@irly65: #riyadh #fyp #النسيم_النظيم #وحيدين_الشرق

B
B
Open In TikTok:
Region: SA
Saturday 22 August 2026 19:21:11 GMT
9016
79
3
214

Music

Download

Comments

sora03_5
ساره الغامدي :
يلااااا عاد مستحيلللل هي نفس الشخص!!
2026-08-24 23:31:11
3
a.776009
A 77 :
يا لبي ايه
2026-08-22 21:12:58
0
To see more videos from user @irly65, please go to the Tikwm homepage.

Other Videos

Prompt engineering isn't dead. It just stopped being the job. ⚡ Three eras of getting an LLM to do real work — and most people are still stuck in the first one 👇 🟢 1. PROMPT ENGINEERING One window in, one answer out. System prompt + user message → model → response. You control: the words. That's the whole lever. Better phrasing, better examples, better structure. Real skill — just a small surface. The catch: anything that doesn't fit gets cut, and you have no say in what. 🔵 2. CONTEXT ENGINEERING Now there's a pool of things you COULD send — docs, tools, memory files, rules, history — and a window that fits maybe a fraction of it. You control: what goes in. The window becomes a budget. Every doc you include is a doc you're paying for on every single turn. The skill shifts from writing to curating. The model can now call a tool, and the result comes back as history. 🟣 3. HARNESS ENGINEERING The model becomes one node in a loop. Window → model →
Prompt engineering isn't dead. It just stopped being the job. ⚡ Three eras of getting an LLM to do real work — and most people are still stuck in the first one 👇 🟢 1. PROMPT ENGINEERING One window in, one answer out. System prompt + user message → model → response. You control: the words. That's the whole lever. Better phrasing, better examples, better structure. Real skill — just a small surface. The catch: anything that doesn't fit gets cut, and you have no say in what. 🔵 2. CONTEXT ENGINEERING Now there's a pool of things you COULD send — docs, tools, memory files, rules, history — and a window that fits maybe a fraction of it. You control: what goes in. The window becomes a budget. Every doc you include is a doc you're paying for on every single turn. The skill shifts from writing to curating. The model can now call a tool, and the result comes back as history. 🟣 3. HARNESS ENGINEERING The model becomes one node in a loop. Window → model → "done?" → no → tool call → environment → result → back into the window. Repeat until done. You control: the loop. Long-term memory decides what survives between turns. The tools define what's even possible. 👀 The one nobody says: at stage 3, the model is the least interesting part of your system. Two teams on the same model ship wildly different products, because the difference is the loop, the tools, and what you chose to keep in the window. That's engineering, not prompting. 🎯 The rule: you stopped writing prompts. You build loops. Notice the scissors in all three panels. The window still truncates at every stage. What changes is how much control you have over what survives. 📸 Screenshot it. Figure out which stage you're actually working at. Follow @hackproduct — we turn scary AI-engineering concepts into things you can ship. ⚡ . . #contextengineering #promptengineering #AIagents #AIengineering #LLM

About