GlobSynk
GlobSynk Academy™
Module 3 of 9~33 minutes

Giving Context & Memory (conceptual)

Context is what the assistant could not work out for itself. Supplying the right amount is a discipline: more personal detail is not automatically better, and anything remembered can go stale.

Objectives

  1. Explain in plain words what context is, and why an assistant cannot supply it for itself.
  2. Tell a preference, a constraint, and a piece of background apart in their own writing.
  3. Identify the information a task genuinely depends on.
  4. Decide which context should persist and which belongs to one task only.
  5. Explain why more personal context is not automatically better, and name two costs.
  6. Recognize when stored context has gone stale and needs review.
  7. Explain why remembered information is not the same as verified information.

Introduction

An assistant with a clear role and no context produces confident, generic work. It is the most common failure in everyday use, and the most misdiagnosed: people conclude the tool is weak when the tool was simply never told anything only they knew.

This module is about supplying that — and about the discipline that has to come with it. Context is the one input where more genuinely feels safer and usually is not.

Every extra detail is something you have shared, something that can go stale, and something that can quietly steer an answer in a direction you did not choose.

By the end of this module you will be able to supply what a task genuinely depends on, decide what should persist, and recognize when something remembered has stopped being true.

Lesson 1

What Context Actually Is

Context is what the assistant would have to be told because it could not work it out. Separate it cleanly from the instruction, which is the work itself.

Lesson 2

Preferences, Constraints, and Background

Three different kinds of context that behave differently. A preference shapes the result; a constraint rules options out; background explains the situation.

Lesson 3

What Should Persist, and What Should Not

The conceptual heart of "memory": some context is true every time, and some is true only today. Confusing them is what makes assistants feel oddly wrong later.

Lesson 4

More Context Is Not Better Context

Teach selection over accumulation, with the two real costs: the detail that matters gets buried, and information you never needed to share has been shared.

Lesson 5

Stored Context Goes Stale

A job changes, a class ends, a preference shifts. Anything remembered needs a review habit, because nothing tells you when it stopped being true.

Lesson 6

Remembered Is Not Verified

The closing principle: an assistant repeating something back accurately is evidence it stored it, and evidence of nothing else.

Three kinds of context, kept apart

These behave differently, and mixing them is why some context helps and some quietly distorts.

Three kinds of context, kept apart

A preference you can ignore and still be correct. A constraint you cannot. Most of what learners supply turns out to be background.

Three kinds of context distinguished by what each one does to the result. A preference shapes an acceptable result — short paragraphs rather than bullet lists — and can be ignored while the result is still correct. A constraint rules options out — it must fit on one page — and cannot be ignored. Background explains the situation, such as who the reader is. Mixing them is why some context helps and some quietly distorts; learners who label their own context this way usually find most of it was background, the kind that feels helpful and changes nothing.

Preference

Shapes an acceptable result.

  • “Short paragraphs, no bullet lists.”
  • Ignore it and the result can still be correct

Constraint

Rules options out.

  • “It must fit on one page.”
  • Ignore it and the result is wrong

Background

Explains the situation.

  • “This goes to people who missed the meeting.”
  • Often the largest share — and the one that changes leastIt feels helpful. Check whether it actually changed anything.
  • Preference — shapes an acceptable result. "Short paragraphs, no bullet lists."
  • Constraint — rules options out. "It must fit on one page."
  • Background — explains the situation. "This goes to people who missed the meeting."

A preference you can ignore and still be correct. A constraint you cannot. Learners who label their own context this way usually discover that most of what they supplied was background — the kind that feels helpful and changes nothing.

The context filter

Run each piece of context through this before supplying it. Read it as a sequence of questions; the first no ends it.

Does the task depend on it? → Could the assistant work it out without me? → Is it mine to share? → Will it still be true next time?

  • Fails the first — leave it out; it is background, not context.
  • Fails the second — leave it out; you are telling it something it already handles.
  • Fails the third — leave it out, and consider whether the task itself needs changing.
  • Fails the fourth — supply it for this task only. Do not let it persist.

