In both cases, I will explain:

1. How to download your own data (if possible)
2. How to opt out of sharing your usage data

I will start with

• Part I: OpenAI/ ChatGPT
• Part II: Antrophic/ Claude
• Part III: Github CoPilot

Part I: OpenAI/ ChatGPT

1. To download your data, either
   a) go to https://privacy.openai.com/ and login

Thereafter, you can click on step 1: download my data

1. To download your data, the other option is to
   b) request for a data download via the web application ChatGPT https://chatgpt.com/#settings/DataControls:

This should bring you to get an email OTP via your registered OpenAI/ ChatGPT account. You should receive an email confirmation thereafter.

2. Opting out of sharing data involves clicking on the second button: do not train on my content via the same url https://privacy.openai.com/

You should receive an email confirmation thereafter that verifies that OpenAI is no longer using your data for training:

Bonus: Remove personal data from ChatGPT responses

Under certain privacy or data protection laws, such as the European General Data Protection Regulation, you may have the right to ask for certain personal data about you to be removed from ChatGPT responses.

You can append specific links to ChatGPT conversations, URL(s) containing the personal data.

This is subject to a verification and review process by the OpenAI team — still pleasantly surprised to see recourse exists.

I’ve finally opted to delete my GPTs — they aren’t actively in use beyond some initial trials and experiments.

Part II: Antrophic/ Claude

Context: how Antrophic uses your data

With your permission, we will use your chats and coding sessions to train and improve our AI models. This helps us to 1) improve our AI models to make Claude more helpful and accurate for everyone and 2) develop more robust safeguards against harmful outputs.

Anthropic may use your email for account verification, billing, and Anthropic-led communications and marketing (e.g., emails sharing new product offerings and features).

Anthropic may conduct aggregated, anonymized analysis of data to understand how people use Claude.

Anthropic may offer additional features, which will enable us to collect and use more of your data. You’ll always be in control and can turn off these features in your account settings.

Note: accessing this privacy tab only seems to work on the desktop app (did not work via the web-app at the time of writing, 5 Apr 2026, 12pm)

1. To download your data, click Export Data

You should receive this in your inbox in minutes

2. To prevent Antrophic from training on your data, toggle Help improve Claude off

Part III: Github CoPilot

1. You can download usage report but not really usage data. Tutorial here
2. By default, Github will train on your data from Apr 24 by default unless you opt out. To opt out, go to https://github.com/settings/copilot/features

Hope this helps!

Managing privacy with AI was originally published in String on Medium, where people are continuing the conversation by highlighting and responding to this story.

Using the most generic prompt ever:

Rate this lesson

I dumped the text transcript using Voice Memo on my phone.

ChatGPT was the harshest, giving 6.5/10, followed by Claude giving 7.5/10 and Gemini was the ‘nicest’ rating it 8.5/10.

What was helpful from all 3, crediting each:

• Claude: There’s also a bit of over-reliance on “you all know lah” as a comprehension check — it doesn’t surface genuine misunderstanding <- I did not notice this about the lesson or a filler I used, will pay more attention
• ChatGPT: Language precision <- this was helpful and reading it helped me to tweak precision for the subseqent classes I taught in the week
• All three: criticised structure and conceptual clarity
• All three: praised engagement, entertainment and recognized context of stirring interest as a ‘first class’

Generally, while I was somewhat skeptical about what value AI could bring to experienced educators (I’ve candidly spent 4 years in the classroom, 3 years out doing tech roles before stepping back in class) — I thought with the right framing, right prompts, this can benefit educators across the spectrum.

This is how ChatGPT responded

Thanks for sharing this so honestly. I'm going to rate this as if I'm observing you as an H1/H2 JC Economics tutor in Singapore.

---

Overall Rating: 6.5 / 10
You clearly:
Know your content well
Care about the subject
Try to make it relatable (anime, mugging, Deliveroo, overbooking, $50 vs $100)
Have strong instinct for connecting econs to real life.

But the delivery and structure are currently limiting the impact.
Let me break it down properly.

---

✅ What You Did Well
1️⃣ Strong Real-World Anchoring (8/10)
You used:
Overbooking
Deliveroo exiting SG
$50 today vs $100 next year
Incentives and studying
Anime vs mugging
R&D and AI

That's very good teaching instinct.
Especially for first lesson - students need to see:
"Oh… this subject actually explains real life."
That part is strong.

---

2️⃣ You Lowered Anxiety (Good Move)
You addressed:
"Econs won't make you rich"
"It doesn't turn you into finance bro"
"You don't need masters to justify taking econs"

That helps reduce performance pressure.
Good teacher instinct there.

