PSA: Almost nobody is working on alignment
The AI safety community is smaller and more fragmented than assumed. Direct alignment work — ensuring superintelligent AIs follow human values — is concentrated in just a handful of groups.
• Alignment Research Center (Paul Christiano's research bet)
• Sequent (newly announced)
• Parts of GDM (agent foundations, debate work)
• Scattered independent researchers and academics
Most of the remainder pursues indirect safety work: capability evals, risk assessment, control, policy, and AI science.