Practical exercise

≈7 min

The minimum that actually works. Take one real task you would use an assistant for this week and write down everything you think it needs to know. Do not filter yet — get it all down.

Label each item preference, constraint, or background. Then cross out every piece of background that does not change what a good result looks like. Most people cross out half.

Mark what is left as persists or this task only. Then look at your "persists" list and ask the hardest question: if this were still being applied in six months, would it still be true?

Nothing is submitted or stored. If you would rather not use your own task, the supplied worked example below teaches the same filtering skill.

One worked task with eleven candidate pieces of context is supplied below, drawn from an ordinary situation — organizing a community event. Label each, filter, and mark persistence, then compare against the supplied reasoning. Three of the eleven are deliberately borderline, and the answer says so: the point is not a score but seeing the same piece of information argued both ways and having to hold the judgment yourself.

Your progress

0 of 2 required activities complete in this module · course progress 0%

  • ○ GlobSynk Lab™ · optional
  • ○ Reflection
  • ○ Checkpoint

GlobSynk Lab™

optional, ≈5 min

About GlobSynk Lab™. GlobSynk Lab™ is the hands-on practice experience used throughout GlobSynk Academy. This Lab is optional hands-on practice: complete it now, skip it and continue the module, or return to it later. Skipping this Lab does not prevent you from continuing the course.

Take your filtered context and give an assistant the task twice: once with your full original list, once with only what survived the filter.

Compare the two results honestly. Most learners find them equivalent, and some find the shorter one better. Then remove one more item from the short version and try again — the goal is to find the point where the answer actually degrades, because that tells you what was doing the work.

Reflection

≈2 min

Think of something a person or a service remembered about you that had stopped being true. How did you find out? What would have caught it earlier — and what would that habit look like applied to an assistant that remembers your preferences?

This reflection is yours alone — it is never sent to GlobSynk or stored. Only the fact that you completed it is saved.

Checkpoint

5 questions · unscored gate · instant feedback · retry as often as you like.

1. A prompt contains four sentences. Which one is context rather than instruction?
2. "It must be under 300 words" and "I prefer a warm tone." Which one can be ignored while the result is still correct?
3. A learner pastes an entire group chat to ask one scheduling question. What are the two costs?
4. Which piece of context should NOT persist between sessions?
5. An assistant correctly recalls a date you told it last month. What does that establish?

Answer all 5 questions to continue.

Key takeaways

  • Context is what the assistant could not work out for itself — separate from the instruction, which is the work.
  • Preferences shape a result, constraints rule options out, and background explains the situation.
  • More context is not better context: the detail that matters gets buried, and you share more than you needed to.
  • Some context belongs to one task only; letting it persist is what makes an assistant feel wrong later.
  • Remembered is not verified — accurate recall says nothing about whether the fact is true.

Practice in Prompt Lab — optional

Optional

Practice giving only the context a task actually needs.

In Prompt Lab, give an assistant a task with the minimum context you think it needs. Then remove one piece and run it again. If nothing meaningful changed, that piece was never carrying the answer. Repeat until something does change — that is the boundary of what the task genuinely depends on, and it is almost always smaller than it feels.

Optional, and never required. Prompt Lab is not GlobSynk Lab™. Opening it is not part of module completion, the checkpoint, course progress, the Final Assessment, the certificate or Reward Points, and a learner who never opens it completes this course exactly as normal. Academy never depends on Prompt Lab being reachable.

Practice Prompting in the Real World

Opens in a new tab. Optional practice — never required for this module, the checkpoint, your progress, the Final Assessment, the certificate or Reward Points.

Before you move on

You can now give an assistant enough to be useful without giving it more than it needs — and you know which of that should stick around.

Doing this once, for one task, is worth the effort. Doing it from scratch every Monday for the same task is not.

What changes when a task stops being a request and becomes something you do every week is Module 4.