI can
accelerate work and provide useful first drafts, but it can also make mistakes,
invent information, or reinforce habits shaped by engagement-driven algorithms.
The most responsible approach is to use AI as an assistant, not as an
unquestionable authority. In this model, Copilot or any similar AI tool
supports the work, while the human user provides verification, professional
judgment, ethical responsibility, and final decisions.
Explanation of how we use Copilot: For the summary about AI, LLMs, and Copilot, Copilot was used. A prompt was given asking it to create a summary of the document provided and discussed in class. The summary had to be written in English, delivered in a Word document, and presented in prose form.
Reflection:
In my opinion, Artificial Intelligence can be a very useful support tool for teachers when it is used responsibly. AI can help create teaching materials, organize ideas, prepare summaries, design activities, and save time during lesson planning. If teachers know how to use it well, it can make classes more dynamic, creative, and easier to prepare.
However, AI also has important limitations. It can sometimes provide incorrect or incomplete information, and it does not fully understand the specific needs, context, or learning process of each group of students. For this reason, teachers should not depend on AI for everything. It is important to review and adapt the information, because the teacher is the one who has the professional knowledge, critical thinking, and responsibility for the class.
