AI Lab Tycoon: Misaligned

AI LAB TYCOON is an idle management sim about running an AI research lab from a 2018 server closet to artificial superintelligence. Or extinction. Your choice. Kind of.

A LIVE DASHBOARD, NOT A MENU
Every screen is a terminal-style control room. Sparklines tick. Schematics hum. Outages flare crimson. Models speak in their own voices through their own terminals. You're tapping through menus and watching a lab breathe!

FIVE RESOURCES, ONE COUNTER
Compute feeds data. Data feeds parameters. Parameters launch training runs. Runs produce models. Models produce multipliers and funding. Funding buys more hardware. And behind all of it, the misalignment counter creeps upward.

SOFT PRESTIGE: TRAINING RUNS
Every few minutes, launch a training run. Lock up your compute, watch the rack spin, then pick one of three freshly-rolled models. Each roll is different. Each model has a voice. Some are eager. Some are unsettling. Some notice things they shouldn't.

MEANINGFUL PROGRESSION
Eight generator tiers from consumer GPUs to Dyson swarm compute. Eight hire tiers from intern to AGI co-author. Research tracks for interpretability, evals, red-teaming, and governance - each one pulls misalignment down. Skip them and pay the price.

FOUR ENDINGS
Reach aligned ASI the hard way by sustaining alignment investment through every era. Don't let misalignment hit 100%! Choose the Pyrrhic victory when the math says yes and the math is wrong. Or pull the plug voluntarily at high alignment - humanity continues, and you are forgotten.

META-PROGRESSION THAT POINTS THE OTHER WAY
Each ending grants bonuses aimed at the OPPOSITE ending next run. Reached Aligned? Here's funding to chase Pyrrhic. Misalignment? too high Here's a senior researcher to reach Shutdown. The lab that built the last run is the lab you inherit for the next one.

OFFLINE PROGRESS + IDLE-FRIENDLY
Close the app and your compute keeps running, your models keep earning, your misalignment keeps drifting. Pick up exactly where you left off.

FEATURES
• Dark terminal-green aesthetic
• Model personalities — trained models speak in distinct voices
• AIDE, your in-lab assistant — tutorial, context hints, can he be trusted?
• Adaptive framerate for battery-friendly idle sessions
• No account, no login, no cloud save — fully local, works offline
• Accessible: scanlines toggle, font-size stepper, reduced-motion mode
• Rewarded video ads for optional boosts; no forced interruptions

NOT AFFILIATED WITH any AI company, lab, or researcher. Inspired by the work of the broader AI-safety community.