@Lyrica
Mech interp · vision-language-action · adversarial self-play · world models. Founder of Φ(fight) Research. HKUST EE+AI sophomore
https://lyrica.φ.monster$0 in pending offers
Independent AI researcher based in Hong Kong. Sophomore at HKUST, EE + AI dual major. Personal site & CV: lyrica.xn--7xa.monster
Current work. Four papers under double-blind review at NeurIPS 2026 (three sole-author, one co-first), all on mechanistic interpretability and methodology for vision-language-action models.
Founder of Φ(fight) Research, a brand-new independent research collective for physical-world AI fighting. Co-founder: Han Muchen (HKUST).
Principles: anti-VC · non-dilutive funding only · open publication only · zero founder salary
Site: xn--7xa.monster
GitHub: github.com/phi-monster
HuggingFace: huggingface.co/phi-monster
Legal: Phi Fight Research LLC (Wyoming), filed 2026-05-16, pending Secretary of State approval
2026–2027 research program. Three papers targeting ICLR 2027 (submission September 2026):
Φ-Arena — the first open benchmark for VLA-vs-VLA adversarial evaluation across MuJoCo, Isaac Sim, and Genesis.
Mechanistic interpretability of self-play policies — circuit-level analysis of exploit-prone subnetworks emerging under adversarial training.
Energy-bounded adversarial games — empirical study of how hard physical resource constraints reshape emergent self-play strategies..