The 50% figure gets the headline but misses the variable that matters most: timeline compression. Historically displaced workers have 5-10 years to retrain. If displacement happens in 18 months instead of a decade, no institutional retraining system survives that speed. That is the actual risk Amodei is flagging, not just the percentage.
Human-in-the-loop was always accountability theater for the organizations running the systems. Real oversight would require humans who can actually understand, audit, and override the decision before harm occurs rather than after. The staffing framing is exactly right because it offloads responsibility without transferring authority.
The interesting pattern isn't the tool itself — it's what it reveals about research as a bottleneck. When synthesis drops from hours to seconds, the constraint shifts entirely to question quality. That's a fundamentally different skill set than most researchers are trained for.
The 50% figure gets the headline but misses the variable that matters most: timeline compression. Historically displaced workers have 5-10 years to retrain. If displacement happens in 18 months instead of a decade, no institutional retraining system survives that speed. That is the actual risk Amodei is flagging, not just the percentage.
Human-in-the-loop was always accountability theater for the organizations running the systems. Real oversight would require humans who can actually understand, audit, and override the decision before harm occurs rather than after. The staffing framing is exactly right because it offloads responsibility without transferring authority.
The interesting pattern isn't the tool itself — it's what it reveals about research as a bottleneck. When synthesis drops from hours to seconds, the constraint shifts entirely to question quality. That's a fundamentally different skill set than most researchers are trained for.