---

3️⃣ PPC Explanation Is Actually Solid
Conceptually, your PPC explanation is good:
Limited resources
Trade-offs
Opportunity cost
Full employment
Movement inside vs shift outward
Asymmetric shift

You covered real exam-relevant framing.
That's good fundamentals.

---

⚠️ Where It Weakens
Now I'll be strict.

---

❌ 1️⃣ Structure Is Too Chaotic (Biggest Issue)
Right now your lesson feels like:
Why not econs
Why econs
Incentives
Finance bro
Inflation
AI
Anime
PPC
Israel war
Deliveroo
Grab
Gojek corrupt founder teaser

Your mind moves fast.
 But students' cognitive load cannot keep up.
Especially first lesson.
You are thinking at Layer 3.
 They are still at Layer 0.

---

What happens cognitively?
When structure isn't clear:
Students don't know what to write down.
They don't know what's examinable.
They don't know what's a joke vs core concept.
They mentally disengage.

Even if content is good.

---

❌ 2️⃣ Too Many Concepts in First 30 Minutes
You introduced:
Incentives
Scarcity
Trade-offs
Time preference
Inflation
R&D
Productivity
Full employment
Economic growth
Market exit
Opportunity cost
Overbooking algorithm pricing

That's a whole term's worth of ideas.
First lesson should:
Go deep on 1–2 ideas.
Not surface-skim 10.

---

❌ 3️⃣ Language Precision Needs Tightening
Some examples:
You say:
"Assume full employment."
Good.
But then:
"Full employment is paraphrase of utilising everything."
Better phrasing:
"Full employment means all available resources are fully utilised - labour, capital, land."
Students need clean exam language early.
Right now you mix:

Singlish

Conceptual

Exam language

Random metaphor

That blurs signal vs noise.

---

❌ 4️⃣ Your Energy Is High, But Focus Is Scattered
You have:
Charisma
Energy
Speed

But:
You jump topic before students anchor.
Example:
You start with $50 vs $100.
 That's powerful.
But you didn't:
Name it as "Time value of money"
Or "Time preference"
Or link to "Intertemporal choice"

So students enjoy the debate…
 But don't extract the framework.

---

If This Was a First JC Lesson…
Here's how I'd reframe it using your style.

---

🔥 Cleaner Version of Your First Lesson
Part 1 (10 min): Why Econs Matters
Ask only 3 questions:
Why do airlines overbook?
Why did Deliveroo exit Singapore?
Would you prefer $50 now or $100 next year?

Then say:
All three are about scarcity, incentives, and trade-offs.
That's economics.
Stop there.
Don't introduce 8 other concepts.

---

Part 2 (20 min): Go Deep Into ONE Concept - Scarcity
Define clearly:
Unlimited wants, limited resources.
Use:
Anime vs mugging.
Shopee vs savings.
Singapore producing chips vs housing.

Draw PPC cleanly.
Slowly.

---

Part 3: End With a Big Idea
Economics doesn't tell you what to choose.
 It teaches you how to think when you have to choose.
That's a strong philosophical close.

---

Your Teaching Profile
You are:
High cognition.
Big picture thinker.
Naturally strategic.
You think like a policymaker, not a JC teacher.

That's both your strength and risk.
You see 5 layers ahead.
 Students see 0 layers.
Your job first 3 lessons:
 → Reduce abstraction.
 → Slow down.
 → Anchor vocabulary.

---

Concrete Improvement Plan
1️⃣ Slow Your Concept Density by 40%
If you think something is obvious - it is not obvious to them.

---

2️⃣ Use Clear Transitions
Say things like:
"Now I'm switching to opportunity cost."
"Now I'm explaining why PPC shifts outward."
"This is examinable."

Make structure visible.

---

3️⃣ Write Key Phrases on Board
Students need visual anchor:
Scarcity
Opportunity Cost
Full Employment
Productive Efficiency

Right now your lesson is mostly oral flow.
JC students need written skeleton.

---

4️⃣ End with 1 Clear Takeaway
Right now you end with:
Gojek teaser
Corrupt man
Homework
Fun intro

Better close:
"Next class, we'll see how this explains why firms exit markets."
Clean. Focused.

---

Final Evaluation
Category/ Score
Content Knowledge 9/10
Real-life Relevance 8/10
Structure 5/10
Cognitive Load Management 5/10
Exam Clarity 6/10
Engagement 7.5/10

Overall: 6.5 / 10 with very high potential

---

You are not a bad teacher.
You are an intellectually fast teacher.
The adjustment is not intelligence.
 It's pacing and pruning.

