Keep up.

You can’t keep up with new AI.

chatgpt‑o3, chatgpt‑o4‑mini, gemini‑2.5‑flash, gpt‑4.1, llama‑4, claude‑sonnet‑3.7‑deep‑thinking, mistral‑pixtral‑large…

You think I'm messing with you. But all of these got launched (or got loud) since... March.

Instead of chasing them, I sat with one model—ChatGPT‑o3—and wrote this entire blog with it, in front of you.

Why document the process?

1. Show how few‑shot prompting works in real life.
2. Prove that a single model can take you from idea to published post.
3. Give you a replicable checklist you can use at home, with any model (I do love o3).

The exact prompts we used

Prompt #0 — Few‑shot setup

I want to see how good you are at writing articles for me.  

Below are two blogs I wrote and the opening of a third.  Finish Blog #3 in the same voice.

## blog 1

## end

## blog 2

## end

## blog 3

## end

I pasted the full text of Blog #1, Blog #2, and the opening lines of Blog #3.

That single block is the “few‑shot” context—o3 now knows my tone, structure, and length.

o3 reply (trimmed): “Proposed direction for Blog #3…” (outline suggestions)

Prompt #1 — Outline collaboration Not sure where we’re going.  Let’s refine the outline together—keep it short, bullets, same style.   o3 returned a concise 8‑section outline.

Prompt #2 — Make it meta I want this blog to be a live case study of us writing together.  No screenshots—quote text blocks only.   o3 rebuilt the outline around our actual chat.

Prompt #3 — Draft pass Write the full post following that outline.  Keep paragraphs short, use checkboxes where helpful, avoid analogies.  

My redlines (sample):

‍• Cut “fire‑hose” metaphor

• Add step‑by‑step so readers can replicate

• Include the first few‑shot prompt in the post itself

o3 adjusted and produced the version you’re reading now.

Step‑by‑step: replicate at home

1. Collect context
   • Copy 1‑2 past pieces of writing you like (your “few shots”).
   • Add any opening lines you already have.
2. Prompt the model

I want you to finish this new piece in the same voice.  Here are examples of my style and the opening lines:  [paste examples + opening]  

1. Ask for an outline first
   • “Give me a bullet outline, 6‑8 sections, same tone.”
   • Edit the outline until it feels right.
2. Generate the first draft
   • “Write the post following that outline. Short paragraphs, occasional checkboxes, no jargon.”
3. Redline
   • Point out anything off: length, tone, banned words.
   • Let the model revise.
4. Polish
   • Quick grammar pass (Grammarly or your eyes).
   • Publish.

Total tools: ChatGPT‑o3 + a spell‑checker.

What this live run shows

☑ Few‑shot context is enough. Two past blogs + an intro gave o3 everything it needed.

☑ Iterative prompts > one‑and‑done. Outline → draft → redline → publish.

☑ You don’t need every shiny model. Master one workflow first.

You don't believe I wrote 99% of this blog with o3?

Well, here's the full chat transcript: ​https://chatgpt.com/share/6803428c-8fcc-800f-95dc-6a23048f3f3c​.

AI is very real, my friend.

PS: What AI do I (actually) use the most?

Here's my April 2025 list of tools I use every day:

​https://www.linkedin.com/posts/ruben-hassid_8-tools-i-actually-use-to-do-the-work-of-activity-7318951434954117121-qyQE​