How a language model continues text

Where we are on the map

Lesson 68 of 72. AI and project release block (63–72). The assistant keeps its earlier reminder rules and local database; the new feature only suggests a task category and can abstain.

Situation and question

A phone suggests the next word. A large language model does a far more complex version over a longer context, yet a fluent sentence can still be wrong.

New words without gaps

A token is a piece of text processed by a model; it is not always a whole word. Context is the part of the conversation and instructions available to the model. Training on large corpora fits parameters for predicting continuation. Generation selects further tokens in steps. A prompt is user input. The model does not automatically gain a verified source of truth.

The lesson’s support signal

Read the three cells left to right, then cover the third. State the input and action in each cell; predict how the path ends and which observable fact checks it. Compare the drawing with a project file or worked example. If a number on the map cannot be derived from the data, correct the map rather than changing data for a pretty picture.

Work through it step by step

Write two completions of “Tomorrow we have…”: “class” and “a holiday”. Both are grammatical; the timetable decides which is true. Compare with our step-07 classifier: it returns only math, reading or review, while a generative model creates free text. Do not send someone else’s tasks or details to an outside service without permission.

Predict and check

Write the expected output before running code. Execute it in step-06 or step-07, compare every character and explain each line. Then change one safe fictional input and repeat “predict — run — change — explain — verify.” Do not turn an observation on a tiny dataset into a promise about real people.

from collections import Counter
next_words = Counter(("class", "holiday"))
print(next_words["class"], next_words["holiday"])

Expected output

1 1

Catch the error

A polite, coherent answer is no guarantee of fact. “Large” does not grant a model authority to decide for a person. Generating text and locating a current official document are different operations.

Project change

Release 2.1 preserves the DB, report, reminders and backup. labelled_tasks.csv contains fictional examples only. classifier.py suggests math, reading or review but writes no decision to the database. Check each model output against held-out examples and let a person reconsider it.

Task and evidence

For three statements separate “plausible” from “verified.” Name the document or observation needed for each check.

Transfer to a new setting

A model gives a convincing date for a school competition. What should a family verify before buying tickets?

Return after 1, 7, and 30 days

After 1 day redraw the map from memory and name one boundary of the idea. After 7 days explain it to a friend without reading the lesson and answer 7 of 10 block questions; analyse mistakes with concrete inputs. After 30 days open a fresh project copy, reproduce the check and transfer the idea to a new case. Keep a record of input, expected output and actual output.

Primary reference to check

Official reference

Next lesson

A confident answer is not a reliable answer