AI is not a living being.
A program designed by people builds sentences by calculating the probability of the next words (tokens), from patterns it learned and the current context.
Sign up to keep track of your learning progress.
Log in
A space to learn at your own pace
Your learning journey,
Your learning journey,
with Lyen.
New to AI? You belong here. Explore the basics, enjoy creating, and learn one step at a time.
Everyone is equal before AI
Your first step today
All guides
Understand. Create. Check.
Not sure where to begin? Explore these steps in order.
AI is a fill-in-the-blanks champion!
Here, we mean language models: the kind of AI you chat with in writing. Let's explore how they produce a reply using an everyday sentence.
What would you put in the blank? Tap an example.
This morning, I __________. This morning, I ate breakfast. This morning, I went for a walk. This morning, I read a book.
These choices are prepared examples. Tap your selected example again to bring back the blank.
Does AI really make replies this way?
Yes, predicting what comes next is the basic idea! A language model uses patterns learned during training, your input, and the text so far to calculate what could come next. It chooses a small piece of text called a token, which can be a word or part of one, then repeats the process to build a reply.
Filling a blank is an analogy. In a chat reply, the model usually keeps adding tokens at the end. It doesn't always choose the most likely one: it can also sample from several candidates according to their probabilities.
Then what does “thinking” or “reasoning” mean?
People use these words when AI works through a problem in steps. Underneath, it runs calculations using learned patterns and its input. Fluent language alone is no reason to treat it as a person with feelings or intentions.
A plausible continuation can still be wrong. Check important answers.
What AI really is
Understand AI. Check its answers.
How a language model produces an answer
Generation uses context and learned weights.
A neural language model learns numerical weights from training examples. At generation time, those weights and the available context determine scores for candidate tokens. The model does not simply look up a stored internet sentence or count only the word immediately before it.
Always remember this.
What you are reading is not the thought of a living being, but a result made of words likely to fit the context. Plausible does not mean true.
PES LEARNING MAP
What would you like to learn?
Starting with AI, every subject tailored to you.
My Learning
Track your skills, understanding, and progress with data, and get a personalized learning path.
Learn Python
Learn Python in 9 steps, coding right in your browser — it runs as you type, no install needed.
AI
From prompt engineering to building your own AI — easy and fun.
Content Studio
Anyone can create and share learning content.
01 /
Absolutely easy
You can start with zero background. Complexity is a flaw of the system, not your fault.
02 /
Absolutely fun
Gamification, 3D tutors, and visuals make learning feel like play. Boredom is also a defect the system must fix.
03 /
Personalized for you
Dashboards, chatbots, stats, and progress adapt to you — learn at your own pace.