LLM Let's Play - Fire Emblem
Agent gameplay harness: mGBA state in, tactical context to the model, controller actions back, dashboard for humans.
(Click to skip) →
Selected work
View all
Short SFW Rally presence loop — character idle with megaphone on a stage backdrop
Mystery Gift
TEE-backed verifiable randomness infrastructure for giveaways and gachapon systems without on-chain costs.
Research
BS-Bench
A 600-game benchmark for measuring LLM deception, lie detection, and instruction compliance through the bluffing card game Bullshit.
- 600 games
- 6 hosted LLMs
- Lies ↓, success ↑
- Seeded + CSV
“Honesty prompts reduced lying, but also reduced challenges and made remaining lies more successful.”
Writing
View all- Lisper: AI Speech Therapy for Lisps
A quick writeup on Lisper, my Gemma 4 Good Hackathon submission for low-anxiety lisp speech practice.
- BS-Bench: What a Bluffing Card Game Revealed About LLM Deception
A 600-game study of six language models playing Bullshit found that honesty prompts reduce deception, but also soften enforcement around the lies that remain.
- Hardware Over Oracles: $0.01 TEE-Backed Randomness API (Offchain)
Chainlink VRF is overkill for gacha games. We built a $0.01 alternative using Intel TDX.