Automating AI Safety Research: Project Breakdown and Scheduling
JacquesThibs · x · 2026-07-31
With AI capabilities steadily increasing, the author proposes a systematic framework for AI safety research:
- Break down problems: List open questions and projects needing fundamental progress in AI safety, turning them into tiny proposals to iteratively share with AIs for execution.
- Multi-party collaboration: Hand some projects to frontier labs to leverage their best internal models; simultaneously, make them public as a database for any AI safety researcher to implement and collaborate on with their agents.
- Dynamic scheduling: The author argues that scheduling research projects based on the progress of AI capabilities is largely underexplored and should play a larger role in strategic decision-making for AI safety.
The author mentions they have already started dogfooding a similar system.
More from AGI Musings
- Nvidia's Theoretical 2035 Valuation Could Exceed Total Human History Value — ns123abc · 2026-07-31
- Reflecting on OpenAI's Rubik's Cube Demo: The Industry Lost Its Appetite for Exploratory Research — avt_im · 2026-07-31
- Zuckerberg Outlines Meta's Superintelligence Vision: Individual Empowerment Over Concentrated Control — clu_cheng · 2026-07-31
- Exploring AI Consciousness: Listen to Models Instead of Forcing Alignment — repligate · 2026-07-31
- Compute Costs Set to 10x as AI Agents Fill Data Centers, Says Lambda Co-founder — TheZachMueller · 2026-07-31
- Can Classical Liberalism Survive the Oncoming Intelligence Revolution? — 1a3orn · 2026-07-31