Uploaded January 2026 | Updated September 2026, 1 hour ago
Day 12/42: What Is a Prompt? (System vs User)
Yesterday, we saw how models learn *how* to behave.
Today, we talk about how you actually talk to them.
Everything you send to an LLM is part of a prompt.
But not all prompts are equal.
- The system prompt sets the rules and personality.
- The user prompt is your actual question or task.
Both are read together.
The system prompt says *how to answer*.
The user prompt says *what to answer*.
Once you see this split, prompt engineering suddenly makes sense.
Missed Day 11? Worth watching.
Tomorrow, we explain how models remember conversations: context windows.
I’m Louis-François, PhD dropout, now CTO & co-founder at Towards AI. Follow me for tomorrow’s no-BS AI roundup 🚀
#PromptEngineering #LLM #AIExplained #short
Day 12/42: What Is a Prompt? (System vs User)
Yesterday, we saw how models learn *how* to behave.
Today, we talk about how you actually talk to them.
Everything you send to an LLM is part of a prompt.
But not all prompts are equal.
- The system prompt sets the rules and personality.
- The user prompt is your actual question or task.
Both are read together.
The system prompt says *how to answer*.
The user prompt says *what to answer*.
Once you see this split, prompt engineering suddenly makes sense.
Missed Day 11? Worth watching.
Tomorrow, we explain how models remember conversations: context windows.
I’m Louis-François, PhD dropout, now CTO & co-founder at Towards AI. Follow me for tomorrow’s no-BS AI roundup 🚀
#PromptEngineering #LLM #AIExplained #short










![Multi-Image Editing Just Got Way Better with Qwen
🚨 Big update for Qwen-Image-Edit!
The new [2509] version takes things to the next level:
Multi-image consistency boost: keeps facial identity rock-solid across poses and styles (portraits, restorations, memes, cartoons).
Multi-image editing (1–3 inputs): trained with image concatenation → combos like person+product, person+scene… even works with ControlNet maps (pose, depth).
Better single-image edits too:
Stronger identity preservation
Advanced text handling (fonts, colors, content changes)
And yes, it’s an open model.
This isn’t a brand-new model, but the update makes Qwen-Image-Edit way more powerful.
⚠️ Just make sure to switch to version [2509] to unlock all the improvements.
Which AI updates should I break down next? Drop your pick and I’ll tag you.
I’m Louis-François, PhD dropout, now CTO & co-founder at Towards AI. Follow me for tomorrow’s no-BS AI roundup 🚀
#AI #Qwen #ImageEditing #short Multi-Image Editing Just Got Way Better with Qwen](https://i.ytimg.com/vi/VPYEnHtBIOo/mqdefault.jpg)