Frontier Watch: monitoring our advancement towards superintelligence

Good decisions about AI risk depend on a calibrated read of how fast capability is actually moving. Today that read is dominated by the actors with the largest financial stake in it: timelines from labs and investors move with funding rounds, not evidence (SoftBank, OpenAI, and xAI have each moved dates or definitions as financing required). Both overstatement and understatement of progress degrade governance.

An independent, incentive-isolated measure improves the epistemic substrate that policymakers, journalists, and the public rely on. It provides a reference reading produced by people with nothing riding on a particular date being believed. Reporting a median and a spread communicates genuine uncertainty rather than false precision, reducing both hype and complacency. And the monitoring side adds a watchdog function: it surfaces when labs weaken safety or data commitments — backsliding that is easy to miss one announcement at a time, and that bears directly on whether safety norms hold under competitive pressure.

I am not an alignment researcher and do not claim to substitute for that work. This improves the shared situational awareness that alignment and governance both depend on — a niche getting little attention from existing funders.