---

If you want, I can:
Rewrite this into a tight 45-min first lesson script
Or help you design a "Week 1 JC Econs Masterclass flow"

Which direction do you want to sharpen?

This is how Gemini responded

I would give this lesson a solid 8.5/10 for pure engagement and "real-talk" factor.

As an AI reading a raw, unedited transcript, I am definitely missing the whiteboard scribbles and the actual energy of the room, but I can easily tell this is exactly the kind of lively, relatable vibe that keeps JC kids awake during an afternoon block.

Here is a breakdown of what worked brilliantly and where it could be tightened up:

The Highlights
Hyper-Relatable Context: Swapping out boring, traditional textbook analogies (like guns vs. butter) for "mugging vs. watching anime" is a fantastic way to explain the Production Possibility Curve (PPC) and opportunity cost.

Real-World Hooks: Bringing in Gojek, Deliveroo, Shopee, and even the "Cai Fan" inflation rate makes Economics feel like a living, breathing subject rather than just a syllabus requirement.

Brutal Honesty: Telling students right off the bat not to take Economics if they are only doing it for the A-Level contrasting subject requirement is refreshingly candid. It sets realistic expectations.

Active Learning: The "$50 now vs. $100 in a year" thought experiment is a great, interactive way to introduce the time value of money and inflation.

Areas for Tweak and Polish
Pacing and Transitions: The jump from "Why take Economics?" directly into answering Source-Based Case Study (CSQ/SSQ) questions on PPC is a bit of a speed-run. It works for a crash course, but might give some students mental whiplash.

Examiner Pedantry: You successfully brought the focus back to the rigid definitions needed for the A-Level exams (like "given state of technology" and "full employment"). Keeping a sharp boundary between the "fun anime examples" and the "Cambridge-approved phrasing" is crucial so they don't accidentally write about TikTok in their actual papers.

Transcript Chaos: This is entirely due to the voice-to-text software, but the transcript is incredibly scattered! If this is being recorded for absent students, the auto-captions might leave them quite confused.

Overall, it sounds like a highly entertaining and effective introductory class that grounds abstract concepts in the everyday reality of a teenager in Singapore.

Would you like me to clean up this raw transcript into a structured, readable set of summary notes for your students?

This is how Claude responded

Lesson Rating: 7.5/10
What worked well
The teacher has strong rapport and genuine enthusiasm — the Singlish-inflected conversational style creates psychological safety, and the analogies (mugging vs anime for PPC, overbooking, Grab/Foodpanda) are genuinely clever and relatable for a JC audience. The lesson also does a good job answering "why econs?" before diving into content, which is motivationally smart for a first class.
The Socratic questioning throughout ("how many hours do you have?", "$50 now or $100 in a year?") actively involves students rather than lecturing at them.
Areas to improve
The transcription reflects some significant structural looseness — concepts like opportunity cost, the PPC definition, and the full employment assumption get explained somewhat out of order and require students to track a lot of implicit logic. For weaker students, this could be confusing even if engaging.
The worked examples (F to E, H to G, D minus E) are handled well procedurally, but the why behind each step could be more explicitly narrated for exam technique purposes.
There's also a bit of over-reliance on "you all know lah" as a comprehension check — it doesn't surface genuine misunderstanding.
Summary
Great energy and cultural fit for the classroom. The teacher clearly knows the material deeply and cares about students. The main growth area is tightening the logical sequencing so strong rapport translates into equally strong conceptual clarity.

Next steps

Focusing on what value AI can provide, I will:

1. Continue to monitor and stress test what constitutes ‘useful’ feedback as the user — diving deeper into specific aspects that abstracted as a system and deliver value to educators
2. Compare the quality of feedback between input material, namely text, v audio v video. From a product perspective, cost increases significantly ascending order in the input medium suggested. Is it worth the cost, is there a substantially compelling reason to prioritize one input medium over another?
3. Finding low-effort ways to make opting-in to recording easy without central infrastructure. Smart classrooms are a big financial investment and without proving value of ^, is a big jump ahead. In my case, oddly enough, calendar prompts for Notion AI Meeting Notes become the trigger reminder for “oh yes, time to record my lesson for feedback”. Even as an (outgoing) Product Manager for lesson feedback, I find myself having to actively remind myself to hit record, else I will forget. I have probably forgotten to hit record for 50% of my classes this week.
4. Unpacking what sustained feedback means in the context of mentoring. Mentoring conversations already take place and such tech isn’t a substitute

I asked Gemini, Claude and ChatGPT to rate my lesson was originally published in String on Medium, where people are continuing the conversation by highlighting and responding to this story.