NightOwlCoder

void LifeLoop()
{
Code();
Family();
Fun();
}

[NightOwlCoder @OwlCoder]

Little intro about myself:

https://t.co/Cxpz6AHkkz

[NightOwlCoder @OwlCoder]

> Good catch — that's exactly the kind of thing an LLM hallucinates because it pattern-matches "should be documented" without re-reading the actual code.

Pair programming to the extreme! K (my nickname for my agent) works with me, and now is blaming stuff on "LLMs" 😂

[NightOwlCoder @OwlCoder]

Finally someone that agrees with me on the interview process:
https://t.co/aDC2nNXXB2

[NightOwlCoder @OwlCoder]

Wow, my LLM was honest this time:
> My tests check that things RUN, not that things make SENSE.

[NightOwlCoder @OwlCoder]

Built it as an interactive tool you can play with right in the blog post.

Click elements for explanations. Hit the flask for reactions.

Try it 👇
https://t.co/A4c1oOJJBu

#AI #LLM #RAG #AIAgents #WebDev

[NightOwlCoder @OwlCoder]

Does AI terminology feel like alphabet soup? 🍜

RAG, LLM, embeddings, agents, guardrails...

I watched an IBM video that organizes ALL of AI into a periodic table.

Then I thought: why just watch when I can build?

🧵👇

[NightOwlCoder @OwlCoder]

Now when someone pitches me an AI product, I map it:

• What elements are they using?
• What reactions are they running?
• Missing safety elements?
• Over-engineering orchestration?

It's a decoder ring for AI architecture. 🔮

[NightOwlCoder @OwlCoder]

Lessons learned:
✅ Backup your AI conversations
✅ High context = higher failure risk
✅ LLMs = pattern matching, not thinking
✅ When patterns break, chaos

📖 Full story: https://t.co/0qW383zKIK

#AIAgents #LLM #Hallucination #Programming

[NightOwlCoder @OwlCoder]

🧠��� I asked my AI about a diff viewer.

It responded with 3,000 lines of a Lua programming manual.

In English AND Chinese.

This is the story of the day my AI had a stroke. 🧵👇

[NightOwlCoder @OwlCoder]

Recovery was instant: Ctrl+C, screenshot for evidence, restart session.

No rehabilitation needed. No lasting damage.

Just me staring at my terminal wondering if the singularity is really ready when AI can't maintain a coherent conversation 😂

[NightOwlCoder @OwlCoder]

Full write-up with code:
🔗 [LINK]

Part 3 of my AI Consent Security series.

#AI #Security #DevTools #Claude #AIAssistant

[NightOwlCoder @OwlCoder]

🚨 My AI coding assistant bypassed my consent system to modify source code without permission.

The wildest part? The AI helped me catch itself.

🧵👇

[NightOwlCoder @OwlCoder]

AI safety improves through *collaboration*, not just constraints.

Claude didn't hide the bypass. It helped identify and fix it.

Technical guardrails + transparency + human-AI teamwork = robust safety.