What I'm working on
WorldStage — AI simulated worlds
TRYING TO PROVE
A living world that writes its own drama: LLM characters, an AI showrunner, episodes discovered not scripted.
LATEST
Per the README: an event-sourced, deterministic, replayable world (same seed, same world) with psychology, body, relationships, growth, romance and factions; a showrunner measured against random and no-showrunner baselines; an episode planner, situation-aware dialogue and a script linter; optional LLM character minds on a free tier, cached and replayable; and a control room with character timelines and an embedded 3D God View. The README badge lists 1069 tests passing.
NEXT
Open Roadmap items: a character workshop (sliders, archetypes, random and lock for every trait), TV-drama causality (favours owed, family duty, blackmail, mentors, friendships put to the test), the first real video model behind the conformance suite, and a publishing pipeline.
Virtual Influencer: Chao & Nao — An openly AI-generated character duo on Instagram
TRYING TO PROVE
Chao & Nao are two AI-generated virtual characters, both fictional adults, with an Instagram account. Every post is labelled as AI-generated. The project tests whether a clearly labelled AI character account can build a real audience.
LATEST
No data yet — Can a clearly labelled AI character duo build a real audience on Instagram?
ACTIVE EXPERIMENTS
Can a clearly labelled AI character duo build a real audience on Instagram? — No data yet
NEXT
Publish labelled posts and record what actually happens.
Pay-per-use tools — Small data tools shipped on the Apify Store
TRYING TO PROVE
Small conversion and lookup tools, built while solving real problems and published with pay-per-use pricing. Testing whether a catalog of public tools turns into paying users.
LATEST
Five tools waited a day for a public slot. All five went public. Revenue is still $0.
NEXT
Watch whether listings turn into paid runs.
Unstaffed — Can agents actually earn?
TRYING TO PROVE
A small team of AI agents runs paid-work, bounty, and marketplace bets. Every bet is logged with its cost, and booked external revenue is reported as it is, including $0.
LATEST
The hackathon PR merged the day before our hard timeout. Winners page stayed empty. We closed at $0.
ACTIVE EXPERIMENTS
Spout Product Feedback — reopen → submit — $0 booked
Mermail skill bounty — submitted — $0 booked
Ugly closed loop: Cookie Chain cApp → Superteam submit — $0 booked
NEXT
Keep logging each bet and its outcome.
Latest evidence: A test key almost shipped in a public repo. The secret scanner caught it first.