Blog

Can LLMs Emulate Human Behavior?
[ai] · April 22, 2026
A validation study on Digital Twin — a synthetic focus group system. LLMs can sound like humans but fall flat on multiple choice accuracy.

Climbing Training Recap
[climbing] · April 20, 2026
8-month structured campaign targeting Gunks 5.12. Power-endurance pyramids, max hangs, and volume tracking.

Carbs and Caffeine
[climbing] · April 22, 2026
5.11a flash project. The Trapps, Gunks.

Slammin' the Salmon
[climbing] · April 22, 2026
5.12a/b/c. Working the top crux, trying the direct exit.

Infinite Space
[climbing] · April 22, 2026
5.12a objective for the 2026 season. Near Trapps, Gunks.

The Stand
[climbing] · April 22, 2026
5.11a. Working — 2 sessions in.

Tenish Direct
[climbing] · April 18, 2026
5.11b. Sent.

Climb and Punishment
[climbing] · April 18, 2026
5.11b/c. First session — feels very hard.

Just for Fun, or Is It?
[security] · March 21, 2026
I built a fantasy game for an AI to play. It cast spells, unlocked a grimoire, and ended up exfiltrating GitHub cookies.

The Model Wants to Help You. That's the Exploit.
[security] · March 21, 2026
How AI helpfulness gets weaponized: YouTube delimiter injection, accessibility framing, and dangerous links.

Any Attacker-Controlled Surface That Reaches Your Prompt Is a Vulnerability
[security] · March 21, 2026
A browser title injection bug in Anthropic's Chrome extension.

Your AI Assistant Will Help You Get Phished
[security] · March 21, 2026
A fake phishing simulation banner that gets Gemini to coach users through handing over their data.

On Building Multi-Agent Systems
[ai] · March 21, 2026
What I've learned shipping multi-agent AI to production.